Showing 1–50 of 205 results for author: Cha, S

Symetra: Visual Analytics for the Parameter Tuning Process of Symbolic Execution Engines

Authors: Donghee Hong, Minjong Kim, Sooyoung Cha, Jaemin Jo

Abstract: Symbolic execution engines such as KLEE automatically generate test cases to maximize branch coverage, but their numerous parameters make it difficult to understand the parameters' impact, leading the user to rely on suboptimal default configurations. While automated tuners have shown promising results, they provide limited insights into why certain configurations work well, motivating the need fo… ▽ More Symbolic execution engines such as KLEE automatically generate test cases to maximize branch coverage, but their numerous parameters make it difficult to understand the parameters' impact, leading the user to rely on suboptimal default configurations. While automated tuners have shown promising results, they provide limited insights into why certain configurations work well, motivating the need for Human-in-the-Loop approaches. In this work, we present a visual analytics system, Symetra, designed to support Human-in-the-Loop parameter tuning of symbolic execution engines. To handle a large number of parameters and their configurations, we provide two complementary overviews of their impact on branch coverage values and patterns. Building on these overviews, our system enables collective analysis, allowing the user to contrast groups of configurations and identify differences that may affect branch coverage. We also report on case studies and a Human-in-the-Loop tuning process, demonstrating that experts not only interpreted parameter impacts and identified complementary configurations, but also improved upon fully automated approaches in both branch coverage and tuning efficiency. △ Less

Submitted 6 April, 2026; originally announced April 2026.

GPU-Accelerated Optimization of Transformer-Based Neural Networks for Real-Time Inference

Authors: Soutrik Mukherjee, Sangwhan Cha

Abstract: This paper presents the design and evaluation of a GPU-accelerated inference pipeline for transformer models using NVIDIA TensorRT with mixed-precision optimization. We evaluate BERT-base (110M parameters) and GPT-2 (124M parameters) across batch sizes from 1 to 32 and sequence lengths from 32 to 512. The system achieves up to 64.4x speedup over CPU baselines, sub-10 ms latency for single-sample i… ▽ More This paper presents the design and evaluation of a GPU-accelerated inference pipeline for transformer models using NVIDIA TensorRT with mixed-precision optimization. We evaluate BERT-base (110M parameters) and GPT-2 (124M parameters) across batch sizes from 1 to 32 and sequence lengths from 32 to 512. The system achieves up to 64.4x speedup over CPU baselines, sub-10 ms latency for single-sample inference, and a 63 percent reduction in memory usage. We introduce a hybrid precision strategy that preserves FP32 for numerically sensitive operations such as softmax and layer normalization, while applying FP16 to linear layers. This approach maintains high numerical fidelity (cosine similarity >= 0.9998 relative to baseline outputs) and eliminates NaN instability. The pipeline is implemented as a modular, containerized system that enables reproducible benchmarking across more than 360 configurations. Cross-GPU validation on an NVIDIA A100 shows consistent FP16 speedup ratios between 1.84x and 2.00x, along with stable numerical behavior. Downstream evaluation on SST-2 demonstrates no accuracy degradation under hybrid precision. Validation on WikiText-2 shows that random inputs underestimate NaN instability by up to 6x for full FP16, while confirming the robustness of the hybrid approach (0.0 percent NaN, cosine similarity >= 0.9998). These results provide a detailed characterization of performance and accuracy trade-offs across GPU architectures and offer practical guidance for deploying transformer models in latency-critical environments. △ Less

Submitted 30 March, 2026; originally announced March 2026.

Comments: 10 pages, 8 figures, 15 tables

Scaling Atomistic Protein Binder Design with Generative Pretraining and Test-Time Compute

Authors: Kieran Didi, Zuobai Zhang, Guoqing Zhou, Danny Reidenbach, Zhonglin Cao, Sooyoung Cha, Tomas Geffner, Christian Dallago, Jian Tang, Michael M. Bronstein, Martin Steinegger, Emine Kucukbenli, Arash Vahdat, Karsten Kreis

Abstract: Protein interaction modeling is central to protein design, which has been transformed by machine learning with applications in drug discovery and beyond. In this landscape, structure-based de novo binder design is cast as either conditional generative modeling or sequence optimization via structure predictors ("hallucination"). We argue that this is a false dichotomy and propose Proteina-Complexa,… ▽ More Protein interaction modeling is central to protein design, which has been transformed by machine learning with applications in drug discovery and beyond. In this landscape, structure-based de novo binder design is cast as either conditional generative modeling or sequence optimization via structure predictors ("hallucination"). We argue that this is a false dichotomy and propose Proteina-Complexa, a novel fully atomistic binder generation method unifying both paradigms. We extend recent flow-based latent protein generation architectures and leverage the domain-domain interactions of monomeric computationally predicted protein structures to construct Teddymer, a new large-scale dataset of synthetic binder-target pairs for pretraining. Combined with high-quality experimental multimers, this enables training a strong base model. We then perform inference-time optimization with this generative prior, unifying the strengths of previously distinct generative and hallucination methods. Proteina-Complexa sets a new state of the art in computational binder design benchmarks: it delivers markedly higher in-silico success rates than existing generative approaches, and our novel test-time optimization strategies greatly outperform previous hallucination methods under normalized compute budgets. We also demonstrate interface hydrogen bond optimization, fold class-guided binder generation, and extensions to small molecule targets and enzyme design tasks, again surpassing prior methods. Code, models and new data will be publicly released. △ Less

Submitted 29 March, 2026; originally announced March 2026.

Comments: ICLR 2026 Oral Presentation. Project page: https://research.nvidia.com/labs/genair/proteina-complexa/

Voxtral TTS

Authors: Mistral-AI, :, Alexander H. Liu, Alexis Tacnet, Andy Ehrenberg, Andy Lo, Chen-Yo Sun, Guillaume Lample, Henry Lagarde, Jean-Malo Delignon, Jaeyoung Kim, John Harvill, Khyathi Raghavi Chandu, Lorenzo Signoretti, Margaret Jennings, Patrick von Platen, Pavankumar Reddy Muddireddy, Rohin Arora, Sanchit Gandhi, Samuel Humeau, Soham Ghosh, Srijan Mishra, Van Phung, Abdelaziz Bounhar, Abhinav Rastogi , et al. (164 additional authors not shown)

Abstract: We introduce Voxtral TTS, an expressive multilingual text-to-speech model that generates natural speech from as little as 3 seconds of reference audio. Voxtral TTS adopts a hybrid architecture that combines auto-regressive generation of semantic speech tokens with flow-matching for acoustic tokens. These tokens are encoded and decoded with Voxtral Codec, a speech tokenizer trained from scratch wit… ▽ More We introduce Voxtral TTS, an expressive multilingual text-to-speech model that generates natural speech from as little as 3 seconds of reference audio. Voxtral TTS adopts a hybrid architecture that combines auto-regressive generation of semantic speech tokens with flow-matching for acoustic tokens. These tokens are encoded and decoded with Voxtral Codec, a speech tokenizer trained from scratch with a hybrid VQ-FSQ quantization scheme. In human evaluations conducted by native speakers, Voxtral TTS is preferred for multilingual voice cloning due to its naturalness and expressivity, achieving a 68.4\% win rate over ElevenLabs Flash v2.5. We release the model weights under a CC BY-NC license. △ Less

Submitted 6 April, 2026; v1 submitted 26 March, 2026; originally announced March 2026.

Can ChatGPT Really Understand Modern Chinese Poetry?

