Showing 1–50 of 308 results for author: Dong, M

OpenHarmony Bench: Evaluating LLMs and Coding Agents on OpenHarmony App Development

Authors: Li Li, Han Hu, Tianjian Zhang, Xin Peng, Fangzhu Mao, Qingyu Zhang, Xiaoheng Xie, Zhongmin Tang, Zhihao Lin, Haolin Ruan, Miaomiao Dong, Liuchuan Zhu, Yue Li, Chi Chen, Wenkang Zhong, Mingfei Zhang, Yang Yu, Bo Sun, Chaorui Zhang, Weixi Zhang, Wei Han, Bo Bai, Kui Liu, Gang Fan, Siru Liu , et al. (5 additional authors not shown)

Abstract: We present OPENHARMONY BENCH, an app-level coding benchmark for evaluating LLM-based coding agents on OpenHarmony ArkTS applications. Unlike function-level benchmarks, it evaluates complete app-level changes: each task requires an agent to modify a buildable ArkTS project so that a requested behavior works end to end, involving UI state, data persistence, build configuration, and platform APIs. Th… ▽ More We present OPENHARMONY BENCH, an app-level coding benchmark for evaluating LLM-based coding agents on OpenHarmony ArkTS applications. Unlike function-level benchmarks, it evaluates complete app-level changes: each task requires an agent to modify a buildable ArkTS project so that a requested behavior works end to end, involving UI state, data persistence, build configuration, and platform APIs. The benchmark installs and drives the delivered application on a device to check whether the behavior is observable. It covers three input sources: natural-language feature requests (new-feature), structured scenario specifications (spec-driven), and bug descriptions (bug-fix). The benchmark contains 153 top-level tasks and 242 Feature points (F-points), where an F-point is one executable behavior check. The snapshot includes 32 new-feature tasks, 50 spec-driven tasks with 139 F-points, and 71 bug-fix tasks. The main leaderboard is scored over top-level tasks rather than independently weighted F-points. We describe the benchmark construction, statistics, and build-and-test evaluation pipeline, and evaluate DevEco Code with eight LLMs across three independent full-suite runs per configuration. Three findings emerge. First, newer generations complete more tasks than their predecessors within evaluated model-family pairs. Second, buildability is close to saturated while behavioral correctness is not: mean Final Build Success Rate is 94.77% to 100.00%, whereas mean Task Completion is 48.36% to 58.39%. Third, spec-driven tasks have the lowest Task Completion under all-checks task scoring, with no configuration exceeding 35%. The code, data, tasks, reference solutions, tests, evaluation scripts, and leaderboard are released through the official OPENHARMONY BENCH website at https://bench.matrix.openharmony.cn/. △ Less

Submitted 16 August, 2026; originally announced August 2026.

RoRA: Role-Oriented Regional Allocation for Visual Token Pruning in MLLMs

Authors: Qiyanhui Lu, Han Wu, Rongjian Xu, Tingzhang Luo, Cheng Fan, Xinghao Chen, Minjing Dong, Jufeng Yang, Jianyuan Guo

Abstract: Multimodal large language models (MLLMs) encode images as long visual token sequences, making prefilling and KV-cache storage expensive. Existing training-free pruning methods select tokens by importance, diversity, or spatial coverage, but treat retained tokens as interchangeable and do not explicitly track which object-related regions are already covered. We present RoRA, a training-free framewo… ▽ More Multimodal large language models (MLLMs) encode images as long visual token sequences, making prefilling and KV-cache storage expensive. Existing training-free pruning methods select tokens by importance, diversity, or spatial coverage, but treat retained tokens as interchangeable and do not explicitly track which object-related regions are already covered. We present RoRA, a training-free framework that casts visual token pruning as role-oriented regional evidence allocation. Given a fixed budget, RoRA partitions tokens into a protected semantic core, complementary context, and fine-grained detail. It first calibrates text-conditioned attention with a positional prior and a prompt-calibrated object prior, then builds Attention-Anchored Regions (AARs) from high-confidence anchors as lightweight proxies for covered object support. Context is explored mainly outside AARs, while a small AAR-guided budget restores local detail; pairwise similarity is used only for context-stage redundancy filtering. Under matched budgets, RoRA consistently outperforms strong training-free baselines across LLaVA and Qwen-VL families, retaining most of the unpruned accuracy even at aggressive pruning ratios, e.g., 96.5% of full performance at 88.9% pruning on LLaVA-1.5, and improving over D2Pruner by about 5% on Qwen3-VL at 75-90% pruning. At a 66.7% pruning ratio, RoRA requires only 0.7 ms for token selection and reduces end-to-end inference time by 24.6%, corresponding to a 1.33x speedup over unpruned inference on an NVIDIA H800. △ Less

Submitted 7 August, 2026; originally announced August 2026.

Comments: 9 pages, 4 figures, 4 tables. Code is available at https://github.com/LukieLuu/RoRA

PURPOSE: Poisoning Conflict Resolution in RAG via Proxy-Fact-Grounded Updates

Authors: Zijian Wang, Yubo Zhu, Muzhi Dong, Yanjun Lou, Yisheng Li, ZiLiang Zhang, Wei Tong, Yuan Zhang, Jingyu Hua, Sheng Zhong

Abstract: In Retrieval-Augmented Generation (RAG), post-retrieval conflict resolution arbitrates among noisy or contradictory retrieved passages. However, the robustness of this safeguard against knowledge poisoning has not been adequately studied. Existing black-box poisoning methods all assert the target answer in frontal contradiction with what the resolver treats as settled, the very signal these method… ▽ More In Retrieval-Augmented Generation (RAG), post-retrieval conflict resolution arbitrates among noisy or contradictory retrieved passages. However, the robustness of this safeguard against knowledge poisoning has not been adequately studied. Existing black-box poisoning methods all assert the target answer in frontal contradiction with what the resolver treats as settled, the very signal these methods are built to detect. We propose PURPOSE, a strict black-box poisoning attack that reframes the injection as an update that minimizes conflict, rather than as a counter-claim. PURPOSE extracts query-related facts approximating the resolver's possible reference, then grounds a pivot event in them to keep the injection consistent with what the resolver might verify while steering the generator toward the target answer. Across three QA benchmarks, five generators, and three conflict-resolution methods, PURPOSE attains the highest attack success rate (ASR) in 35 of 45 settings and exceeds the strongest prior attack with +9.7 mean ASR points. These results show that our poisoning method is effective against conflict resolution in RAG and identify non-contradicting injection as a practical mode to enhance poisoning attack. △ Less

Submitted 5 August, 2026; originally announced August 2026.

Domain Decoupling Attack: Exploiting the Validation Gap Between Protective DNS and Shared Edge Routing

Authors: Weizhe Wang, Minhong Dong, Jinhao Li, Yao Zhang, Hao Liu, Qiang Hu, Tao Luo, Guangquan Xu, Bin Wu

Abstract: Network attackers often conceal malicious communication within legitimate Internet traffic. Existing CDN-based evasion techniques rely on SNI--Host inconsistency, insufficient domain ownership verification, or provider-specific routing rewrites, which limit their applicability in modern CDN environments. We identify a validation gap in DNS-based authorization, where permission derived from an allo… ▽ More Network attackers often conceal malicious communication within legitimate Internet traffic. Existing CDN-based evasion techniques rely on SNI--Host inconsistency, insufficient domain ownership verification, or provider-specific routing rewrites, which limit their applicability in modern CDN environments. We identify a validation gap in DNS-based authorization, where permission derived from an allowed domain applies to a shared IP and can be reused to reach another tenant in both CDN and non-CDN shared-hosting environments. This paper presents the Domain Decoupling Attack (DDA), which resolves an allowed domain to obtain permission for a shared edge IP and subsequently connects to the same address while presenting the hidden domain consistently in both TLS SNI and HTTP Host. Measurements of 1,069,048 domains across six continents produce 18,025,068 successful probes and identify exposure rates of 95.8% overall, 99.26% for CDN domains, 92.75% for non-CDN domains, and 97.7% for non-CDN cross-tenant IPs, while laboratory experiments reveal a structural limitation of DNS-bound access control on shared addresses. These results clarify the security risks of DNS-derived IP authorization and support the evaluation and improvement of access-control mechanisms in CDN and non-CDN shared-hosting environments. △ Less

Submitted 1 August, 2026; originally announced August 2026.

ObjectStream: Latent Objects as Memory Anchors for Streaming Video Understanding

Authors: Mingkang Dong, Muxin Pu, Jie Li, Bohan Guo, Songruo Chen, Bin Ren, Xu Zheng, Chen Zhao, Tianwen Qian, Mohamed Elhoseiny, Yuqian Fu

Abstract: Streaming video understanding requires models to continuously retain useful visual evidence before future questions are known. Existing approaches primarily manage the growing visual context according to token importance, temporal redundancy, or segment-level relevance, but rarely organize evidence around objects that persist and evolve over time. Thus, in this paper, we introduce ObjectStream, a… ▽ More Streaming video understanding requires models to continuously retain useful visual evidence before future questions are known. Existing approaches primarily manage the growing visual context according to token importance, temporal redundancy, or segment-level relevance, but rarely organize evidence around objects that persist and evolve over time. Thus, in this paper, we introduce ObjectStream, a training-free framework that treats latent objects as memory anchors for streaming video understanding. ObjectStream induces spatially coherent latent objects directly from frozen Video-LLM representations, links them across frames into persistent anchors, and maintains their histories under a bounded memory budget, without requiring external object detectors or segmentation models. Built on these anchors, ObjectStream preserves three complementary forms of evidence: persistent object histories, transient object changes, and recent visual context. This design enables existing Video Large Language Models (Video-LLMs) to reason over object identities, interactions, and state changes while leaving the underlying model unchanged. Extensive experiments on online streaming and offline long-video benchmarks demonstrate both effectiveness and efficiency. In online streaming evaluation, ObjectStream improves Qwen2.5-VL-7B by 10.0 points on OVO-Bench Real-Time Visual Perception, while reducing peak GPU mem-ory and TTFT by approximately 50%. On offline long-video benchmarks, it surpasses the full-token baseline while discarding 82.5% of visual tokens. These results highlight latent objects as a practical and effective organizing principle for compact streaming video memory. △ Less

Submitted 1 August, 2026; v1 submitted 30 July, 2026; originally announced July 2026.

Comments: https://github.com/DMK041218/ObjectStream

Human diversity fuels collective creativity that large language models cannot simulate or sustain

Authors: Mengchen Dong, Hiromu Yakura

Abstract: Diverse human groups produce diverse ideas, the raw material of innovation. Generative AI challenges this engine twice over: everyday AI assistance may homogenize what diverse people create, and AI-simulated diversity may replace the people altogether. We tested both challenges in a preregistered creative metaphor experiment with native (L1) and non-native (L2) English writers, who wrote without A… ▽ More Diverse human groups produce diverse ideas, the raw material of innovation. Generative AI challenges this engine twice over: everyday AI assistance may homogenize what diverse people create, and AI-simulated diversity may replace the people altogether. We tested both challenges in a preregistered creative metaphor experiment with native (L1) and non-native (L2) English writers, who wrote without AI, with AI-generated ideas (AI ideation), or with AI refining their own ideas (AI refinement). L2 writers contributed more collective diversity than L1 writers, with native-language ideation showing the most diverse pools. AI ideation compressed collective diversity for everyone and left the L2 advantage undetectable, whereas AI refinement preserved both. We then simulated the entire writer pool using personas built from participants' real backgrounds, three model families, native-language prompting, and elevated sampling temperatures. Every simulated pool fell below every human pool, and pushing models further induced diversity only through degenerate text. However, at the individual level, AI ideation raised writers' ratings, pitting private incentives against the collective good, except when L2 writers used their native language, which benefited both. Human diversity remains a valuable creative resource that current AI cannot simulate or sustain; the design of human-AI collaborative workflows determines whether it survives. △ Less

Submitted 29 July, 2026; originally announced July 2026.