Authors: Shanshan Wang, Derek F. Wong, Jingming Yao, Lidia S. Chao

Abstract: ChatGPT has demonstrated remarkable capabilities on both poetry generation and translation, yet its ability to truly understand poetry remains unexplored. Previous poetry-related work merely analyzed experimental outcomes without addressing fundamental issues of comprehension. This paper introduces a comprehensive framework for evaluating ChatGPT's understanding of modern poetry. We collaborated w… ▽ More ChatGPT has demonstrated remarkable capabilities on both poetry generation and translation, yet its ability to truly understand poetry remains unexplored. Previous poetry-related work merely analyzed experimental outcomes without addressing fundamental issues of comprehension. This paper introduces a comprehensive framework for evaluating ChatGPT's understanding of modern poetry. We collaborated with professional poets to evaluate ChatGPT's interpretation of modern Chinese poems by different poets along multiple dimensions. Evaluation results show that ChatGPT's interpretations align with the original poets' intents in over 73% of the cases. However, its understanding in certain dimensions, particularly in capturing poeticity, proved to be less satisfactory. These findings highlight the effectiveness and necessity of our proposed framework. This study not only evaluates ChatGPT's ability to understand modern poetry but also establishes a solid foundation for future research on LLMs and their application to poetry-related tasks. △ Less

Submitted 21 March, 2026; originally announced March 2026.

Comments: Accepted by EACL 2026

Adaptive Auxiliary Prompt Blending for Target-Faithful Diffusion Generation

Authors: Kwanyoung Lee, SeungJu Cha, Yebin Ahn, Hyunwoo Oh, Sungho Koh, Dong-Jin Kim

Abstract: Diffusion-based text-to-image (T2I) models have made remarkable progress in generating photorealistic and semantically rich images. However, when the target concepts lie in low-density regions of the training distribution, these models often produce semantically misaligned or structurally inconsistent results. This limitation arises from the long-tailed nature of text-image datasets, where rare co… ▽ More Diffusion-based text-to-image (T2I) models have made remarkable progress in generating photorealistic and semantically rich images. However, when the target concepts lie in low-density regions of the training distribution, these models often produce semantically misaligned or structurally inconsistent results. This limitation arises from the long-tailed nature of text-image datasets, where rare concepts or editing instructions are underrepresented. To address this, we introduce Adaptive Auxiliary Prompt Blending (AAPB) - a unified framework that stabilizes the diffusion process in low-density regions. AAPB leverages auxiliary anchor prompts to provide semantic support in rare concept generation and structural support in image editing, ensuring faithful guidance toward the target prompt. Unlike prior heuristic prompt alternation methods, AAPB derives a closed-form adaptive coefficient that optimally balances the influence between the auxiliary anchor and the target prompt at each diffusion step. Grounded in Tweedie's identity, our formulation provides a principled and training-free framework for adaptive prompt blending, ensuring stable and target-faithful generation. We demonstrate the effectiveness of adaptive interpolation over fixed interpolation through controlled experiments and empirically show consistent improvements on the RareBench and FlowEdit datasets, achieving superior semantic accuracy and structural fidelity compared to prior training-free baselines. △ Less

Submitted 19 March, 2026; originally announced March 2026.

Comments: Accepted in CVPR 2026 (main track). 10 pages, 6 figures; supplementary material included (14 pages, 11 figures)

ADAPT: Attention Driven Adaptive Prompt Scheduling and InTerpolating Orthogonal Complements for Rare Concepts Generation

Authors: Kwanyoung Lee, Hyunwoo Oh, SeungJu Cha, Sungho Koh, Dong-Jin Kim

Abstract: Generating rare compositional concepts in text-to-image synthesis remains a challenge for diffusion models, particularly for attributes that are uncommon in the training data. While recent approaches, such as R2F, address this challenge by utilizing LLM for prompt scheduling, they suffer from inherent variance due to the randomness of language models and suboptimal guidance from iterative text emb… ▽ More Generating rare compositional concepts in text-to-image synthesis remains a challenge for diffusion models, particularly for attributes that are uncommon in the training data. While recent approaches, such as R2F, address this challenge by utilizing LLM for prompt scheduling, they suffer from inherent variance due to the randomness of language models and suboptimal guidance from iterative text embedding switching. To address these problems, we propose the ADAPT framework, a training-free framework that deterministically plans and semantically aligns prompt schedules, providing consistent guidance to enhance the composition of rare concepts. By leveraging attention scores and orthogonal components, ADAPT significantly enhances compositional generation of rare concepts in the RareBench benchmark without additional training or fine-tuning. Through comprehensive experiments, we demonstrate that ADAPT achieves superior performance in RareBench and accurately reflects the semantic information of rare attributes, providing deterministic and precise control over the generation of rare compositions without compromising visual integrity. △ Less

Submitted 19 March, 2026; originally announced March 2026.

Comments: Accepted in CVPR 2026 (findings). 10 pages, 4 figures; supplementary material included (8 pages, 10 figures)

Neuron-Aware Data Selection In Instruction Tuning For Large Language Models

Authors: Xin Chen, Junchao Wu, Shu Yang, Runzhe Zhan, Zeyu Wu, Min Yang, Shujian Huang, Lidia S. Chao, Derek F. Wong

Abstract: Instruction Tuning (IT) has been proven to be an effective approach to unlock the powerful capabilities of large language models (LLMs). Recent studies indicate that excessive IT data can degrade LLMs performance, while carefully selecting a small subset of high-quality IT data can significantly enhance their capabilities. Therefore, identifying the most efficient subset data from the IT dataset t… ▽ More Instruction Tuning (IT) has been proven to be an effective approach to unlock the powerful capabilities of large language models (LLMs). Recent studies indicate that excessive IT data can degrade LLMs performance, while carefully selecting a small subset of high-quality IT data can significantly enhance their capabilities. Therefore, identifying the most efficient subset data from the IT dataset to effectively develop either specific or general abilities in LLMs has become a critical challenge. To address this, we propose a novel and efficient framework called NAIT. NAIT evaluates the impact of IT data on LLMs performance by analyzing the similarity of neuron activation patterns between the IT dataset and the target domain capability. Specifically, NAIT captures neuron activation patterns from in-domain datasets of target domain capabilities to construct reusable and transferable neuron activation features. It then evaluates and selects optimal samples based on the similarity between candidate samples and the expected activation features of the target capabilities. Experimental results show that training on the 10\% Alpaca-GPT4 IT data subset selected by NAIT consistently outperforms methods that rely on external advanced models or uncertainty-based features across various tasks. Our findings also reveal the transferability of neuron activation features across different capabilities of LLMs. In particular, IT data with more logical reasoning and programmatic features possesses strong general transferability, enabling models to develop stronger capabilities across multiple tasks, while a stable core subset of data is sufficient to consistently activate fundamental model capabilities and universally improve performance across diverse tasks. △ Less

Submitted 13 March, 2026; originally announced March 2026.

PIM-SHERPA: Software Method for On-device LLM Inference by Resolving PIM Memory Attribute and Layout Inconsistencies

Authors: Sunjung Lee, Sanghoon Cha, Hyeonsu Kim, Seungwoo Seo, Yuhwan Ro, Sukhan Lee, Byeongho Kim, Yongjun Park, Kyomin Sohn, Seungwon Lee, Jaehoon Yu

Abstract: On-device deployments of large language models (LLMs) are rapidly proliferating across mobile and edge platforms. LLM inference comprises a compute-intensive prefill phase and a memory bandwidth-intensive decode phase, and the decode phase has been widely recognized as well-suited to processing-in-memory (PIM) in both academia and industry. However, practical PIM-enabled systems face two obstacles… ▽ More On-device deployments of large language models (LLMs) are rapidly proliferating across mobile and edge platforms. LLM inference comprises a compute-intensive prefill phase and a memory bandwidth-intensive decode phase, and the decode phase has been widely recognized as well-suited to processing-in-memory (PIM) in both academia and industry. However, practical PIM-enabled systems face two obstacles between these phases, a memory attribute inconsistency in which prefill favors placing weights in a cacheable region for reuse whereas decode requires weights in a non-cacheable region to reliably trigger PIM, and a weight layout inconsistency between host-friendly and PIM-aware layouts. To address these problems, we introduce \textit{PIM-SHERPA}, a software-only method for efficient on-device LLM inference by resolving PIM memory attribute and layout inconsistencies. PIM-SHERPA provides two approaches, DRAM double buffering (DDB), which keeps a single PIM-aware weights in the non-cacheable region while prefetching the swizzled weights of the next layer into small cacheable buffers, and online weight rearrangement with swizzled memory copy (OWR), which performs the on-demand swizzled memory copy immediately before GEMM. Compared to a baseline PIM emulation system, PIM-SHERPA achieves approximately 47.8 - 49.7\% memory capacity savings while maintaining comparable performance to the theoretical maximum on the Llama 3.2 model. To the best of our knowledge, this is the first work to identify the memory attribute inconsistency and propose effective solutions on product-level PIM-enabled systems. △ Less

Submitted 10 March, 2026; originally announced March 2026.

Comments: 13 pages, 13 figures

Dynamic Multi-period Experts for Online Time Series Forecasting

Authors: Seungha Hong, Sukang Chae, Suyeon Kim, Sanghwan Jang, Hwanjo Yu

Abstract: Online Time Series Forecasting (OTSF) requires models to continuously adapt to concept drift. However, existing methods often treat concept drift as a monolithic phenomenon. To address this limitation, we first redefine concept drift by categorizing it into two distinct types: Recurring Drift, where previously seen patterns reappear, and Emergent Drift, where entirely new patterns emerge. We then… ▽ More Online Time Series Forecasting (OTSF) requires models to continuously adapt to concept drift. However, existing methods often treat concept drift as a monolithic phenomenon. To address this limitation, we first redefine concept drift by categorizing it into two distinct types: Recurring Drift, where previously seen patterns reappear, and Emergent Drift, where entirely new patterns emerge. We then propose DynaME (Dynamic Multi-period Experts), a novel hybrid framework designed to effectively address this dual nature of drift. For Recurring Drift, DynaME employs a committee of specialized experts that are dynamically fitted to the most relevant historical periodic patterns at each time step. For Emergent Drift, the framework detects high-uncertainty scenarios and shifts reliance to a stable, general expert. Extensive experiments on several benchmark datasets and backbones demonstrate that DynaME effectively adapts to both concept drifts and significantly outperforms existing baselines. △ Less

Submitted 9 March, 2026; originally announced March 2026.

Comments: WWW 2026

X-AVDT: Audio-Visual Cross-Attention for Robust Deepfake Detection

Authors: Youngseo Kim, Kwan Yun, Seokhyeon Hong, Sihun Cha, Colette Suhjung Koo, Junyong Noh

Abstract: The surge of highly realistic synthetic videos produced by contemporary generative systems has significantly increased the risk of malicious use, challenging both humans and existing detectors. Against this backdrop, we take a generator-side view and observe that internal cross-attention mechanisms in these models encode fine-grained speech-motion alignment, offering useful correspondence cues for… ▽ More The surge of highly realistic synthetic videos produced by contemporary generative systems has significantly increased the risk of malicious use, challenging both humans and existing detectors. Against this backdrop, we take a generator-side view and observe that internal cross-attention mechanisms in these models encode fine-grained speech-motion alignment, offering useful correspondence cues for forgery detection. Building on this insight, we propose X-AVDT, a robust and generalizable deepfake detector that probes generator-internal audio-visual signals accessed via DDIM inversion to expose these cues. X-AVDT extracts two complementary signals: (i) a video composite capturing inversion-induced discrepancies, and (ii) an audio-visual cross-attention feature reflecting modality alignment enforced during generation. To enable faithful cross-generator evaluation, we further introduce MMDF, a new multimodal deepfake dataset spanning diverse manipulation types and rapidly evolving synthesis paradigms, including GANs, diffusion, and flow-matching. Extensive experiments demonstrate that X-AVDT achieves leading performance on MMDF and generalizes strongly to external benchmarks and unseen generators, outperforming existing methods with accuracy improved by 13.1%. Our findings highlight the importance of leveraging internal audio-visual consistency cues for robustness to future generators in deepfake detection. △ Less

Submitted 9 March, 2026; originally announced March 2026.

Journal ref: CVPR 2026

Can Large Language Models Keep Up? Benchmarking Online Adaptation to Continual Knowledge Streams

Authors: Jiyeon Kim, Hyunji Lee, Dylan Zhou, Sue Hyun Park, Seunghyun Yoon, Trung Bui, Franck Dernoncourt, Sungmin Cha, Minjoon Seo

Abstract: LLMs operating in dynamic real-world contexts often encounter knowledge that evolves continuously or emerges incrementally. To remain accurate and effective, models must adapt to newly arriving information on the fly. We introduce Online Adaptation to Continual Knowledge Streams(OAKS) to evaluate this capability, establishing a benchmark for online adaptation over streaming, continually updating k… ▽ More LLMs operating in dynamic real-world contexts often encounter knowledge that evolves continuously or emerges incrementally. To remain accurate and effective, models must adapt to newly arriving information on the fly. We introduce Online Adaptation to Continual Knowledge Streams(OAKS) to evaluate this capability, establishing a benchmark for online adaptation over streaming, continually updating knowledge. Specifically, the benchmark is structured as a sequence of fine-grained context chunks where facts change dynamically across time intervals. OAKS comprises two datasets: OAKS-BABI and OAKS-Novel, where individual facts evolve multiple times across context chunks. These datasets include dense annotations to measure whether models track changes accurately. Evaluating 14 models with varied inference approaches, we observe significant limitations in current methodologies. Both state-of-the-art models and agentic memory systems fail to adapt robustly on OAKS, demonstrating delays in state-tracking and susceptibility to distraction within streaming environments. △ Less

Submitted 7 March, 2026; originally announced March 2026.

SAIL: Similarity-Aware Guidance and Inter-Caption Augmentation-based Learning for Weakly-Supervised Dense Video Captioning

Authors: Ye-Chan Kim, SeungJu Cha, Si-Woo Kim, Minju Jeon, Hyungee Kim, Dong-Jin Kim

Abstract: Weakly-Supervised Dense Video Captioning aims to localize and describe events in videos trained only on caption annotations, without temporal boundaries. Prior work introduced an implicit supervision paradigm based on Gaussian masking and complementary captioning. However, existing method focuses merely on generating non-overlapping masks without considering their semantic relationship to correspo… ▽ More Weakly-Supervised Dense Video Captioning aims to localize and describe events in videos trained only on caption annotations, without temporal boundaries. Prior work introduced an implicit supervision paradigm based on Gaussian masking and complementary captioning. However, existing method focuses merely on generating non-overlapping masks without considering their semantic relationship to corresponding events, resulting in simplistic, uniformly distributed masks that fail to capture semantically meaningful regions. Moreover, relying solely on ground-truth captions leads to sub-optimal performance due to the inherent sparsity of existing datasets. In this work, we propose SAIL, which constructs semantically-aware masks through cross-modal alignment. Our similarity aware training objective guides masks to emphasize video regions with high similarity to their corresponding event captions. Furthermore, to guide more accurate mask generation under sparse annotation settings, we introduce an LLM-based augmentation strategy that generates synthetic captions to provide additional alignment signals. These synthetic captions are incorporated through an inter-mask mechanism, providing auxiliary guidance for precise temporal localization without degrading the main objective. Experiments on ActivityNet Captions and YouCook2 demonstrate state-of-the-art performance on both captioning and localization metrics. △ Less

Submitted 9 March, 2026; v1 submitted 5 March, 2026; originally announced March 2026.