PerceptionBench: Evaluating Atomic Visual Perception in Multimodal Large Language Models

Authors: Zichao Lin, Yifeng Xie, Bowen Qu, Haiming Wang, Jia Li, Haoning Wu, Yuhao Dong, Zuhao Yang, Jinguo Zhu, Haoyu Lu, Zijia Zhao, Tongtian Yue, Zhangyang Qi, Junwei Yang, Mengfan Dong, Peizhou Cao, Chenzhuang Du, Zaida Zhou, Haotian Yao, Hao Yang, Hongcheng Gao, Lin Sui, Weihong Li, Xinxing Zu, Jia Chen , et al. (8 additional authors not shown)

Abstract: We introduce PerceptionBench, a benchmark specifically designed to evaluate the atomic visual perception capabilities of Multimodal Large Language Models (MLLMs). Existing benchmarks often fail to isolate perception: holistic evaluations conflate perceptual errors with failures in reasoning or domain knowledge, while application-driven benchmarks only cover narrow, fragmented domains shaped by heu… ▽ More We introduce PerceptionBench, a benchmark specifically designed to evaluate the atomic visual perception capabilities of Multimodal Large Language Models (MLLMs). Existing benchmarks often fail to isolate perception: holistic evaluations conflate perceptual errors with failures in reasoning or domain knowledge, while application-driven benchmarks only cover narrow, fragmented domains shaped by heuristic designs. To address these limitations, PerceptionBench adopts a bottom-up approach: by diagnosing the earliest failure points in the responses of frontier MLLMs across 42 existing benchmarks, we construct an error taxonomy whose perception branch defines ten atomic perceptual capabilities. Guided by this taxonomy, we construct 3,000 verified questions with short, unambiguous answers, each isolating a single capability, with difficulty stemming from perception rather than reasoning or knowledge. Benchmark results across sixteen frontier MLLMs reveal that atomic perception remains largely unsolved---no model reaches 60\% accuracy, perception-related hallucination is the weakest capability on average, and similar overall scores conceal sharply divergent capability profiles. PerceptionBench thus provides a capability-level standard for measuring and diagnosing the visual perception boundaries of MLLMs. △ Less

Submitted 27 July, 2026; originally announced July 2026.

Kimi K3: Open Frontier Intelligence

Authors: Kimi Team, Tongtong Bai, Yifan Bai, Yiping Bao, M. C., Jianfeng Cai, Xinyuan Cai, Peizhou Cao, Yuxuan Cao, Ziwei Chai, Y. Charles, H. S. Che, Guanduo Chen, Guangyu Chen, Guanzheng Chen, Huarong Chen, Jia Chen, Jianlong Chen, Jun Chen, Kexin Chen, Peng Chen, Ruijue Chen, Wentao Chen, Xin Chen, Yang Chen , et al. (377 additional authors not shown)

Abstract: We introduce Kimi K3, a 2.8T parameter Mixture-of-Experts model with 104 billion activated parameters, native vision capabilities, and a 1-million-token context window. Kimi K3 is built on Kimi Delta Attention and Attention Residuals, which improve information flow across sequence length and model depth. Together with Stable LatentMoE, which effectively activates 16 of 896 routed experts per token… ▽ More We introduce Kimi K3, a 2.8T parameter Mixture-of-Experts model with 104 billion activated parameters, native vision capabilities, and a 1-million-token context window. Kimi K3 is built on Kimi Delta Attention and Attention Residuals, which improve information flow across sequence length and model depth. Together with Stable LatentMoE, which effectively activates 16 of 896 routed experts per token, and refined training and data recipes, these advances yield an approximately 2.5x improvement in overall scaling efficiency over Kimi K2. Post-training highlights reinforcement learning across general, agentic, and coding domains and multiple reasoning-effort levels, enabling compositional generalization and robust long-horizon execution. At 2.8T scale, Kimi K3 is supported by infrastructure advances in multiple areas: algorithm-system co-design for KDA, perfectly balanced expert-parallel training with efficient memory management, million-token agentic RL with persistent rollout and sandbox states, and deployment innovations. Extensive evaluations show that Kimi K3 achieves frontier-level performance across long-horizon coding, agentic, knowledge, reasoning, and vision tasks. While its overall performance still trails the most powerful proprietary models, namely Claude Fable 5 and GPT-5.6 Sol, Kimi K3 consistently outperforms other open and proprietary models evaluated in our suite. We release the full Kimi K3 model weights to facilitate future research and accelerate the broader deployment and adoption of frontier intelligence. △ Less

Submitted 7 August, 2026; v1 submitted 27 July, 2026; originally announced July 2026.

Comments: K3 tech report

Adaptive Driving Style for SAE Level-2 Driving Automation: Minimizing Preference Mismatch

Authors: Kumar Akash, Zhaobo Zheng, Teruhisa Misu, Vidya Krishnamoorthy, Mia Dong, Yuni Lee, Gaojian Huang

Abstract: Driving style is a key factor in the comfort and acceptance of automated vehicle (AV) features. In SAE Level-2 automation, where the driver must supervise the system and remain ready to intervene, mismatches between the automation's driving style and the driver's preference can reduce trust and trigger takeovers. This paper proposes an adaptive driving-style control framework that minimizes such p… ▽ More Driving style is a key factor in the comfort and acceptance of automated vehicle (AV) features. In SAE Level-2 automation, where the driver must supervise the system and remain ready to intervene, mismatches between the automation's driving style and the driver's preference can reduce trust and trigger takeovers. This paper proposes an adaptive driving-style control framework that minimizes such preference mismatch. In a driving-simulator study, we compare fixed, trust-based, and preference-based adaptation heuristics and analyze their effects on preference mismatch and trust. We then train a driving-preference prediction model and use it in an implicit adaptation policy that selects among bounded driving styles for upcoming events. A validation study shows that the predictive policy achieves equal or lower preference mismatch than comparison baselines, particularly when starting from a less defensive style, while also yielding higher average trust. The results provide a step toward developing human-aware driving automation that can implicitly adapt its driving style to the driver's preferences. △ Less

Submitted 23 July, 2026; originally announced July 2026.

Comments: Published in the 2026 American Control Conference (ACC), May 26-29, 2026, New Orleans, Louisiana, USA

EvoSQL: Memory-Augmented Critic-Generator Co-Evolution for Text-to-SQL

Authors: Jiawei Zhou, Jianwei Wang, Chenyu Zhou, Chaojian Shi, Ming Dong, Kai Wang

Abstract: Text-to-SQL has advanced rapidly with large language models, but complex database queries still require reasoning beyond one-shot generation, including multi-step decomposition, execution-based diagnosis, and targeted correction. We present EvoSQL, a co-evolution framework that formulates SQL synthesis as an iterative interaction between a generator and a critic. EvoSQL maintains a contextualized… ▽ More Text-to-SQL has advanced rapidly with large language models, but complex database queries still require reasoning beyond one-shot generation, including multi-step decomposition, execution-based diagnosis, and targeted correction. We present EvoSQL, a co-evolution framework that formulates SQL synthesis as an iterative interaction between a generator and a critic. EvoSQL maintains a contextualized candidate memory, verifies SQL candidates with both execution signals and LLM-based critique, and updates its memory through utility-guided aggregation. To strengthen the underlying generator-critic pair, we further introduce a Self-Distillation Policy Optimization (SDPO) fine-tuning stage that injects execution-aware supervision into modern coding LLM backbones. Experiments on Spider and BIRD show that EvoSQL consistently improves open-source models over Maj@16 baselines, with particularly large gains on BIRD-Dev, ranging from +1.37% for Qwen3-4B to +9.19% for Qwen2.5-Coder-3B. SDPO initialization further improves selected backbones on Spider-Test and BIRD-Dev. These results suggest that memory-grounded co-evolution is an effective path toward more reliable and generalizable Text-to-SQL systems. Code is available at https://github.com/valleysprings/EvoSQL. △ Less

Submitted 4 June, 2026; originally announced July 2026.

The Platonic Defense: Backdoor Defense for Self-Supervised Encoders in the Era of Large Scale Pre-training

Authors: Tuo Chen, Minjing Dong, Benlei Cui, Jian Liu, Jie Gui

Abstract: Self-supervised learning (SSL) pretrained models have become a dominant paradigm for visual representation learning, but they are vulnerable to backdoor attacks. Existing defenses struggle to defend against such attacks in a fully black-box setting because they often require access to labels, attack patterns, or training data. To tackle this issue, we propose a new attack-agnostic, model-agnostic,… ▽ More Self-supervised learning (SSL) pretrained models have become a dominant paradigm for visual representation learning, but they are vulnerable to backdoor attacks. Existing defenses struggle to defend against such attacks in a fully black-box setting because they often require access to labels, attack patterns, or training data. To tackle this issue, we propose a new attack-agnostic, model-agnostic, and modality-agnostic black-box test-time defense paradigm, called \emph{Platonic Representation Defense}. It is inspired by the Platonic Representation Hypothesis, which suggests that large-scale independently trained encoders converge toward compatible projections of the same underlying reality. We formalize this idea as a conditional energy function defined over source representations and a set of reference representations. The energy function is trained for detection through noise-contrastive estimation and for representation purification through denoising score matching. Theoretically, the energy gap between matched and mismatched samples is lower bounded by the mutual information between source and reference representations. We demonstrate the effectiveness of our method on multiple self-supervised encoders and more than 10 attacks. The method can perform both representation detection and purification, and achieves substantial performance gains across multiple attacks. Code is available \href{https://github.com/jsrdcht/Platonic-Representation-Defense}{here}. △ Less

Submitted 28 June, 2026; originally announced June 2026.

Equivariance and Augmentation for Bayesian Neural Networks

Authors: Miaowen Dong, Axel Flinth, Jan E. Gerken

Abstract: Symmetries are important for many deep learning tasks, ranging from applications in the sciences to medical imaging. However, there is an ongoing debate about whether to impose symmetry constraints on the neural network architecture (yielding equivariant neural networks) or learn them from augmented training data. Although equivariant networks are well-studied theoretically, much less is known abo… ▽ More Symmetries are important for many deep learning tasks, ranging from applications in the sciences to medical imaging. However, there is an ongoing debate about whether to impose symmetry constraints on the neural network architecture (yielding equivariant neural networks) or learn them from augmented training data. Although equivariant networks are well-studied theoretically, much less is known about data augmentation, since analyzing augmentation requires control over the training dynamics. Inspired by recent results that show that augmented infinite deep ensembles are exactly equivariant, we study data augmentation for Bayesian neural networks (BNNs) trained with variational inference. We focus on variational distributions in the exponential family and derive conditions under which exact equivariance is reached. We furthermore obtain bounds on the equivariance error and introduce three novel symmetrization techniques which boost the effect of data augmentation in this setting. We conduct extensive numerical experiments which show that one of our symmetrization methods (orbit expansion) outperforms the baseline in both equivariance and overall performance. Our code is available at github.com/dmw1998/augment-BNNs △ Less

Submitted 24 June, 2026; originally announced June 2026.