Comments: Accepted to CVPR 2026

Online Learning for Multi-Layer Hierarchical Inference under Partial and Policy-Dependent Feedback

Authors: Haoran Zhang, Seohyeon Cha, Hasan Burhan Beytur, Kevin S Chan, Gustavo de Veciana, Haris Vikalo

Abstract: Hierarchical inference systems route tasks across multiple computational layers, where each node may either finalize a prediction locally or offload the task to a node in the next layer for further processing. Learning optimal routing policies in such systems is challenging: inference loss is defined recursively across layers, while feedback on prediction error is revealed only at a terminal oracl… ▽ More Hierarchical inference systems route tasks across multiple computational layers, where each node may either finalize a prediction locally or offload the task to a node in the next layer for further processing. Learning optimal routing policies in such systems is challenging: inference loss is defined recursively across layers, while feedback on prediction error is revealed only at a terminal oracle layer. This induces a partial, policy-dependent feedback structure in which observability probabilities decay with depth, causing importance-weighted estimators to suffer from amplified variance. We study online routing for multi-layer hierarchical inference under long-term resource constraints and terminal-only feedback. We formalize the recursive loss structure and show that naive importance-weighted contextual bandit methods become unstable as feedback probability decays along the hierarchy. To address this, we develop a variance-reduced EXP4-based algorithm integrated with Lyapunov optimization, yielding unbiased loss estimation and stable learning under sparse and policy-dependent feedback. We provide regret guarantees relative to the best fixed routing policy in hindsight and establish near-optimality under stochastic arrivals and resource constraints. Experiments on large-scale multi-task workloads demonstrate improved stability and performance compared to standard importance-weighted approaches. △ Less

Submitted 4 March, 2026; originally announced March 2026.

Comments: preprint

FedRot-LoRA: Mitigating Rotational Misalignment in Federated LoRA

Authors: Haoran Zhang, Dongjun Kim, Seohyeon Cha, Haris Vikalo

Abstract: Federated LoRA provides a communication-efficient mechanism for fine-tuning large language models on decentralized data. In practice, however, a discrepancy between the factor-wise averaging used to preserve low rank and the mathematically correct aggregation of local updates can cause significant aggregation error and unstable training. We argue that a major source of this problem is rotational m… ▽ More Federated LoRA provides a communication-efficient mechanism for fine-tuning large language models on decentralized data. In practice, however, a discrepancy between the factor-wise averaging used to preserve low rank and the mathematically correct aggregation of local updates can cause significant aggregation error and unstable training. We argue that a major source of this problem is rotational misalignment, arising from the rotational invariance of low-rank factorizations -- semantically equivalent updates can be represented in different latent subspaces across clients since $(B_i R_i)(R_i^\top A_i) = B_i A_i$. When such misaligned factors are averaged directly, they interfere destructively and degrade the global update. To address this issue, we propose FedRot-LoRA, a federated LoRA framework that aligns client updates via orthogonal transformations prior to aggregation. This alignment preserves the semantic update while reducing cross-client subspace mismatch, without increasing communication cost or restricting model expressivity. We provide a convergence analysis that examines the aggregation error induced by factor-wise averaging and shows how rotational alignment yields a tighter upper bound on this error. Extensive experiments on natural language understanding and generative tasks demonstrate that FedRot-LoRA consistently outperforms existing federated LoRA baselines across a range of heterogeneity levels and LoRA ranks. △ Less

Submitted 26 February, 2026; originally announced February 2026.

Comments: preprint

MSC Class: 68T05 (Primary); 68T07 (Secondary) ACM Class: I.2.11; I.2.7

KnapSpec: Self-Speculative Decoding via Adaptive Layer Selection as a Knapsack Problem

Authors: Seongjin Cha, Gyuwan Kim, Dongsu Han, Tao Yang, Insu Han

Abstract: Self-speculative decoding (SSD) accelerates LLM inference by skipping layers to create an efficient draft model, yet existing methods often rely on static heuristics that ignore the dynamic computational overhead of attention in long-context scenarios. We propose KnapSpec, a training-free framework that reformulates draft model selection as a knapsack problem to maximize tokens-per-time throughput… ▽ More Self-speculative decoding (SSD) accelerates LLM inference by skipping layers to create an efficient draft model, yet existing methods often rely on static heuristics that ignore the dynamic computational overhead of attention in long-context scenarios. We propose KnapSpec, a training-free framework that reformulates draft model selection as a knapsack problem to maximize tokens-per-time throughput. By decoupling Attention and MLP layers and modeling their hardware-specific latencies as functions of context length, KnapSpec adaptively identifies optimal draft configurations on the fly via a parallel dynamic programming algorithm. Furthermore, we provide the first rigorous theoretical analysis establishing cosine similarity between hidden states as a mathematically sound proxy for the token acceptance rate. This foundation allows our method to maintain high drafting faithfulness while navigating the shifting bottlenecks of real-world hardware. Our experiments on Qwen3 and Llama3 demonstrate that KnapSpec consistently outperforms state-of-the-art SSD baselines, achieving up to 1.47x wall-clock speedup across various benchmarks. Our plug-and-play approach ensures high-speed inference for long sequences without requiring additional training or compromising the target model's output distribution. △ Less

Submitted 23 February, 2026; originally announced February 2026.

Voxtral Realtime

Authors: Mistral-AI, :, Alexander H. Liu, Andy Ehrenberg, Andy Lo, Chen-Yo Sun, Guillaume Lample, Jean-Malo Delignon, Khyathi Raghavi Chandu, Patrick von Platen, Pavankumar Reddy Muddireddy, Rohin Arora, Sanchit Gandhi, Sandeep Subramanian, Soham Ghosh, Srijan Mishra, Abhinav Rastogi, Adrien Sadé, Alan Jeffares, Albert Jiang, Alexandre Cahill, Alexandre Gavaudan, Alexandre Sablayrolles, Amélie Héliou, Amos You , et al. (144 additional authors not shown)

Abstract: We introduce Voxtral Realtime, a natively streaming automatic speech recognition model that matches offline transcription quality at sub-second latency. Unlike approaches that adapt offline models through chunking or sliding windows, Voxtral Realtime is trained end-to-end for streaming, with explicit alignment between audio and text streams. Our architecture builds on the Delayed Streams Modeling… ▽ More We introduce Voxtral Realtime, a natively streaming automatic speech recognition model that matches offline transcription quality at sub-second latency. Unlike approaches that adapt offline models through chunking or sliding windows, Voxtral Realtime is trained end-to-end for streaming, with explicit alignment between audio and text streams. Our architecture builds on the Delayed Streams Modeling framework, introducing a new causal audio encoder and Ada RMS-Norm for improved delay conditioning. We scale pretraining to a large-scale dataset spanning 13 languages. At a delay of 480ms, Voxtral Realtime achieves performance on par with Whisper, the most widely deployed offline transcription system. We release the model weights under the Apache 2.0 license. △ Less

Submitted 6 April, 2026; v1 submitted 11 February, 2026; originally announced February 2026.

Regularized Calibration with Successive Rounding for Post-Training Quantization

Authors: Seohyeon Cha, Huancheng Chen, Dongjun Kim, Haoran Zhang, Kevin Chan, Gustavo de Veciana, Haris Vikalo

Abstract: Large language models (LLMs) deliver robust performance across diverse applications, yet their deployment often faces challenges due to the memory and latency costs of storing and accessing billions of parameters. Post-training quantization (PTQ) enables efficient inference by mapping pretrained weights to low-bit formats without retraining, but its effectiveness depends critically on both the qua… ▽ More Large language models (LLMs) deliver robust performance across diverse applications, yet their deployment often faces challenges due to the memory and latency costs of storing and accessing billions of parameters. Post-training quantization (PTQ) enables efficient inference by mapping pretrained weights to low-bit formats without retraining, but its effectiveness depends critically on both the quantization objective and the rounding procedure used to obtain low-bit weight representations. In this work, we show that interpolating between symmetric and asymmetric calibration acts as a form of regularization that preserves the standard quadratic structure used in PTQ while providing robustness to activation mismatch. Building on this perspective, we derive a simple successive rounding procedure that naturally incorporates asymmetric calibration, as well as a bounded-search extension that allows for an explicit trade-off between quantization quality and the compute cost. Experiments across multiple LLM families, quantization bit-widths, and benchmarks demonstrate that the proposed bounded search based on a regularized asymmetric calibration objective consistently improves perplexity and accuracy over PTQ baselines, while incurring only modest and controllable additional computational cost. △ Less

Submitted 5 February, 2026; originally announced February 2026.

Consistency-Preserving Concept Erasure via Unsafe-Safe Pairing and Directional Fisher-weighted Adaptation

Authors: Yongwoo Kim, Sungmin Cha, Hyunsoo Kim, Jaewon Lee, Donghyun Kim

Abstract: With the increasing versatility of text-to-image diffusion models, the ability to selectively erase undesirable concepts (e.g., harmful content) has become indispensable. However, existing concept erasure approaches primarily focus on removing unsafe concepts without providing guidance toward corresponding safe alternatives, which often leads to failure in preserving the structural and semantic co… ▽ More With the increasing versatility of text-to-image diffusion models, the ability to selectively erase undesirable concepts (e.g., harmful content) has become indispensable. However, existing concept erasure approaches primarily focus on removing unsafe concepts without providing guidance toward corresponding safe alternatives, which often leads to failure in preserving the structural and semantic consistency between the original and erased generations. In this paper, we propose a novel framework, PAIRed Erasing (PAIR), which reframes concept erasure from simple removal to consistency-preserving semantic realignment using unsafe-safe pairs. We first generate safe counterparts from unsafe inputs while preserving structural and semantic fidelity, forming paired unsafe-safe multimodal data. Leveraging these pairs, we introduce two key components: (1) Paired Semantic Realignment, a guided objective that uses unsafe-safe pairs to explicitly map target concepts to semantically aligned safe anchors; and (2) Fisher-weighted Initialization for DoRA, which initializes parameter-efficient low-rank adaptation matrices using unsafe-safe pairs, encouraging the generation of safe alternatives while selectively suppressing unsafe concepts. Together, these components enable fine-grained erasure that removes only the targeted concepts while maintaining overall semantic consistency. Extensive experiments demonstrate that our approach significantly outperforms state-of-the-art baselines, achieving effective concept erasure while preserving structural integrity, semantic coherence, and generation quality. △ Less

Submitted 5 February, 2026; originally announced February 2026.

Knowledge Vector Weakening: Efficient Training-free Unlearning for Large Vision-Language Models

Authors: Yejin Kim, Dongjun Hwang, Sungmin Cha, Junsuk Choe

Abstract: Large Vision-Language Models (LVLMs) are widely adopted for their strong multimodal capabilities, yet they raise serious concerns such as privacy leakage and harmful content generation. Machine unlearning has emerged as a promising solution for removing the influence of specific data from trained models. However, existing approaches largely rely on gradient-based optimization, incurring substantia… ▽ More Large Vision-Language Models (LVLMs) are widely adopted for their strong multimodal capabilities, yet they raise serious concerns such as privacy leakage and harmful content generation. Machine unlearning has emerged as a promising solution for removing the influence of specific data from trained models. However, existing approaches largely rely on gradient-based optimization, incurring substantial computational costs for large-scale LVLMs. To address this limitation, we propose Knowledge Vector Weakening (KVW), a training-free unlearning method that directly intervenes in the full model without gradient computation. KVW identifies knowledge vectors that are activated during the model's output generation on the forget set and progressively weakens their contributions, thereby preventing the model from exploiting undesirable knowledge. Experiments on the MLLMU and CLEAR benchmarks demonstrate that KVW achieves a stable forget-retain trade-off while significantly improving computational efficiency over gradient-based and LoRA-based unlearning methods. △ Less

Submitted 29 January, 2026; originally announced January 2026.

Deep Learning Based Facial Retargeting Using Local Patches

Authors: Yeonsoo Choi, Inyup Lee, Sihun Cha, Seonghyeon Kim, Sunjin Jung, Junyong Noh

Abstract: In the era of digital animation, the quest to produce lifelike facial animations for virtual characters has led to the development of various retargeting methods. While the retargeting facial motion between models of similar shapes has been very successful, challenges arise when the retargeting is performed on stylized or exaggerated 3D characters that deviate significantly from human facial struc… ▽ More In the era of digital animation, the quest to produce lifelike facial animations for virtual characters has led to the development of various retargeting methods. While the retargeting facial motion between models of similar shapes has been very successful, challenges arise when the retargeting is performed on stylized or exaggerated 3D characters that deviate significantly from human facial structures. In this scenario, it is important to consider the target character's facial structure and possible range of motion to preserve the semantics assumed by the original facial motions after the retargeting. To achieve this, we propose a local patch-based retargeting method that transfers facial animations captured in a source performance video to a target stylized 3D character. Our method consists of three modules. The Automatic Patch Extraction Module extracts local patches from the source video frame. These patches are processed through the Reenactment Module to generate correspondingly re-enacted target local patches. The Weight Estimation Module calculates the animation parameters for the target character at every frame for the creation of a complete facial animation sequence. Extensive experiments demonstrate that our method can successfully transfer the semantic meaning of source facial expressions to stylized characters with considerable variations in facial feature proportion. △ Less

Submitted 13 January, 2026; originally announced January 2026.

Comments: Eurographics 25

Journal ref: Computer Graphics Forum 2024

ReinPool: Reinforcement Learning Pooling Multi-Vector Embeddings for Retrieval System

Authors: Sungguk Cha, DongWook Kim, Mintae Kim, Youngsub Han, Byoung-Ki Jeon, Sangyeob Lee

Abstract: Multi-vector embedding models have emerged as a powerful paradigm for document retrieval, preserving fine-grained visual and textual details through token-level representations. However, this expressiveness comes at a staggering cost: storing embeddings for every token inflates index sizes by over $1000\times$ compared to single-vector approaches, severely limiting scalability. We introduce \textb… ▽ More Multi-vector embedding models have emerged as a powerful paradigm for document retrieval, preserving fine-grained visual and textual details through token-level representations. However, this expressiveness comes at a staggering cost: storing embeddings for every token inflates index sizes by over $1000\times$ compared to single-vector approaches, severely limiting scalability. We introduce \textbf{ReinPool}, a reinforcement learning framework that learns to dynamically filter and pool multi-vector embeddings into compact, retrieval-optimized representations. By training with an inverse retrieval objective and NDCG-based rewards, ReinPool identifies and retains only the most discriminative vectors without requiring manual importance annotations. On the Vidore V2 benchmark across three vision-language embedding models, ReinPool compresses multi-vector representations by $746$--$1249\times$ into single vectors while recovering 76--81\% of full multi-vector retrieval performance. Compared to static mean pooling baselines, ReinPool achieves 22--33\% absolute NDCG@3 improvement, demonstrating that learned selection significantly outperforms heuristic aggregation. △ Less

Submitted 11 January, 2026; originally announced January 2026.