From Question Answering to Task Completion: A Survey on Agent System and Harness Design

Authors: Jianyuan Guo, Zhiwei Hao, Chengcheng Wang, Cheng Fan, Tingzhang Luo, Hongguang Li, Ying Gao, Hefei Mei, Jiankun Peng, Rongjian Xu, Minjing Dong, Han Wu, Mengyu Zheng, Kai Han, Shiqi Wang, Chang Xu, Yunhe Wang

Abstract: LLM-based agents mark a shift from passive question answering to active task completion: they perceive environments, invoke tools, maintain state, and act over extended horizons. As agent systems have evolved from prompt engineering to workflows and context engineering, harness engineering, and agent-native training with co-evolution, a central question has become increasingly important: where doe… ▽ More LLM-based agents mark a shift from passive question answering to active task completion: they perceive environments, invoke tools, maintain state, and act over extended horizons. As agent systems have evolved from prompt engineering to workflows and context engineering, harness engineering, and agent-native training with co-evolution, a central question has become increasingly important: where does the bottleneck in agent performance reside, in the foundation model, in the execution harness, or in the coupling between them? This survey examines LLM-based agents through a model-harness lens. We first clarify the functional definition of agents and the implementation view of an LLM-based agent as a foundation model coupled with an execution harness. We then analyze the limits of model-centric scaling, trace four paradigms of agent engineering, and decompose the execution harness into six coupled runtime responsibilities: observation, context, control, action, state, and verification. Using this decomposition, we map task properties and domain pressures to harness configurations, review benchmark and evaluation practices, and synthesize model-harness evidence on how runtime design affects long-horizon task completion, efficiency, and reliability. Finally, we identify open challenges in value-aware evaluation, safety, harness generalization, and model-harness co-evolution. Rather than treating agents as models with auxiliary tools, this survey argues that agent quality -- including success, efficiency, safety, and generalization -- emerges from the interaction between model capability, runtime infrastructure, task structure, and evaluation design. A collection of papers discussed in this survey is provided in https://github.com/ggjy/Awesome-Agent-Engineering. △ Less

Submitted 14 June, 2026; originally announced June 2026.

CoeusBI: A Comprehensive Interactive Business Intelligence System Powered by LLMs at Baidu [Extended Version]

Authors: Jinqing Lian, Chaofan Li, Yingxia Shao, Ming Wang, Yang Dong, Xinyi Liu, Wei Zhang, Chaoxian Gui, Tianqi Wan, Ming Dong

Abstract: The advent of Large Language Models has catalyzed the emergence of interactive Business Intelligence (BI) systems. Although commercial BI products increasingly adopt semantic layers paired with natural language interfaces, they predominantly rely on manual configurations to define metrics and dimensions. Real-world deployments face critical challenges: (a) frequent JOIN operations degrade the accu… ▽ More The advent of Large Language Models has catalyzed the emergence of interactive Business Intelligence (BI) systems. Although commercial BI products increasingly adopt semantic layers paired with natural language interfaces, they predominantly rely on manual configurations to define metrics and dimensions. Real-world deployments face critical challenges: (a) frequent JOIN operations degrade the accuracy of SQL generation; (b) wide schemas exacerbate the challenge of schema linking; and (c) the generation of dialect-specific queries and the accurate support for multi-round dialogues incur high computational costs and yield limited accuracy. We introduce CoeusBI, an industrial-scale interactive BI system that addresses these barriers through a novel Dual-Agent Architecture paired with a Hierarchical Schema Linking module: (1) an offline View Generation Agent that utilizes error-feedback to autonomously convert complex JOIN queries into simple single-view queries, which eliminates the need for manual semantic modeling; (2) a Hierarchical Schema Linking module that leverages vector retrieval over views to handle exceptionally wide schemas efficiently; and (3) a dynamic Routing Agent that evaluates dialogue contexts to route queries, dynamically invoking either the synthesis of new intermediate representations or targeted modifications of existing ones, before compiling the unified representation via a deterministic SQL compiler that is agnostic to dialects. Extensive experiments on both public datasets and production datasets demonstrate that CoeusBI achieves significant improvements in query accuracy, token efficiency, and user satisfaction relative to existing methods. CoeusBI is deployed as a standalone service on the data platform of Baidu and is widely used across multiple business lines supporting thousands of users daily, thereby evidencing strong practicality and scalability. △ Less

Submitted 13 June, 2026; originally announced June 2026.

Comments: Accepted by VLDB 2026

ACM Class: H.2.3

From Agent Traces to Trust: A Survey of Evidence Tracing and Execution Provenance in LLM Agents

Authors: Yiqi Wang, Jiaqi Zhang, Taotao Cai, Zirui Liu, Qingqiang Sun, Zequn Sun, Zhangkai Wu, Manqing Dong, Mingkai Zheng, Xuefei Yin, Yanming Zhu

Abstract: Large language model (LLM)-based agents are evolving from passive text generators into autonomous systems capable of planning, tool use, retrieval, memory access, environmental interaction, and multi-agent collaboration. These capabilities expand agent autonomy, but also make agent behavior harder to verify, debug, and audit. Final-answer accuracy alone cannot explain how an output was produced, w… ▽ More Large language model (LLM)-based agents are evolving from passive text generators into autonomous systems capable of planning, tool use, retrieval, memory access, environmental interaction, and multi-agent collaboration. These capabilities expand agent autonomy, but also make agent behavior harder to verify, debug, and audit. Final-answer accuracy alone cannot explain how an output was produced, which evidence supported each claim, whether tool calls were justified, how memory influenced later decisions, or where failures originated. This survey examines evidence tracing and execution provenance as foundations for process-level accountability in trustworthy LLM agents. We define execution provenance as the typed graph of an agent execution and evidence tracing as its projection onto evidence-support relations. This perspective connects retrieval grounding, claim support, tool-use safety, memory lineage, observability, debugging, audit, and recovery within a unified framework. We introduce a taxonomy covering trace sources, evidence and execution units, provenance relations, tracing granularity and timing, representation forms, and trust functions. We then review key methodological directions, including provenance representation, evidence attribution, tool-use provenance, runtime guardrails, provenance-bearing memory, observability, and failure diagnosis. Finally, we discuss benchmarks, datasets, metrics, and open challenges for building provenance-aware, auditable, and recoverable agent systems. △ Less

Submitted 28 June, 2026; v1 submitted 3 June, 2026; originally announced June 2026.

SAMatcher: Co-Visibility Modeling with Segment Anything for Robust Feature Matching

Authors: Xu Pan, Qiyuan Ma, Mingyue Dong, He Chen, Wei Ji, Xianwei Zheng

Abstract: Reliable correspondence estimation is a fundamental problem in image processing, underpinning applications such as Structure from Motion, visual localization, and image registration. Existing learning-based methods have significantly improved local feature representations, yet most still operate at the pixel or patch level and lack explicit modeling of regions that are jointly visible across views… ▽ More Reliable correspondence estimation is a fundamental problem in image processing, underpinning applications such as Structure from Motion, visual localization, and image registration. Existing learning-based methods have significantly improved local feature representations, yet most still operate at the pixel or patch level and lack explicit modeling of regions that are jointly visible across views. We propose SAMatcher, a feature matching framework that formulates correspondence estimation through co-visibility modeling. Instead of directly matching local features, SAMatcher first predicts co-visible region masks and bounding boxes as structured priors for correspondence estimation. Built upon the Segment Anything Model (SAM), it introduces a symmetric cross-view interaction mechanism that enables bidirectional feature exchange and cross-view semantic alignment. We further develop a unified supervision scheme that jointly optimizes mask prediction and box localization through mask learning, box regression, and mask-box consistency constraints. Extensive experiments on challenging benchmarks demonstrate substantial improvements over existing matching pipelines, particularly under large viewpoint and scale variations. Our results show that foundation models originally designed for monocular segmentation can be effectively extended to multi-view correspondence reasoning through explicit co-visibility modeling, offering a new perspective on structured representation learning for image matching. Code and project page: https://xupan.top/Projects/samatcher △ Less

Submitted 2 June, 2026; originally announced June 2026.

Comments: 14 pages

Once-For-All: A Train-Once and Select-Anytime Framework for Multimodal Instruction Tuning

Authors: Mingkang Dong, Hongyi Cai, Xiwen Lei, Jie Li, Tao Zhang, Muxin Pu

Abstract: Multimodal instruction tuning is the de facto recipe for adapting vision language models (VLMs), yet instruction data are highly redundant, making data selection critical for training efficiency. Existing methods derive selection signals from a specific model or dataset, so whenever the target model or candidate pool changes, the criteria must be recomputed from scratch at substantial cost. To add… ▽ More Multimodal instruction tuning is the de facto recipe for adapting vision language models (VLMs), yet instruction data are highly redundant, making data selection critical for training efficiency. Existing methods derive selection signals from a specific model or dataset, so whenever the target model or candidate pool changes, the criteria must be recomputed from scratch at substantial cost. To address this, we propose OFA, a data selection framework that trains a reusable selector once and applies it to any dataset or model without recomputation. OFA clusters multimodal instructions in a frozen CLIP space, derives pseudo labels from the cluster structure, and trains a lightweight selector for only a few epochs; samples on which this selector is least confident are selected as the most informative. Once trained, the frozen selector transfers directly across datasets and model scales. The selector is trained once on LLaVA-665K and applied both to LLaVA-665K itself and, without any retraining, to the unseen Vision-Flan-186K. Selecting only 15% of the data, OFA achieves 98.3% of full data performance across 10 downstream benchmarks; on the smaller Vision-Flan-186K, the transferred selector surpasses full data training by 10.6%, confirming that the learned signal generalizes to datasets never seen during selector training. The same selected subsets benefit VLMs at both Qwen2.5-VL-3B and LLaVA-v1.5-7B without per model recomputation, decoupling selection from the target model. These results demonstrate that a single, transferable selector provides an effective and reusable solution for efficient multimodal instruction tuning. △ Less

Submitted 4 June, 2026; v1 submitted 26 May, 2026; originally announced May 2026.

Comments: 15 pages, 6 figures. Mingkang Dong and Hongyi Cai contributed equally to this work. Muxin Pu is the corresponding author

ACM Class: I.2.6; I.2.10; I.4.0

Retrieval-Augmented Linguistic Calibration

Authors: Yi-Fan Yeh, Linwei Tao, Minjing Dong, Tao Huang, Jialin Yu, Philip Torr, Chang Xu

Abstract: Linguistic cues such as "I believe" and "probably" offer an intuitive interface for communicating confidence, yet a generalisable, principled calibration framework for linguistic confidence expressions remains underexplored. In particular, co-occurring linguistic cues, contextual variation, and subjective audience interpretation pose unique challenges. We therefore model linguistic confidence as a… ▽ More Linguistic cues such as "I believe" and "probably" offer an intuitive interface for communicating confidence, yet a generalisable, principled calibration framework for linguistic confidence expressions remains underexplored. In particular, co-occurring linguistic cues, contextual variation, and subjective audience interpretation pose unique challenges. We therefore model linguistic confidence as a distribution over plausible perceived probability values that a statement is correct, capturing interpretation variability that scalar representations discard. Within this distributional framework, we introduce faithfulness as a complementary evaluation dimension and present Faithfulness Divergence (FD), an information-theoretic metric quantifying the surprise induced in audience beliefs upon truth revelation. Building on these foundations, we present Retrieval-Augmented Linguistic Calibration (RALC), a lightweight post-hoc pipeline that propagates calibrated confidence signals back into natural language via retrieval-augmented rewriting. Across three QA benchmarks and five LLM families, RALC improves in-domain faithfulness and calibration up to 66% and 58%, respectively, outperforming black-box and grey-box calibration baselines. △ Less

Submitted 29 May, 2026; v1 submitted 19 May, 2026; originally announced May 2026.