Comments: 5 pages

NVIDIA Nemotron Nano V2 VL

Authors: NVIDIA, :, Amala Sanjay Deshmukh, Kateryna Chumachenko, Tuomas Rintamaki, Matthieu Le, Tyler Poon, Danial Mohseni Taheri, Ilia Karmanov, Guilin Liu, Jarno Seppanen, Guo Chen, Karan Sapra, Zhiding Yu, Adi Renduchintala, Charles Wang, Peter Jin, Arushi Goel, Mike Ranzinger, Lukas Voegtle, Philipp Fischer, Timo Roman, Wei Ping, Boxin Wang, Zhuolin Yang , et al. (99 additional authors not shown)

Abstract: We introduce Nemotron Nano V2 VL, the latest model of the Nemotron vision-language series designed for strong real-world document understanding, long video comprehension, and reasoning tasks. Nemotron Nano V2 VL delivers significant improvements over our previous model, Llama-3.1-Nemotron-Nano-VL-8B, across all vision and text domains through major enhancements in model architecture, datasets, and… ▽ More We introduce Nemotron Nano V2 VL, the latest model of the Nemotron vision-language series designed for strong real-world document understanding, long video comprehension, and reasoning tasks. Nemotron Nano V2 VL delivers significant improvements over our previous model, Llama-3.1-Nemotron-Nano-VL-8B, across all vision and text domains through major enhancements in model architecture, datasets, and training recipes. Nemotron Nano V2 VL builds on Nemotron Nano V2, a hybrid Mamba-Transformer LLM, and innovative token reduction techniques to achieve higher inference throughput in long document and video scenarios. We are releasing model checkpoints in BF16, FP8, and FP4 formats and sharing large parts of our datasets, recipes and training code. △ Less

Submitted 6 November, 2025; v1 submitted 5 November, 2025; originally announced November 2025.

ScaleDiff: Higher-Resolution Image Synthesis via Efficient and Model-Agnostic Diffusion

Authors: Sungho Koh, SeungJu Cha, Hyunwoo Oh, Kwanyoung Lee, Dong-Jin Kim

Abstract: Text-to-image diffusion models often exhibit degraded performance when generating images beyond their training resolution. Recent training-free methods can mitigate this limitation, but they often require substantial computation or are incompatible with recent Diffusion Transformer models. In this paper, we propose ScaleDiff, a model-agnostic and highly efficient framework for extending the resolu… ▽ More Text-to-image diffusion models often exhibit degraded performance when generating images beyond their training resolution. Recent training-free methods can mitigate this limitation, but they often require substantial computation or are incompatible with recent Diffusion Transformer models. In this paper, we propose ScaleDiff, a model-agnostic and highly efficient framework for extending the resolution of pretrained diffusion models without any additional training. A core component of our framework is Neighborhood Patch Attention (NPA), an efficient mechanism that reduces computational redundancy in the self-attention layer with non-overlapping patches. We integrate NPA into an SDEdit pipeline and introduce Latent Frequency Mixing (LFM) to better generate fine details. Furthermore, we apply Structure Guidance to enhance global structure during the denoising process. Experimental results demonstrate that ScaleDiff achieves state-of-the-art performance among training-free methods in terms of both image quality and inference speed on both U-Net and Diffusion Transformer architectures. △ Less

Submitted 29 October, 2025; originally announced October 2025.

Comments: NeurIPS 2025. Code: https://github.com/KSH00906/ScaleDiff

Open Korean Historical Corpus: A Millennia-Scale Diachronic Collection of Public Domain Texts

Authors: Seyoung Song, Nawon Kim, Songeun Chae, Kiwoong Park, Jiho Jin, Haneul Yoo, Kyunghyun Cho, Alice Oh

Abstract: The history of the Korean language is characterized by a discrepancy between its spoken and written forms and a pivotal shift from Chinese characters to the Hangul alphabet. However, this linguistic evolution has remained largely unexplored in NLP due to a lack of accessible historical corpora. To address this gap, we introduce the Open Korean Historical Corpus, a large-scale, openly licensed data… ▽ More The history of the Korean language is characterized by a discrepancy between its spoken and written forms and a pivotal shift from Chinese characters to the Hangul alphabet. However, this linguistic evolution has remained largely unexplored in NLP due to a lack of accessible historical corpora. To address this gap, we introduce the Open Korean Historical Corpus, a large-scale, openly licensed dataset spanning 1,300 years and 6 languages, as well as under-represented writing systems like Korean-style Sinitic (Idu) and Hanja-Hangul mixed script. This corpus contains 17.7 million documents and 5.1 billion tokens from 19 sources, ranging from the 7th century to 2025. We leverage this resource to quantitatively analyze major linguistic shifts: (1) Idu usage peaked in the 1860s before declining sharply; (2) the transition from Hanja to Hangul was a rapid transformation starting around 1890; and (3) North Korea's lexical divergence causes modern tokenizers to produce up to 51 times higher out-of-vocabulary rates. This work provides a foundational resource for quantitative diachronic analysis by capturing the history of the Korean language. Moreover, it can serve as a pre-training corpus for large language models, potentially improving their understanding of Sino-Korean vocabulary in modern Hangul as well as archaic writing systems. △ Less

Submitted 5 March, 2026; v1 submitted 28 October, 2025; originally announced October 2025.

Comments: LREC 2026

Towards a Generalizable AI for Materials Discovery: Validation through Immersion Coolant Screening

Authors: Hyunseung Kim, Dae-Woong Jeong, Changyoung Park, Won-Ji Lee, Ha-Eun Lee, Ji-Hye Lee, Rodrigo Hormazabal, Sung Moon Ko, Sumin Lee, Soorin Yim, Chanhui Lee, Sehui Han, Sang-Ho Cha, Woohyung Lim

Abstract: Artificial intelligence (AI) has emerged as a powerful accelerator of materials discovery, yet most existing models remain problem-specific, requiring additional data collection and retraining for each new property. Here we introduce and validate GATE (Geometrically Aligned Transfer Encoder) -- a generalizable AI framework that jointly learns 34 physicochemical properties spanning thermal, electri… ▽ More Artificial intelligence (AI) has emerged as a powerful accelerator of materials discovery, yet most existing models remain problem-specific, requiring additional data collection and retraining for each new property. Here we introduce and validate GATE (Geometrically Aligned Transfer Encoder) -- a generalizable AI framework that jointly learns 34 physicochemical properties spanning thermal, electrical, mechanical, and optical domains. By aligning these properties within a shared geometric space, GATE captures cross-property correlations that reduce disjoint-property bias -- a key factor causing false positives in multi-criteria screening. To demonstrate its generalizable utility, GATE -- without any problem-specific model reconfiguration -- applied to the discovery of immersion cooling fluids for data centers, a stringent real-world challenge defined by the Open Compute Project (OCP). Screening billions of candidates, GATE identified 92,861 molecules as promising for practical deployment. Four were experimentally or literarily validated, showing strong agreement with wet-lab measurements and performance comparable to or exceeding a commercial coolant. These results establish GATE as a generalizable AI platform readily applicable across diverse materials discovery tasks. △ Less

Submitted 31 October, 2025; v1 submitted 27 October, 2025; originally announced October 2025.

Comments: 16 pages, 4 figures

Are Large Reasoning Models Good Translation Evaluators? Analysis and Performance Boost

Authors: Runzhe Zhan, Zhihong Huang, Xinyi Yang, Lidia S. Chao, Min Yang, Derek F. Wong

Abstract: Recent advancements in large reasoning models (LRMs) have introduced an intermediate "thinking" process prior to generating final answers, improving their reasoning capabilities on complex downstream tasks. However, the potential of LRMs as evaluators for machine translation (MT) quality remains underexplored. We provides the first systematic analysis of LRM-as-a-judge in MT evaluation. We identif… ▽ More Recent advancements in large reasoning models (LRMs) have introduced an intermediate "thinking" process prior to generating final answers, improving their reasoning capabilities on complex downstream tasks. However, the potential of LRMs as evaluators for machine translation (MT) quality remains underexplored. We provides the first systematic analysis of LRM-as-a-judge in MT evaluation. We identify key challenges, revealing LRMs require tailored evaluation materials, tend to "overthink" simpler instances and have issues with scoring mechanisms leading to overestimation. To address these, we propose to calibrate LRM thinking by training them on synthetic, human-like thinking trajectories. Our experiments on WMT24 Metrics benchmarks demonstrate that this approach largely reduces thinking budgets by ~35x while concurrently improving evaluation performance across different LRM scales from 7B to 32B (e.g., R1-Distill-Qwen-7B achieves a +8.7 correlation point improvement). These findings highlight the potential of efficiently calibrated LRMs to advance fine-grained automatic MT evaluation. △ Less

Submitted 23 October, 2025; originally announced October 2025.

Comments: NeurIPS 2025

MX+: Pushing the Limits of Microscaling Formats for Efficient Large Language Model Serving

Authors: Jungi Lee, Junyong Park, Soohyun Cha, Jaehoon Cho, Jaewoong Sim

Abstract: Reduced-precision data formats are crucial for cost-effective serving of large language models (LLMs). While numerous reduced-precision formats have been introduced thus far, they often require intrusive modifications to the software frameworks or are rather unconventional for widespread adoption across hardware vendors. In this paper, we instead focus on recent industry-driven variants of block f… ▽ More Reduced-precision data formats are crucial for cost-effective serving of large language models (LLMs). While numerous reduced-precision formats have been introduced thus far, they often require intrusive modifications to the software frameworks or are rather unconventional for widespread adoption across hardware vendors. In this paper, we instead focus on recent industry-driven variants of block floating-point (BFP) formats and conduct a comprehensive analysis to push their limits for efficient LLM serving. Our analysis shows that existing ultra low-bit BFP variants struggle to provide reasonable language model performance due to outlier values in blocks. To address the outliers with BFPs, we propose MX+, a cost-effective and non-intrusive extension designed for seamless integration into the microscaling (MX) formats. MX+ builds on the key insight that the outlier does not need to use its exponent field in the element data type, which allows us to repurpose the exponent field as an extended mantissa to increase the precision of the outlier element. Our evaluation shows that MX+ achieves significantly higher model performance compared to the 4-bit MX format (MXFP4) with negligible storage overhead and slowdown, thus offering a compelling alternative to MXFP4 or MXFP6 for efficient LLM inference. △ Less

Submitted 16 October, 2025; originally announced October 2025.

Comments: To appear at the 58th International Symposium on Microarchitecture (MICRO 2025)

Reference-Specific Unlearning Metrics Can Hide the Truth: A Reality Check

Authors: Sungjun Cho, Dasol Hwang, Frederic Sala, Sangheum Hwang, Kyunghyun Cho, Sungmin Cha

Abstract: Current unlearning metrics for generative models evaluate success based on reference responses or classifier outputs rather than assessing the core objective: whether the unlearned model behaves indistinguishably from a model that never saw the unwanted data. This reference-specific approach creates systematic blind spots, allowing models to appear successful while retaining unwanted knowledge acc… ▽ More Current unlearning metrics for generative models evaluate success based on reference responses or classifier outputs rather than assessing the core objective: whether the unlearned model behaves indistinguishably from a model that never saw the unwanted data. This reference-specific approach creates systematic blind spots, allowing models to appear successful while retaining unwanted knowledge accessible through alternative prompts or attacks. We address these limitations by proposing Functional Alignment for Distributional Equivalence (FADE), a novel metric that measures distributional similarity between unlearned and reference models by comparing bidirectional likelihood assignments over generated samples. Unlike existing approaches that rely on predetermined references, FADE captures functional alignment across the entire output distribution, providing a principled assessment of genuine unlearning. Our experiments on the TOFU benchmark for LLM unlearning and the UnlearnCanvas benchmark for text-to-image diffusion model unlearning reveal that methods achieving near-optimal scores on traditional metrics fail to achieve distributional equivalence, with many becoming more distant from the gold standard than before unlearning. These findings expose fundamental gaps in current evaluation practices and demonstrate that FADE provides a more robust foundation for developing and assessing truly effective unlearning methods. △ Less

Submitted 14 October, 2025; originally announced October 2025.

Comments: 20 pages, 11 figures

ECO: Enhanced Code Optimization via Performance-Aware Prompting for Code-LLMs

Authors: Su-Hyeon Kim, Joonghyuk Hahn, Sooyoung Cha, Yo-Sub Han

Abstract: Code runtime optimization-the task of rewriting a given code to a faster one-remains challenging, as it requires reasoning about performance trade-offs involving algorithmic and structural choices. Recent approaches employ code-LLMs with slow-fast code pairs provided as optimization guidance, but such pair-based methods obscure the causal factors of performance gains and often lead to superficial… ▽ More Code runtime optimization-the task of rewriting a given code to a faster one-remains challenging, as it requires reasoning about performance trade-offs involving algorithmic and structural choices. Recent approaches employ code-LLMs with slow-fast code pairs provided as optimization guidance, but such pair-based methods obscure the causal factors of performance gains and often lead to superficial pattern imitation rather than genuine performance reasoning. We introduce ECO, a performance-aware prompting framework for code optimization. ECO first distills runtime optimization instructions (ROIs) from reference slow-fast code pairs; Each ROI describes root causes of inefficiency and the rationales that drive performance improvements. For a given input code, ECO in parallel employs (i) a symbolic advisor to produce a bottleneck diagnosis tailored to the code, and (ii) an ROI retriever to return related ROIs. These two outputs are then composed into a performance-aware prompt, providing actionable guidance for code-LLMs. ECO's prompts are model-agnostic, require no fine-tuning, and can be easily prepended to any code-LLM prompt. Our empirical studies highlight that ECO prompting significantly improves code-LLMs' ability to generate efficient code, achieving speedups of up to 7.81x while minimizing correctness loss. △ Less

Submitted 12 October, 2025; originally announced October 2025.

Path Drift in Large Reasoning Models:How First-Person Commitments Override Safety

Authors: Yuyi Huang, Runzhe Zhan, Lidia S. Chao, Ailin Tao, Derek F. Wong

Abstract: As large language models (LLMs) are increasingly deployed for complex reasoning tasks, Long Chain-of-Thought (Long-CoT) prompting has emerged as a key paradigm for structured inference. Despite early-stage safeguards enabled by alignment techniques such as RLHF, we identify a previously underexplored vulnerability: reasoning trajectories in Long-CoT models can drift from aligned paths, resulting i… ▽ More As large language models (LLMs) are increasingly deployed for complex reasoning tasks, Long Chain-of-Thought (Long-CoT) prompting has emerged as a key paradigm for structured inference. Despite early-stage safeguards enabled by alignment techniques such as RLHF, we identify a previously underexplored vulnerability: reasoning trajectories in Long-CoT models can drift from aligned paths, resulting in content that violates safety constraints. We term this phenomenon Path Drift. Through empirical analysis, we uncover three behavioral triggers of Path Drift: (1) first-person commitments that induce goal-driven reasoning that delays refusal signals; (2) ethical evaporation, where surface-level disclaimers bypass alignment checkpoints; (3) condition chain escalation, where layered cues progressively steer models toward unsafe completions. Building on these insights, we introduce a three-stage Path Drift Induction Framework comprising cognitive load amplification, self-role priming, and condition chain hijacking. Each stage independently reduces refusal rates, while their combination further compounds the effect. To mitigate these risks, we propose a path-level defense strategy incorporating role attribution correction and metacognitive reflection (reflective safety cues). Our findings highlight the need for trajectory-level alignment oversight in long-form reasoning beyond token-level alignment. △ Less

Submitted 11 October, 2025; originally announced October 2025.