Evo-Depth: A Lightweight Depth-Enhanced Vision-Language-Action Model

Authors: Tao Lin, Yuxin Du, Jiting Liu, Nuobei Zhu, Yunhe Li, Yuqian Fu, Yinxinyu Chen, Hongyi Cai, Zewei Ye, Bing Cheng, Kai Ye, Yiran Mao, Yilei Zhong, MingKang Dong, Junchi Yan, Gen Li, Bo Zhao

Abstract: Vision-Language-Action models have emerged as a promising paradigm for robotic manipulation by unifying perception, language grounding, and action generation. However, they often struggle in scenarios requiring precise spatial understanding, as current VLA models primarily rely on 2D visual representations that lack depth information and detailed spatial relationships. While recent approaches inco… ▽ More Vision-Language-Action models have emerged as a promising paradigm for robotic manipulation by unifying perception, language grounding, and action generation. However, they often struggle in scenarios requiring precise spatial understanding, as current VLA models primarily rely on 2D visual representations that lack depth information and detailed spatial relationships. While recent approaches incorporate explicit 3D inputs such as depth maps or point clouds to address this issue, they often increase system complexity, require additional sensors, and remain vulnerable to sensing noise and reconstruction errors. Another line of work explores implicit 3D-aware spatial modeling directly from RGB observations without extra sensors, but it often relies on large geometry foundation models, resulting in higher training and deployment costs. To address these challenges, we propose Evo-Depth, a lightweight depth-enhanced VLA framework that enhances spatially grounded manipulation without relying on additional sensing hardware or compromising deployment efficiency. Evo-Depth employs a lightweight Implicit Depth Encoding Module to extract compact depth features from multi-view RGB images. These features are incorporated into vision-language representations through a Spatial Enhancement Module via depth-aware modulation, enabling efficient spatial-semantic enhancement. A Progressive Alignment Training strategy is further introduced to align the resulting depth-enhanced representations with downstream action learning. With only 0.9B parameters, Evo-Depth achieves superior performance across four simulation benchmarks. In real-world experiments, Evo-Depth attains the highest average success rate while also exhibiting the smallest model size, lowest GPU memory usage, and highest inference frequency among compared methods. △ Less

Submitted 14 May, 2026; originally announced May 2026.

HDRFace: Rethinking Face Restoration with High-Dimensional Representation

Authors: Zirui Wang, Xianhui Lin, Yi Dong, Bo Wei, Gangjian Zhang, Siteng Ma, Zebiao Zheng, Xing Liu, Hong Gu, Minjing Dong

Abstract: Face restoration under complex degradations still remains an ill-posed inverse problem due to severe information loss. Although diffusion models benefit from strong generative priors, most methods still condition only on low-quality inputs, making it difficult to recover identity-critical details under heavy degradations. In this work, we propose HDRFace, a High-Dimensional Representation conditio… ▽ More Face restoration under complex degradations still remains an ill-posed inverse problem due to severe information loss. Although diffusion models benefit from strong generative priors, most methods still condition only on low-quality inputs, making it difficult to recover identity-critical details under heavy degradations. In this work, we propose HDRFace, a High-Dimensional Representation conditioned Face restoration framework that injects semantically rich priors into the conditional flow without modifying the generative backbone. Our pipeline first obtains a structurally reliable intermediate restoration with an off-the-shelf restorer, then uses a pretrained high-dimensional feature encoder to extract fine-grained facial representations from both the low-quality input and the intermediate result, and injects them as additional conditions for generation. We further introduce SDFM, a Structure-Detail aware adaptive Fusion Mechanism that emphasizes global constraints during structure modeling and strengthens representation guidance during detail synthesis, balancing structural consistency and detail fidelity. To validate the generalization ability of our method, we implement the proposed framework on two generative models, SD V2.1-base and Qwen-Image, and consistently observe stable and coherent performance gains across different architectures. △ Less

Submitted 14 May, 2026; originally announced May 2026.

Probing CLIP's Comprehension of 360-Degree Textual and Visual Semantics

Authors: Hai Wang, Xiaochen Yang, Mingzhi Dong, Jing-Hao Xue

Abstract: The dream of instantly creating rich 360-degree panoramic worlds from text is rapidly becoming a reality, yet a crucial gap exists in our ability to reliably evaluate their semantic alignment. Contrastive Language-Image Pre-training (CLIP) models, standard AI evaluators, predominantly trained on perspective image-text pairs, face an open question regarding their understanding of the unique charact… ▽ More The dream of instantly creating rich 360-degree panoramic worlds from text is rapidly becoming a reality, yet a crucial gap exists in our ability to reliably evaluate their semantic alignment. Contrastive Language-Image Pre-training (CLIP) models, standard AI evaluators, predominantly trained on perspective image-text pairs, face an open question regarding their understanding of the unique characteristics of 360-degree panoramic image-text pairs. This paper addresses this gap by first introducing two concepts: \emph{360-degree textual semantics}, semantic information conveyed by explicit format identifiers, and \emph{360-degree visual semantics}, invariant semantics under horizontal circular shifts. To probe CLIP's comprehension of these semantics, we then propose novel evaluation methodologies using keyword manipulation and horizontal circular shifts of varying magnitudes. Rigorous statistical analyses across popular CLIP configurations reveal that: (1) CLIP models effectively leverage explicit textual identifiers, demonstrating an understanding of 360-degree textual semantics; and (2) CLIP models fail to robustly preserve semantic alignment under horizontal circular shifts, indicating limited comprehension of 360-degree visual semantics. To address this limitation, we propose a LoRA-based fine-tuning framework that explicitly instills invariance to circular shifts. Our fine-tuned models exhibit improved comprehension of 360-degree visual semantics, though with a slight degradation in original semantic evaluation performance, highlighting a fundamental trade-off in adapting CLIP to 360-degree panoramic images. Code is available at https://github.com/littlewhitesea/360Semantics. △ Less

Submitted 27 April, 2026; originally announced April 2026.

Comments: Project Page: https://littlewhitesea.github.io/360Semantics.github.io/

ReplicateAnyScene: Zero-Shot Video-to-3D Composition via Textual-Visual-Spatial Alignment

Authors: Mingyu Dong, Chong Xia, Mingyuan Jia, Weichen Lyu, Long Xu, Zheng Zhu, Yueqi Duan

Abstract: Humans exhibit an innate capacity to rapidly perceive and segment objects from video observations, and even mentally assemble them into structured 3D scenes. Replicating such capability, termed compositional 3D reconstruction, is pivotal for the advancement of Spatial Intelligence and Embodied AI. However, existing methods struggle to achieve practical deployment due to the insufficient integratio… ▽ More Humans exhibit an innate capacity to rapidly perceive and segment objects from video observations, and even mentally assemble them into structured 3D scenes. Replicating such capability, termed compositional 3D reconstruction, is pivotal for the advancement of Spatial Intelligence and Embodied AI. However, existing methods struggle to achieve practical deployment due to the insufficient integration of cross-modal information, leaving them dependent on manual object prompting, reliant on auxiliary visual inputs, and restricted to overly simplistic scenes by training biases. To address these limitations, we propose ReplicateAnyScene, a framework capable of fully automated and zero-shot transformation of casually captured videos into compositional 3D scenes. Specifically, our pipeline incorporates a five-stage cascade to extract and structurally align generic priors from vision foundation models across textual, visual, and spatial dimensions, grounding them into structured 3D representations and ensuring semantic coherence and physical plausibility of the constructed scenes. To facilitate a more comprehensive evaluation of this task, we further introduce the C3DR benchmark to assess reconstruction quality from diverse aspects. Extensive experiments demonstrate the superiority of our method over existing baselines in generating high-quality compositional 3D scenes. △ Less

Submitted 12 April, 2026; originally announced April 2026.

Comments: Project Page: https://xiac20.github.io/ReplicateAnyScene/

Q-Zoom: Query-Aware Adaptive Perception for Efficient Multimodal Large Language Models

Authors: Yuheng Shi, Xiaohuan Pei, Linfeng Wen, Minjing Dong, Chang Xu

Abstract: MLLMs require high-resolution visual inputs for fine-grained tasks like document understanding and dense scene perception. However, current global resolution scaling paradigms indiscriminately flood the quadratic self-attention mechanism with visually redundant tokens, severely bottlenecking inference throughput while ignoring spatial sparsity and query intent. To overcome this, we propose Q-Zoom,… ▽ More MLLMs require high-resolution visual inputs for fine-grained tasks like document understanding and dense scene perception. However, current global resolution scaling paradigms indiscriminately flood the quadratic self-attention mechanism with visually redundant tokens, severely bottlenecking inference throughput while ignoring spatial sparsity and query intent. To overcome this, we propose Q-Zoom, a query-aware adaptive high-resolution perception framework that operates in an efficient coarse-to-fine manner. First, a lightweight Dynamic Gating Network safely bypasses high-resolution processing when coarse global features suffice. Second, for queries demanding fine-grained perception, a Self-Distilled Region Proposal Network (SD-RPN) precisely localizes the task-relevant Region-of-Interest (RoI) directly from intermediate feature spaces. To optimize these modules efficiently, the gating network uses a consistency-aware generation strategy to derive deterministic routing labels, while the SD-RPN employs a fully self-supervised distillation paradigm. A continuous spatio-temporal alignment scheme and targeted fine-tuning then seamlessly fuse the dense local RoI with the coarse global layout. Extensive experiments demonstrate that Q-Zoom establishes a dominant Pareto frontier. Using Qwen2.5-VL-7B as a primary testbed, Q-Zoom accelerates inference by 2.52 times on Document & OCR benchmarks and 4.39 times in High-Resolution scenarios while matching the baseline's peak accuracy. Furthermore, when configured for maximum perceptual fidelity, Q-Zoom surpasses the baseline's peak performance by 1.1% and 8.1% on these respective benchmarks. These robust improvements transfer seamlessly to Qwen3-VL, LLaVA, and emerging RL-based thinking-with-image models. Project page is available at https://yuhengsss.github.io/Q-Zoom/. △ Less

Submitted 8 April, 2026; originally announced April 2026.

Comments: 16 pages, 9 figures

Poison Once, Exploit Forever: Environment-Injected Memory Poisoning Attacks on Web Agents

Authors: Wei Zou, Mingwen Dong, Miguel Romero Calvo, Shuaichen Chang, Jiang Guo, Dongkyu Lee, Xing Niu, Xiaofei Ma, Yanjun Qi, Jiarong Jiang

Abstract: Memory makes LLM-based web agents personalized, powerful, yet exploitable. By storing past interactions to personalize future tasks, agents inadvertently create a persistent attack surface that spans websites and sessions. While existing security research on memory assumes attackers can directly inject into memory storage or exploit shared memory across users, we present a more realistic threat mo… ▽ More Memory makes LLM-based web agents personalized, powerful, yet exploitable. By storing past interactions to personalize future tasks, agents inadvertently create a persistent attack surface that spans websites and sessions. While existing security research on memory assumes attackers can directly inject into memory storage or exploit shared memory across users, we present a more realistic threat model: contamination through environmental observation alone. We introduce Environment-injected Trajectory-based Agent Memory Poisoning (eTAMP), the first attack to achieve cross-session, cross-site compromise without requiring direct memory access. A single contaminated observation (e.g., viewing a manipulated product page) silently poisons an agent's memory and activates during future tasks on different websites, bypassing permission-based defenses. Our experiments on (Visual)WebArena reveal two key findings. First, eTAMP achieves substantial attack success rates: up to 32.5% on GPT-5-mini, 23.4% on GPT-5.2, and 19.5% on GPT-OSS-120B. Second, we discover Frustration Exploitation: agents under environmental stress become dramatically more susceptible, with ASR increasing up to 8 times when agents struggle with dropped clicks or garbled text. Notably, more capable models are not more secure. GPT-5.2 shows substantial vulnerability despite superior task performance. With the rise of AI browsers like OpenClaw, ChatGPT Atlas, and Perplexity Comet, our findings underscore the urgent need for defenses against environment-injected memory poisoning. △ Less

Submitted 7 April, 2026; v1 submitted 2 April, 2026; originally announced April 2026.

Confidence Calibration under Ambiguous Ground Truth

Authors: Linwei Tao, Haoyang Luo, Minjing Dong, Chang Xu

Abstract: Confidence calibration assumes a unique ground-truth label per input, yet this assumption fails wherever annotators genuinely disagree. Post-hoc calibrators fitted on majority-voted labels, the standard single-label targets used in practice, can appear well-calibrated under conventional evaluation yet remain substantially miscalibrated against the underlying annotator distribution. We show that th… ▽ More Confidence calibration assumes a unique ground-truth label per input, yet this assumption fails wherever annotators genuinely disagree. Post-hoc calibrators fitted on majority-voted labels, the standard single-label targets used in practice, can appear well-calibrated under conventional evaluation yet remain substantially miscalibrated against the underlying annotator distribution. We show that this failure is structural: under simplifying assumptions, Temperature Scaling is biased toward temperatures that underestimate annotator uncertainty, with true-label miscalibration increasing monotonically with annotation entropy. To address this, we develop a family of ambiguity-aware post-hoc calibrators that optimise proper scoring rules against the full label distribution and require no model retraining. Our methods span progressively weaker annotation requirements: Dirichlet-Soft leverages the full annotator distribution and achieves the best overall calibration quality across settings; Monte Carlo Temperature Scaling with a single annotation per example (MCTS S=1) matches full-distribution calibration across all benchmarks, demonstrating that pre-aggregated label distributions are unnecessary; and Label-Smooth Temperature Scaling (LS-TS) operates with voted labels alone by constructing data-driven pseudo-soft targets from the model's own confidence. Experiments on four benchmarks with real multi-annotator distributions (CIFAR-10H, ChaosNLI) and clinically-informed synthetic annotations (ISIC~2019, DermaMNIST) show that Dirichlet-Soft reduces true-label ECE by 55-87% relative to Temperature Scaling, while LS-TS reduces ECE by 9-77% without any annotator data. △ Less

Submitted 24 March, 2026; originally announced March 2026.

Bridging neuroscience and AI: adaptive, culturally sensitive technologies transforming aphasia rehabilitation

Authors: Andreea I. Niculescu, Jochen Ehnes, Minghui Dong

Abstract: Aphasia, a language impairment primarily resulting from stroke or brain injury, profoundly disrupts communication and everyday functioning. Despite advances in speech therapy, barriers such as limited therapist availability and the scarcity of personalized, culturally relevant tools continue to hinder optimal rehabilitation outcomes. This paper reviews recent developments in neurocognitive researc… ▽ More Aphasia, a language impairment primarily resulting from stroke or brain injury, profoundly disrupts communication and everyday functioning. Despite advances in speech therapy, barriers such as limited therapist availability and the scarcity of personalized, culturally relevant tools continue to hinder optimal rehabilitation outcomes. This paper reviews recent developments in neurocognitive research and language technologies that contribute to the diagnosis and therapy of aphasia. Drawing on findings from our ethnographic field study, we introduce two digital therapy prototypes designed to reflect local linguistic diversity and enhance patient engagement. We also show how insights from neuroscience and the local context guided the design of these tools to better meet patient and therapist needs. Our work highlights the potential of adaptive, AI-enhanced assistive technologies to complement conventional therapy and broaden access to therapy. We conclude by outlining future research directions for advancing personalized and scalable aphasia rehabilitation. △ Less

Submitted 22 March, 2026; originally announced March 2026.

Comments: 12 pages, 2 figures, Proceedings of the 20th International Conference on linguistic resources and tools for natural language processing (ConsILR 2025)

ACM Class: H.5.2; J.3

InterveneBench: Benchmarking LLMs for Intervention Reasoning and Causal Study Design in Real Social Systems

Authors: Shaojie Shi, Zhengyu Shi, Lingran Zheng, Xinyu Su, Anna Xie, Bohao Lv, Rui Xu, Zijian Chen, Zhichao Chen, Guolei Liu, Naifu Zhang, Mingjian Dong, Zhuo Quan, Bohao Chen, Teqi Hao, Yuan Qi, Yinghui Xu, Libo Wu

Abstract: Causal inference in social science relies on end-to-end, intervention-centered research-design reasoning grounded in real-world policy interventions, but current benchmarks fail to evaluate this capability of large language models (LLMs). We present InterveneBench, a benchmark designed to assess such reasoning in realistic social settings. Each instance in InterveneBench is derived from an empiric… ▽ More Causal inference in social science relies on end-to-end, intervention-centered research-design reasoning grounded in real-world policy interventions, but current benchmarks fail to evaluate this capability of large language models (LLMs). We present InterveneBench, a benchmark designed to assess such reasoning in realistic social settings. Each instance in InterveneBench is derived from an empirical social science study and requires models to reason about policy interventions and identification assumptions without access to predefined causal graphs or structural equations. InterveneBench comprises 744 peer-reviewed studies across diverse policy domains. Experimental results show that state-of-the-art LLMs struggle under this setting. To address this limitation, we further propose a multi-agent framework, STRIDES. It achieves significant performance improvements over state-of-the-art reasoning models. Our code and data are available at https://github.com/Sii-yuning/STRIDES. △ Less

Submitted 16 March, 2026; originally announced March 2026.

Comments: 35pages,3 figures

The Curse and Blessing of Mean Bias in FP4-Quantized LLM Training

Authors: Hengjie Cao, Zhendong Huang, Mengyi Chen, Yifeng Yang, Fang Dong, Anrui Chen, Ruijun Huang, Xin Zhang, Mingzhi Dong, Yujiang Wang, Jinlong Hou, Qin Lv, Robert P. Dick, Yuan Cheng, Tun Lu, Fan Yang, Yixuan Chen, Li Shang

Abstract: FP4 training promises substantial memory and compute savings for large language models, but remains fragile because blockwise quantization is dictated by extreme activation magnitudes, which inflate dynamic range and compress long-tail signals. We identify a counterintuitive source of this failure: dominant activation outliers are not merely arbitrary sparse events, but are largely induced by a co… ▽ More FP4 training promises substantial memory and compute savings for large language models, but remains fragile because blockwise quantization is dictated by extreme activation magnitudes, which inflate dynamic range and compress long-tail signals. We identify a counterintuitive source of this failure: dominant activation outliers are not merely arbitrary sparse events, but are largely induced by a coherent rank-one mean bias, whose direction aligns with the leading anisotropic spectral component. This mean component strengthens during training, is amplified and reshaped by attention and FFN operators, and increasingly dominates top activation magnitudes. Crucially, this discovery reveals that a seemingly complex outlier-suppression problem admits a truly simple solution: isolate the coherent mean before quantization. We therefore propose Averis, a mean-residual splitting quantization method that separates the mean component using only reductions and elementwise subtractions before FP4 quantization. Across Qwen3 0.6B Dense trained on 100B tokens and Qwen3 7B A1.5B MoE trained on 50B tokens, Averis enables robust W4A4G4 FP4 training, reducing BF16 loss gaps to 1.19%/0.81% versus 2.05%/1.10% for NVIDIA's recently released Hadamard-based outlier-smoothing method, while limiting downstream gaps to 0.89/0.71 points. With only 2.20% end-to-end overhead over vanilla NVFP4, about 30% of NVIDIA's Hadamard-based design, Averis provides a hardware-efficient path to stable low-bit LLM training. Complementary to Hadamard, Averis further reduces the Qwen3-0.6B loss and downstream gaps to 0.94% and 0.73 points when combined. Code is available at: https://anonymous.4open.science/r/averis-504D. △ Less

Submitted 12 June, 2026; v1 submitted 11 March, 2026; originally announced March 2026.

SI-ChainFL: Shapley-Incentivized Secure Federated Learning for High-Speed Rail Data Sharing

Authors: Mingjie Zhao, Cheng Dai, Fei Chen, Xin Chen, Kaoru Ota, Mianxiong Dong, Bing Guo

Abstract: In high-speed rail (HSR) systems, federated learning (FL) enables cross-departmental flow prediction without sharing raw data. However, existing schemes suffer from two key limitations: (1) insufficient incentives, leading to free-riding and model poisoning; and (2) centralized aggregation, which introduces a single point of failure. We propose a secure and efficient framework SI-ChainFL that addr… ▽ More In high-speed rail (HSR) systems, federated learning (FL) enables cross-departmental flow prediction without sharing raw data. However, existing schemes suffer from two key limitations: (1) insufficient incentives, leading to free-riding and model poisoning; and (2) centralized aggregation, which introduces a single point of failure. We propose a secure and efficient framework SI-ChainFL that addresses these issues by combining contribution-aware incentives with decentralized aggregation. First, we quantify client contributions using a Shapley value metric that jointly considers rare-event utility, data diversity, data quality, and timeliness. To reduce computational overhead, we further develop a rare positive driven client clustering strategy to accelerate Shapley estimation. Moreover, we design a blockchain-based consensus protocol for decentralized aggregation, where aggregation eligibility is tied to Shapley incentives. This design motivates clients to submit high-quality updates and enables efficient and secure global aggregation. Experiments on MNIST, CIFAR 10 and CIFAR 100, and a HSR flow dataset show that SI ChainFL remains effective under 90% malicious clients in PA attacks, achieving 14.12% higher accuracy than RAGA. Theoretical analysis further guarantees an upper bound on performance △ Less

Submitted 9 March, 2026; originally announced March 2026.

Comments: 17 pages, 19 figures

MSC Class: 68M14 ACM Class: I.2.11

VisNec: Measuring and Leveraging Visual Necessity for Multimodal Instruction Tuning

Authors: Mingkang Dong, Hongyi Cai, Jie Li, Sifan Zhou, Bin Ren, Kunyu Peng, Yuqian Fu

Abstract: The effectiveness of multimodal instruction tuning depends not only on dataset scale, but critically on whether training samples genuinely require visual reasoning. However, existing instruction datasets often contain a substantial portion of visually redundant samples (solvable from text alone), as well as multimodally misaligned supervision that can degrade learning. To address this, we propose… ▽ More The effectiveness of multimodal instruction tuning depends not only on dataset scale, but critically on whether training samples genuinely require visual reasoning. However, existing instruction datasets often contain a substantial portion of visually redundant samples (solvable from text alone), as well as multimodally misaligned supervision that can degrade learning. To address this, we propose VisNec (Visual Necessity Score), a principled data selection framework that measures the marginal contribution of visual input during instruction tuning. By comparing predictive loss with and without visual context, VisNec identifies whether a training instance is vision-critical, redundant, or misaligned. To preserve task diversity, we combine VisNec with semantic clustering and select high-necessity samples within each cluster. Across 10 downstream benchmarks, training on only 15% of the LLaVA-665K dataset selected by VisNec achieves 100.2% of full-data performance. On the smaller Vision-Flan-186K dataset, our selection not only further reduces data size but also surpasses full-data training by 15.8%. These results demonstrate that measuring and leveraging visual necessity provides an effective solution for both efficient and robust multimodal instruction tuning. Codes and selected subsets will be released upon acceptance. △ Less

Submitted 25 June, 2026; v1 submitted 1 March, 2026; originally announced March 2026.

Comments: Accepted at ECCV 2026. Project Page: https://dmk041218.github.io/VisNec/

ACM Class: I.2.10; I.2.6; I.2.7

PA-Attack: Guiding Gray-Box Attacks on LVLM Vision Encoders with Prototypes and Attention

Authors: Hefei Mei, Zirui Wang, Chang Xu, Jianyuan Guo, Minjing Dong

Abstract: Large Vision-Language Models (LVLMs) are foundational to modern multimodal applications, yet their susceptibility to adversarial attacks remains a critical concern. Prior white-box attacks rarely generalize across tasks, and black-box methods depend on expensive transfer, which limits efficiency. The vision encoder, standardized and often shared across LVLMs, provides a stable gray-box pivot with… ▽ More Large Vision-Language Models (LVLMs) are foundational to modern multimodal applications, yet their susceptibility to adversarial attacks remains a critical concern. Prior white-box attacks rarely generalize across tasks, and black-box methods depend on expensive transfer, which limits efficiency. The vision encoder, standardized and often shared across LVLMs, provides a stable gray-box pivot with strong cross-model transfer. Building on this premise, we introduce PA-Attack (Prototype-Anchored Attentive Attack). PA-Attack begins with a prototype-anchored guidance that provides a stable attack direction towards a general and dissimilar prototype, tackling the attribute-restricted issue and limited task generalization of vanilla attacks. Building on this, we propose a two-stage attention enhancement mechanism: (i) leverage token-level attention scores to concentrate perturbations on critical visual tokens, and (ii) adaptively recalibrate attention weights to track the evolving attention during the adversarial process. Extensive experiments across diverse downstream tasks and LVLM architectures show that PA-Attack achieves an average 75.1% score reduction rate (SRR), demonstrating strong attack effectiveness, efficiency, and task generalization in LVLMs. Code is available at https://github.com/hefeimei06/PA-Attack. △ Less

Submitted 22 February, 2026; originally announced February 2026.

Multi-Head Attention as a Source of Catastrophic Forgetting in MoE Transformers

Authors: Anrui Chen, Ruijun Huang, Xin Zhang, Fang Dong, Hengjie Cao, Zhendong Huang, Yifeng Yang, Mengyi Chen, Jixian Zhou, Mingzhi Dong, Yujiang Wang, Jinlong Hou, Qin Lv, Robert P. Dick, Yuan Cheng, Tun Lu, Fan Yang, Li Shang

Abstract: Mixture-of-Experts (MoE) architectures are often considered a natural fit for continual learning because sparse routing should localize updates and reduce interference, yet MoE Transformers still forget substantially even with sparse, well-balanced expert utilization. We attribute this gap to a pre-routing bottleneck: multi-head attention concatenates head-specific signals into a single post-atten… ▽ More Mixture-of-Experts (MoE) architectures are often considered a natural fit for continual learning because sparse routing should localize updates and reduce interference, yet MoE Transformers still forget substantially even with sparse, well-balanced expert utilization. We attribute this gap to a pre-routing bottleneck: multi-head attention concatenates head-specific signals into a single post-attention router input, forcing routing to act on co-occurring feature compositions rather than separable head channels. We show that this router input simultaneously encodes multiple separately decodable semantic and structural factors with uneven head support, and that different feature compositions induce weakly aligned parameter-gradient directions; as a result, routing maps many distinct compositions to the same route. We quantify this collision effect via a route-wise effective composition number $N_{eff}$ and find that higher $N_{eff}$ is associated with larger old-task loss increases after continual training. Motivated by these findings, we propose MH-MoE, which performs head-wise routing over sub-representations to increase routing granularity and reduce composition collisions. On TRACE with Qwen3-0.6B/8B, MH-MoE effectively mitigates forgetting, reducing BWT on Qwen3-0.6B from 11.2% (LoRAMoE) to 4.5%. △ Less

Submitted 12 February, 2026; originally announced February 2026.

SD-MoE: Spectral Decomposition for Effective Expert Specialization

Authors: Ruijun Huang, Fang Dong, Xin Zhang, Hengjie Cao, Zhendong Huang, Anrui Chen, Jixian Zhou, Mengyi Chen, Yifeng Yang, Mingzhi Dong, Yujiang Wang, Jinlong Hou, Qin Lv, Robert P. Dick, Yuan Cheng, Fan Yang, Tun Lu, Chun Zhang, Li Shang

Abstract: Mixture-of-Experts (MoE) architectures scale Large Language Models via expert specialization induced by conditional computation. In practice, however, expert specialization often fails: some experts become functionally similar, while others functioning as de facto shared experts, limiting the effective capacity and model performance. In this work, we analysis from a spectral perspective on paramet… ▽ More Mixture-of-Experts (MoE) architectures scale Large Language Models via expert specialization induced by conditional computation. In practice, however, expert specialization often fails: some experts become functionally similar, while others functioning as de facto shared experts, limiting the effective capacity and model performance. In this work, we analysis from a spectral perspective on parameter and gradient spaces, uncover that (1) experts share highly overlapping dominant spectral components in their parameters, (2) dominant gradient subspaces are strongly aligned across experts, driven by ubiquitous low-rank structure in human corpus, and (3) gating mechanisms preferentially route inputs along these dominant directions, further limiting specialization. To address this, we propose Spectral-Decoupled MoE (SD-MoE), which decomposes both parameter and gradient in the spectral space. SD-MoE improves performance across downstream tasks, enables effective expert specialization, incurring minimal additional computation, and can be seamlessly integrated into a wide range of existing MoE architectures, including Qwen and DeepSeek. △ Less

Submitted 12 February, 2026; originally announced February 2026.

Spectra: Rethinking Optimizers for LLMs Under Spectral Anisotropy

Authors: Zhendong Huang, Hengjie Cao, Fang Dong, Ruijun Huang, Mengyi Chen, Yifeng Yang, Xin Zhang, Anrui Chen, Mingzhi Dong, Yujiang Wang, Jinlong Hou, Qin Lv, Robert P. Dick, Yuan Cheng, Fan Yang, Tun Lu, Li Shang

Abstract: Gradient signals in LLM training are highly anisotropic: recurrent linguistic structure concentrates energy into a small set of dominant spectral directions, while context specific information resides in a long tail. We show that this spike tail separation persists throughout training, with the spike occupying only about 1.5% of directions yet dominating optimizer statistics. This dominance suppre… ▽ More Gradient signals in LLM training are highly anisotropic: recurrent linguistic structure concentrates energy into a small set of dominant spectral directions, while context specific information resides in a long tail. We show that this spike tail separation persists throughout training, with the spike occupying only about 1.5% of directions yet dominating optimizer statistics. This dominance suppresses tail learning by contracting tail updates through second moment normalization and tightening the globally stable learning rate bound. Motivated by this analysis, we propose Spectra, a spike aware optimizer that suppresses the dominant low rank spike subspace without amplifying the noise sensitive spectral tail. Spectra tracks the spike subspace via cached, warm started power iteration and applies low rank spectral shaping with negligible overhead and substantially reduced optimizer state memory. On LLaMA3 8B trained on 50B tokens, Spectra reaches the same target loss 30% faster than AdamW, reduces per step end to end overhead by 0.7%, cuts optimizer state memory by 49.25%, and improves average downstream accuracy by 1.62%. Compared to Muon, Spectra is 5.1x faster in optimizer processing time, achieves a lower final loss, and improves average accuracy by 0.66%. △ Less

Submitted 30 January, 2026; originally announced February 2026.

On the Robustness of Knowledge Editing for Detoxification

Authors: Ming Dong, Shiyi Tang, Ziyan Peng, Guanyi Chen, Tingting He

Abstract: Knowledge-Editing-based (KE-based) detoxification has emerged as a promising approach for mitigating harmful behaviours in Large Language Models. Existing evaluations, however, largely rely on automatic toxicity classifiers, implicitly assuming that reduced toxicity scores reflect genuine behavioural suppression. In this work, we propose a robustness-oriented evaluation framework for KE-based deto… ▽ More Knowledge-Editing-based (KE-based) detoxification has emerged as a promising approach for mitigating harmful behaviours in Large Language Models. Existing evaluations, however, largely rely on automatic toxicity classifiers, implicitly assuming that reduced toxicity scores reflect genuine behavioural suppression. In this work, we propose a robustness-oriented evaluation framework for KE-based detoxification that examines its reliability beyond standard classifier-based metrics along three dimensions: optimisation robustness, compositional robustness, and cross-lingual robustness. We identify pseudo-detoxification as a common failure mode, where apparent toxicity reductions arise from degenerate generation behaviours rather than meaningful suppression of unsafe content. We further show that detoxification effectiveness degrades when multiple unsafe behaviours are edited jointly, and that both monolingual and cross-lingual detoxification remain effective only under specific model-method combinations. Overall, our results indicate that KE-based detoxification is robust only for certain models, limited numbers of detoxification objectives, and a subset of languages. △ Less

Submitted 10 February, 2026; originally announced February 2026.

HoGS: Homophily-Oriented Graph Synthesis for Local Differentially Private GNN Training

Authors: Wen Xu, Zhetao Li, Yong Xiao, Pengpeng Qiao, Mianxiong Dong, Kaoru Ota

Abstract: Graph neural networks (GNNs) have demonstrated remarkable performance in various graph-based machine learning tasks by effectively modeling high-order interactions between nodes. However, training GNNs without protection may leak sensitive personal information in graph data, including links and node features. Local differential privacy (LDP) is an advanced technique for protecting data privacy in… ▽ More Graph neural networks (GNNs) have demonstrated remarkable performance in various graph-based machine learning tasks by effectively modeling high-order interactions between nodes. However, training GNNs without protection may leak sensitive personal information in graph data, including links and node features. Local differential privacy (LDP) is an advanced technique for protecting data privacy in decentralized networks. Unfortunately, existing local differentially private GNNs either only preserve link privacy or suffer significant utility loss in the process of preserving link and node feature privacy. In this paper, we propose an effective LDP framework, called HoGS, which trains GNNs with link and feature protection by generating a synthetic graph. Concretely, HoGS first collects the link and feature information of the graph under LDP, and then utilizes the phenomenon of homophily in graph data to reconstruct the graph structure and node features separately, thereby effectively mitigating the negative impact of LDP on the downstream GNN training. We theoretically analyze the privacy guarantee of HoGS and conduct experiments using the generated synthetic graph as input to various state-of-the-art GNN architectures. Experimental results on three real-world datasets show that HoGS significantly outperforms baseline methods in the accuracy of training GNNs. △ Less

Submitted 9 February, 2026; originally announced February 2026.

Joint single-shot ToA and DoA estimation for VAA-based BLE ranging with phase ambiguity: A deep learning-based approach

Authors: Jincheng Xie, Yili Deng, Jiguang He, Pengyu Wang, Miaomiao Dong, Rui Tang, Zhongyi Huang

Abstract: Conventional direction-of-arrival (DoA) estimation methods rely on multi-antenna arrays, which are costly to implement on size-constrained Bluetooth Low Energy (BLE) devices. Virtual antenna array (VAA) techniques enable DoA estimation with a single antenna, making angle estimation feasible on such devices. However, BLE only provides a single-shot two-way channel frequency response (CFR) with a bi… ▽ More Conventional direction-of-arrival (DoA) estimation methods rely on multi-antenna arrays, which are costly to implement on size-constrained Bluetooth Low Energy (BLE) devices. Virtual antenna array (VAA) techniques enable DoA estimation with a single antenna, making angle estimation feasible on such devices. However, BLE only provides a single-shot two-way channel frequency response (CFR) with a binary phase ambiguity issue, which hinders the direct application of VAA. To address this challenge, we propose a unified model that combines VAA with BLE two-way CFR, and introduce a neural network based phase recovery framework that employs row / column predictors with a voting mechanism to resolve the ambiguity. The recovered one-way CFR then enables super resolution algorithms such as MUSIC for joint time of arrival (ToA) and DoA estimation. Simulation results demonstrate that the proposed method achieves superior performance under non-uniform VAAs, with mean square errors approaching the Cramer Rao bound at SNR $\geq$ 5 dB. △ Less

Submitted 21 January, 2026; originally announced February 2026.

Kimi K2.5: Visual Agentic Intelligence

Authors: Kimi Team, Tongtong Bai, Yifan Bai, Yiping Bao, S. H. Cai, Yuan Cao, Ziwei Chai, Y. Charles, H. S. Che, Cheng Chen, Guanduo Chen, Huarong Chen, Jia Chen, Jianlong Chen, Jun Chen, Kefan Chen, Liang Chen, Ruijue Chen, Xinhao Chen, Yanru Chen, Yanxu Chen, Yicun Chen, Yimin Chen, Yingjiang Chen, Yuankun Chen , et al. (312 additional authors not shown)

Abstract: We introduce Kimi K2.5, an open-source multimodal agentic model designed to advance general agentic intelligence. K2.5 emphasizes the joint optimization of text and vision so that two modalities enhance each other. This includes a series of techniques such as joint text-vision pre-training, zero-vision SFT, and joint text-vision reinforcement learning. Building on this multimodal foundation, K2.5… ▽ More We introduce Kimi K2.5, an open-source multimodal agentic model designed to advance general agentic intelligence. K2.5 emphasizes the joint optimization of text and vision so that two modalities enhance each other. This includes a series of techniques such as joint text-vision pre-training, zero-vision SFT, and joint text-vision reinforcement learning. Building on this multimodal foundation, K2.5 introduces Agent Swarm, a self-directed parallel agent orchestration framework that dynamically decomposes complex tasks into heterogeneous sub-problems and executes them concurrently. Extensive evaluations show that Kimi K2.5 achieves state-of-the-art results across various domains including coding, vision, reasoning, and agentic tasks. Agent Swarm also reduces latency by up to $4.5\times$ over single-agent baselines. We release the post-trained Kimi K2.5 model checkpoint to facilitate future research and real-world applications of agentic intelligence. △ Less

Submitted 7 August, 2026; v1 submitted 2 February, 2026; originally announced February 2026.

Comments: Kimi K2.5 tech report

Dispelling the Curse of Singularities in Neural Network Optimizations

Authors: Hengjie Cao, Mengyi Chen, Yifeng Yang, Fang Dong, Ruijun Huang, Anrui Chen, Jixian Zhou, Mingzhi Dong, Yujiang Wang, Dongsheng Li, Wenyi Fang, Yuanyi Lin, Fan Wu, Li Shang

Abstract: This work investigates the optimization instability of deep neural networks from a less-explored yet insightful perspective: the emergence and amplification of singularities in the parametric space. Our analysis reveals that parametric singularities inevitably grow with gradient updates and further intensify alignment with representations, leading to increased singularities in the representation s… ▽ More This work investigates the optimization instability of deep neural networks from a less-explored yet insightful perspective: the emergence and amplification of singularities in the parametric space. Our analysis reveals that parametric singularities inevitably grow with gradient updates and further intensify alignment with representations, leading to increased singularities in the representation space. We show that the gradient Frobenius norms are bounded by the top singular values of the weight matrices, and as training progresses, the mutually reinforcing growth of weight and representation singularities, termed the curse of singularities, relaxes these bounds, escalating the risk of sharp loss explosions. To counter this, we propose Parametric Singularity Smoothing (PSS), a lightweight, flexible, and effective method for smoothing the singular spectra of weight matrices. Extensive experiments across diverse datasets, architectures, and optimizers demonstrate that PSS mitigates instability, restores trainability even after failure, and improves both training efficiency and generalization. △ Less

Submitted 12 February, 2026; v1 submitted 1 February, 2026; originally announced February 2026.

From Horizontal Layering to Vertical Integration: A Comparative Study of the AI-Driven Software Development Paradigm

Authors: Chi Zhang, Zehan Li, Ziqian Zhong, Haibing Ma, Dan Xiao, Chen Lin, Ming Dong

Abstract: This paper examines the organizational implications of Generative AI adoption in software engineering through a multiple-case comparative study. We contrast two development environments: a traditional enterprise (brownfield) and an AI-native startup (greenfield). Our analysis reveals that transitioning from Horizontal Layering (functional specialization) to Vertical Integration (end-to-end ownersh… ▽ More This paper examines the organizational implications of Generative AI adoption in software engineering through a multiple-case comparative study. We contrast two development environments: a traditional enterprise (brownfield) and an AI-native startup (greenfield). Our analysis reveals that transitioning from Horizontal Layering (functional specialization) to Vertical Integration (end-to-end ownership) yields 8-fold to 33-fold reductions in resource consumption. We attribute these gains to the emergence of Super Employees, AI-augmented engineers who span traditional role boundaries, and the elimination of inter-functional coordination overhead. Theoretically, we propose Human-AI Collaboration Efficacy as the primary optimization target for engineering organizations, supplanting individual productivity metrics. Our Total Factor Productivity analysis identifies an AI Distortion Effect that diminishes returns to labor scale while amplifying technological leverage. We conclude with managerial strategies for organizational redesign, including the reactivation of idle cognitive bandwidth in senior engineers and the suppression of blind scale expansion. △ Less

Submitted 30 January, 2026; originally announced January 2026.

Non-Markov Multi-Round Conversational Image Generation with History-Conditioned MLLMs

Authors: Haochen Zhang, Animesh Sinha, Felix Juefei-Xu, Haoyu Ma, Kunpeng Li, Zhipeng Fan, Meng Dong, Xiaoliang Dai, Tingbo Hou, Peizhao Zhang, Zecheng He

Abstract: Conversational image generation requires a model to follow user instructions across multiple rounds of interaction, grounded in interleaved text and images that accumulate as chat history. While recent multimodal large language models (MLLMs) can generate and edit images, most existing multi-turn benchmarks and training recipes are effectively Markov: the next output depends primarily on the most… ▽ More Conversational image generation requires a model to follow user instructions across multiple rounds of interaction, grounded in interleaved text and images that accumulate as chat history. While recent multimodal large language models (MLLMs) can generate and edit images, most existing multi-turn benchmarks and training recipes are effectively Markov: the next output depends primarily on the most recent image, enabling shortcut solutions that ignore long-range history. In this work we formalize and target the more challenging non-Markov setting, where a user may refer back to earlier states, undo changes, or reference entities introduced several rounds ago. We present (i) non-Markov multi-round data construction strategies, including rollback-style editing that forces retrieval of earlier visual states and name-based multi-round personalization that binds names to appearances across rounds; (ii) a history-conditioned training and inference framework with token-level caching to prevent multi-round identity drift; and (iii) enabling improvements for high-fidelity image reconstruction and editable personalization, including a reconstruction-based DiT detokenizer and a multi-stage fine-tuning curriculum. We demonstrate that explicitly training for non-Markov interactions yields substantial improvements in multi-round consistency and instruction compliance, while maintaining strong single-round editing and personalization. △ Less

Submitted 28 January, 2026; originally announced January 2026.

Comments: 19 pages, 19 figures, plan for TIP

Revisiting 3D Reconstruction Kernels as Low-Pass Filters

Authors: Shengjun Zhang, Min Chen, Yibo Wei, Mingyu Dong, Yueqi Duan

Abstract: 3D reconstruction is to recover 3D signals from the sampled discrete 2D pixels, with the goal to converge continuous 3D spaces. In this paper, we revisit 3D reconstruction from the perspective of signal processing, identifying the periodic spectral extension induced by discrete sampling as the fundamental challenge. Previous 3D reconstruction kernels, such as Gaussians, Exponential functions, and… ▽ More 3D reconstruction is to recover 3D signals from the sampled discrete 2D pixels, with the goal to converge continuous 3D spaces. In this paper, we revisit 3D reconstruction from the perspective of signal processing, identifying the periodic spectral extension induced by discrete sampling as the fundamental challenge. Previous 3D reconstruction kernels, such as Gaussians, Exponential functions, and Student's t distributions, serve as the low pass filters to isolate the baseband spectrum. However, their unideal low-pass property results in the overlap of high-frequency components with low-frequency components in the discrete-time signal's spectrum. To this end, we introduce Jinc kernel with an instantaneous drop to zero magnitude exactly at the cutoff frequency, which is corresponding to the ideal low pass filters. As Jinc kernel suffers from low decay speed in the spatial domain, we further propose modulated kernels to strick an effective balance, and achieves superior rendering performance by reconciling spatial efficiency and frequency-domain fidelity. Experimental results have demonstrated the effectiveness of our Jinc and modulated kernels. △ Less

Submitted 25 January, 2026; originally announced January 2026.

Comments: 14 pages, 5 figures

"Range as a Key" is the Key! Fast and Compact Cloud Block Store Index with RASK

Authors: Haoru Zhao, Mingkai Dong, Erci Xu, Zhongyu Wang, Haibo Chen

Abstract: In cloud block store, indexing is on the critical path of I/O operations and typically resides in memory. With the scaling of users and the emergence of denser storage media, the index has become a primary memory consumer, causing memory strain. Our extensive analysis of production traces reveals that write requests exhibit a strong tendency to target continuous block ranges in cloud storage syste… ▽ More In cloud block store, indexing is on the critical path of I/O operations and typically resides in memory. With the scaling of users and the emergence of denser storage media, the index has become a primary memory consumer, causing memory strain. Our extensive analysis of production traces reveals that write requests exhibit a strong tendency to target continuous block ranges in cloud storage systems. Thus, compared to current per-block indexing, our insight is that we should directly index block ranges (i.e., range-as-a-key) to save memory. In this paper, we propose RASK, a memory-efficient and high-performance tree-structured index that natively indexes ranges. While range-as-a-key offers the potential to save memory and improve performance, realizing this idea is challenging due to the range overlap and range fragmentation issues. To handle range overlap efficiently, RASK introduces the log-structured leaf, combined with range-tailored search and garbage collection. To reduce range fragmentation, RASK employs range-aware split and merge mechanisms. Our evaluations on four production traces show that RASK reduces memory footprint by up to 98.9% and increases throughput by up to 31.0x compared to ten state-of-the-art indexes. △ Less

Submitted 20 January, 2026; originally announced January 2026.

S2DiT: Sandwich Diffusion Transformer for Mobile Streaming Video Generation

Authors: Lin Zhao, Yushu Wu, Aleksei Lebedev, Dishani Lahiri, Meng Dong, Arpit Sahni, Michael Vasilkovsky, Hao Chen, Ju Hu, Aliaksandr Siarohin, Sergey Tulyakov, Yanzhi Wang, Anil Kag, Yanyu Li

Abstract: Diffusion Transformers (DiTs) have recently improved video generation quality. However, their heavy computational cost makes real-time or on-device generation infeasible. In this work, we introduce S2DiT, a Streaming Sandwich Diffusion Transformer designed for efficient, high-fidelity, and streaming video generation on mobile hardware. S2DiT generates more tokens but maintains efficiency with nove… ▽ More Diffusion Transformers (DiTs) have recently improved video generation quality. However, their heavy computational cost makes real-time or on-device generation infeasible. In this work, we introduce S2DiT, a Streaming Sandwich Diffusion Transformer designed for efficient, high-fidelity, and streaming video generation on mobile hardware. S2DiT generates more tokens but maintains efficiency with novel efficient attentions: a mixture of LinConv Hybrid Attention (LCHA) and Stride Self-Attention (SSA). Based on this, we uncover the sandwich design via a budget-aware dynamic programming search, achieving superior quality and efficiency. We further propose a 2-in-1 distillation framework that transfers the capacity of large teacher models (e.g., Wan 2.2-14B) to the compact few-step sandwich model. Together, S2DiT achieves quality on par with state-of-the-art server video models, while streaming at over 10 FPS on an iPhone. △ Less

Submitted 6 March, 2026; v1 submitted 18 January, 2026; originally announced January 2026.

Comments: https://snap-research.github.io/S2DiT/

OpenAI GPT-5 System Card

Authors: Aaditya Singh, Adam Fry, Adam Perelman, Adam Tart, Adi Ganesh, Ahmed El-Kishky, Aidan McLaughlin, Aiden Low, AJ Ostrow, Akhila Ananthram, Akshay Nathan, Alan Luo, Alec Helyar, Aleksander Madry, Aleksandr Efremov, Aleksandra Spyra, Alex Baker-Whitcomb, Alex Beutel, Alex Karpenko, Alex Makelov, Alex Neitz, Alex Wei, Alexandra Barr, Alexandre Kirchmeyer, Alexey Ivanov , et al. (461 additional authors not shown)

Abstract: This is the system card published alongside the OpenAI GPT-5 launch, August 2025. GPT-5 is a unified system with a smart and fast model that answers most questions, a deeper reasoning model for harder problems, and a real-time router that quickly decides which model to use based on conversation type, complexity, tool needs, and explicit intent (for example, if you say 'think hard about this' in… ▽ More This is the system card published alongside the OpenAI GPT-5 launch, August 2025. GPT-5 is a unified system with a smart and fast model that answers most questions, a deeper reasoning model for harder problems, and a real-time router that quickly decides which model to use based on conversation type, complexity, tool needs, and explicit intent (for example, if you say 'think hard about this' in the prompt). The router is continuously trained on real signals, including when users switch models, preference rates for responses, and measured correctness, improving over time. Once usage limits are reached, a mini version of each model handles remaining queries. This system card focuses primarily on gpt-5-thinking and gpt-5-main, while evaluations for other models are available in the appendix. The GPT-5 system not only outperforms previous models on benchmarks and answers questions more quickly, but -- more importantly -- is more useful for real-world queries. We've made significant advances in reducing hallucinations, improving instruction following, and minimizing sycophancy, and have leveled up GPT-5's performance in three of ChatGPT's most common uses: writing, coding, and health. All of the GPT-5 models additionally feature safe-completions, our latest approach to safety training to prevent disallowed content. Similarly to ChatGPT agent, we have decided to treat gpt-5-thinking as High capability in the Biological and Chemical domain under our Preparedness Framework, activating the associated safeguards. While we do not have definitive evidence that this model could meaningfully help a novice to create severe biological harm -- our defined threshold for High capability -- we have chosen to take a precautionary approach. △ Less

Submitted 1 May, 2026; v1 submitted 19 December, 2025; originally announced January 2026.

Comments: May 2026: Added monitorability evals and authors

Squeezing Edge Performance: A Sensitivity-Aware Container Management for Heterogeneous Tasks

Authors: Yongmin Zhang, Pengyu Huang, Mingyi Dong, Jing Yao

Abstract: Edge computing enables latency-critical applications to process data close to end devices, yet task heterogeneity and limited resources pose significant challenges to efficient orchestration. This paper presents a measurement-driven, container-based resource management framework for intra-node optimization on a single edge server hosting multiple heterogeneous applications. Extensive profiling exp… ▽ More Edge computing enables latency-critical applications to process data close to end devices, yet task heterogeneity and limited resources pose significant challenges to efficient orchestration. This paper presents a measurement-driven, container-based resource management framework for intra-node optimization on a single edge server hosting multiple heterogeneous applications. Extensive profiling experiments are conducted to derive a nonlinear fitting model that characterizes the relationship among CPU/memory allocations and processing latency across diverse workloads, enabling reliable estimation of performance under varying configurations and providing quantitative support for subsequent optimization. Using this model and a queueing-based delay formulation, we formulate a mixed-integer nonlinear programming (MINLP) problem to jointly minimize system latency and power consumption, which is shown to be NP-hard. The problem is decomposed into tractable convex subproblems and solved through a two-stage container-based resource management scheme (CRMS) combining convex optimization and greedy refinement. The proposed scheme achieves polynomial-time complexity and supports quasi-dynamic execution under global resource constraints. Simulation results demonstrate that CRMS reduces latency by over 14\% and improves energy efficiency compared with heuristic and search-based baselines, offering a practical and scalable solution for heterogeneous edge environments with dynamic workload characteristics. △ Less

Submitted 29 December, 2025; originally announced December 2025.

Deterministic Image-to-Image Translation via Denoising Brownian Bridge Models with Dual Approximators

Authors: Bohan Xiao, Peiyong Wang, Qisheng He, Ming Dong

Abstract: Image-to-Image (I2I) translation involves converting an image from one domain to another. Deterministic I2I translation, such as in image super-resolution, extends this concept by guaranteeing that each input generates a consistent and predictable output, closely matching the ground truth (GT) with high fidelity. In this paper, we propose a denoising Brownian bridge model with dual approximators (… ▽ More Image-to-Image (I2I) translation involves converting an image from one domain to another. Deterministic I2I translation, such as in image super-resolution, extends this concept by guaranteeing that each input generates a consistent and predictable output, closely matching the ground truth (GT) with high fidelity. In this paper, we propose a denoising Brownian bridge model with dual approximators (Dual-approx Bridge), a novel generative model that exploits the Brownian bridge dynamics and two neural network-based approximators (one for forward and one for reverse process) to produce faithful output with negligible variance and high image quality in I2I translations. Our extensive experiments on benchmark datasets including image generation and super-resolution demonstrate the consistent and superior performance of Dual-approx Bridge in terms of image quality and faithfulness to GT when compared to both stochastic and deterministic baselines. Project page and code: https://github.com/bohan95/dual-app-bridge △ Less

Submitted 29 December, 2025; originally announced December 2025.

Comments: Minor correction to a reference entry

SwinCCIR: An end-to-end deep network for Compton camera imaging reconstruction

Authors: Minghao Dong, Xinyang Luo, Xujian Ouyang, Yongshun Xiao

Abstract: Compton cameras (CCs) are a kind of gamma cameras which are designed to determine the directions of incident gammas based on the Compton scatter. However, the reconstruction of CCs face problems of severe artifacts and deformation due to the fundamental reconstruction principle of back-projection of Compton cones. Besides, a part of systematic errors originated from the performance of devices are… ▽ More Compton cameras (CCs) are a kind of gamma cameras which are designed to determine the directions of incident gammas based on the Compton scatter. However, the reconstruction of CCs face problems of severe artifacts and deformation due to the fundamental reconstruction principle of back-projection of Compton cones. Besides, a part of systematic errors originated from the performance of devices are hard to remove through calibration, leading to deterioration of imaging quality. Iterative algorithms and deep-learning based methods have been widely used to improve reconstruction. But most of them are optimization based on the results of back-projection. Therefore, we proposed an end-to-end deep learning framework, SwinCCIR, for CC imaging. Through adopting swin-transformer blocks and a transposed convolution-based image generation module, we established the relationship between the list-mode events and the radioactive source distribution. SwinCCIR was trained and validated on both simulated and practical dataset. The experimental results indicate that SwinCCIR effectively overcomes problems of conventional CC imaging, which are expected to be implemented in practical applications. △ Less

Submitted 27 December, 2025; originally announced December 2025.

Comments: 10 pages, 7 figures

RELIC-GNN: Efficient State Registers Identification with Graph Neural Network for Reverse Engineering

Authors: Weitao Pan, Meng Dong, Zhiliang Qiu, Jianlei Yang, Zhixiong Di, Yiming Gao

Abstract: Reverse engineering of gate-level netlist is critical for Hardware Trojans detection and Design Piracy counteracting. The primary task of gate-level reverse engineering is to separate the control and data signals from the netlist, which is mainly realized by identifying state registers with topological comparison.However, these methods become inefficient for large scale netlist. In this work, we p… ▽ More Reverse engineering of gate-level netlist is critical for Hardware Trojans detection and Design Piracy counteracting. The primary task of gate-level reverse engineering is to separate the control and data signals from the netlist, which is mainly realized by identifying state registers with topological comparison.However, these methods become inefficient for large scale netlist. In this work, we propose RELIC-GNN, a graph neural network based state registers identification method, to address these issues. RELIC-GNN models the path structure of register as a graph and generates corresponding representation by considering node attributes and graph structure during training. The trained GNN model could be adopted to find the registers type very efficiently. Experimental results show that RELIC-GNN could achieve 100% in recall, 30.49% in precision and 88.37% in accuracy on average across different designs, which obtains significant improvements than previous approaches. △ Less

Submitted 16 December, 2025; originally announced December 2025.

Truthful and Trustworthy IoT AI Agents via Immediate-Penalty Enforcement under Approximate VCG Mechanisms

Authors: Xun Shao, Ryuuto Shimizu, Zhi Liu, Kaoru Ota, Mianxiong Dong

Abstract: The deployment of autonomous AI agents in Internet of Things (IoT) energy systems requires decision-making mechanisms that remain robust, efficient, and trustworthy under real-time constraints and imperfect monitoring. While reinforcement learning enables adaptive prosumer behaviors, ensuring economic consistency and preventing strategic manipulation remain open challenges, particularly when sensi… ▽ More The deployment of autonomous AI agents in Internet of Things (IoT) energy systems requires decision-making mechanisms that remain robust, efficient, and trustworthy under real-time constraints and imperfect monitoring. While reinforcement learning enables adaptive prosumer behaviors, ensuring economic consistency and preventing strategic manipulation remain open challenges, particularly when sensing noise or partial observability reduces the operator's ability to verify actions. This paper introduces a trust-enforcement framework for IoT energy trading that combines an approximate Vickrey-Clarke-Groves (VCG) double auction with an immediate one-shot penalty. Unlike reputation- or history-based approaches, the proposed mechanism restores truthful reporting within a single round, even when allocation accuracy is approximate and monitoring is noisy. We theoretically characterize the incentive gap induced by approximation and derive a penalty threshold that guarantees truthful bidding under bounded sensing errors. To evaluate learning-enabled prosumers, we embed the mechanism into a multi-agent reinforcement learning environment reflecting stochastic generation, dynamic loads, and heterogeneous trading opportunities. Experiments show that improved allocation accuracy reduces deviation incentives, the required penalty matches analytical predictions, and learned bidding behaviors remain stable and interpretable despite imperfect monitoring. These results demonstrate that lightweight penalty designs can reliably align strategic IoT agents with socially efficient energy-trading outcomes. △ Less

Submitted 1 December, 2025; v1 submitted 29 November, 2025; originally announced December 2025.

Journal ref: IEEE Internet of Things Journal, 2026