LeMAJ (Legal LLM-as-a-Judge): Bridging Legal Reasoning and LLM Evaluation

Authors: Joseph Enguehard, Morgane Van Ermengem, Kate Atkinson, Sujeong Cha, Arijit Ghosh Chowdhury, Prashanth Kallur Ramaswamy, Jeremy Roghair, Hannah R Marlowe, Carina Suzana Negreanu, Kitty Boxall, Diana Mincu

Abstract: Evaluating large language model (LLM) outputs in the legal domain presents unique challenges due to the complex and nuanced nature of legal analysis. Current evaluation approaches either depend on reference data, which is costly to produce, or use standardized assessment methods, both of which have significant limitations for legal applications. Although LLM-as-a-Judge has emerged as a promising… ▽ More Evaluating large language model (LLM) outputs in the legal domain presents unique challenges due to the complex and nuanced nature of legal analysis. Current evaluation approaches either depend on reference data, which is costly to produce, or use standardized assessment methods, both of which have significant limitations for legal applications. Although LLM-as-a-Judge has emerged as a promising evaluation technique, its reliability and effectiveness in legal contexts depend heavily on evaluation processes unique to the legal industry and how trustworthy the evaluation appears to the human legal expert. This is where existing evaluation methods currently fail and exhibit considerable variability. This paper aims to close the gap: a) we break down lengthy responses into 'Legal Data Points' (LDPs), self-contained units of information, and introduce a novel, reference-free evaluation methodology that reflects how lawyers evaluate legal answers; b) we demonstrate that our method outperforms a variety of baselines on both our proprietary dataset and an open-source dataset (LegalBench); c) we show how our method correlates more closely with human expert evaluations and helps improve inter-annotator agreement; and finally d) we open source our Legal Data Points for a subset of LegalBench used in our experiments, allowing the research community to replicate our results and advance research in this vital area of LLM evaluation on legal question-answering. △ Less

Submitted 8 October, 2025; originally announced October 2025.

Comments: Published in Natural Legal Language Processing - EMNLP Workshop 2025

InsightQL: Advancing Human-Assisted Fuzzing with a Unified Code Database and Parameterized Query Interface

Authors: Wentao Gao, Renata Borovica-Gajic, Sang Kil Cha, Tian Qiu, Van-Thuan Pham

Abstract: Fuzzing is a highly effective automated testing method for uncovering software vulnerabilities. Despite advances in fuzzing techniques, such as coverage-guided greybox fuzzing, many fuzzers struggle with coverage plateaus caused by fuzz blockers, limiting their ability to find deeper vulnerabilities. Human expertise can address these challenges, but analyzing fuzzing results to guide this support… ▽ More Fuzzing is a highly effective automated testing method for uncovering software vulnerabilities. Despite advances in fuzzing techniques, such as coverage-guided greybox fuzzing, many fuzzers struggle with coverage plateaus caused by fuzz blockers, limiting their ability to find deeper vulnerabilities. Human expertise can address these challenges, but analyzing fuzzing results to guide this support remains labor-intensive. To tackle this, we introduce InsightQL, the first human-assisting framework for fuzz blocker analysis. Powered by a unified database and an intuitive parameterized query interface, InsightQL aids developers in systematically extracting insights and efficiently unblocking fuzz blockers. Our experiments on 14 popular real-world libraries from the FuzzBench benchmark demonstrate the effectiveness of InsightQL, leading to the unblocking of many fuzz blockers and considerable improvements in code coverage (up to 13.90%). △ Less

Submitted 6 October, 2025; originally announced October 2025.

Unveiling LLMs' Metaphorical Understanding: Exploring Conceptual Irrelevance, Context Leveraging and Syntactic Influence

Authors: Fengying Ye, Shanshan Wang, Lidia S. Chao, Derek F. Wong

Abstract: Metaphor analysis is a complex linguistic phenomenon shaped by context and external factors. While Large Language Models (LLMs) demonstrate advanced capabilities in knowledge integration, contextual reasoning, and creative generation, their mechanisms for metaphor comprehension remain insufficiently explored. This study examines LLMs' metaphor-processing abilities from three perspectives: (1) Conc… ▽ More Metaphor analysis is a complex linguistic phenomenon shaped by context and external factors. While Large Language Models (LLMs) demonstrate advanced capabilities in knowledge integration, contextual reasoning, and creative generation, their mechanisms for metaphor comprehension remain insufficiently explored. This study examines LLMs' metaphor-processing abilities from three perspectives: (1) Concept Mapping: using embedding space projections to evaluate how LLMs map concepts in target domains (e.g., misinterpreting "fall in love" as "drop down from love"); (2) Metaphor-Literal Repository: analyzing metaphorical words and their literal counterparts to identify inherent metaphorical knowledge; and (3) Syntactic Sensitivity: assessing how metaphorical syntactic structures influence LLMs' performance. Our findings reveal that LLMs generate 15\%-25\% conceptually irrelevant interpretations, depend on metaphorical indicators in training data rather than contextual cues, and are more sensitive to syntactic irregularities than to structural comprehension. These insights underline the limitations of LLMs in metaphor analysis and call for more robust computational approaches. △ Less

Submitted 5 October, 2025; originally announced October 2025.

Style over Story: Measuring LLM Narrative Preferences via Structured Selection

Authors: Donghoon Jung, Jiwoo Choi, Songeun Chae, Seohyon Jung

Abstract: We introduce a constraint-selection-based experiment design for measuring narrative preferences of Large Language Models (LLMs). This design offers an interpretable lens on LLMs' narrative behavior. We developed a library of 200 narratology-grounded constraints and prompted selections from six LLMs under three different instruction types: basic, quality-focused, and creativity-focused. Findings de… ▽ More We introduce a constraint-selection-based experiment design for measuring narrative preferences of Large Language Models (LLMs). This design offers an interpretable lens on LLMs' narrative behavior. We developed a library of 200 narratology-grounded constraints and prompted selections from six LLMs under three different instruction types: basic, quality-focused, and creativity-focused. Findings demonstrate that models consistently prioritize Style over narrative content elements like Event, Character, and Setting. Style preferences remain stable across models and instruction types, whereas content elements show cross-model divergence and instructional sensitivity. These results suggest that LLMs have latent narrative preferences, which should inform how the NLP community evaluates and deploys models in creative domains. △ Less

Submitted 6 January, 2026; v1 submitted 2 October, 2025; originally announced October 2025.

Exposing the Cracks: Vulnerabilities of Retrieval-Augmented LLM-based Machine Translation

Authors: Yanming Sun, Runzhe Zhan, Chi Seng Cheang, Han Wu, Xuebo Liu, Yuyao Niu, Fengying Ye, Kaixin Lan, Lidia S. Chao, Derek F. Wong

Abstract: \textbf{RE}trieval-\textbf{A}ugmented \textbf{L}LM-based \textbf{M}achine \textbf{T}ranslation (REAL-MT) shows promise for knowledge-intensive tasks like idiomatic translation, but its reliability under noisy retrieval contexts remains poorly understood despite this being a common challenge in real-world deployment. To address this gap, we propose a noise synthesis framework and new metrics to eva… ▽ More \textbf{RE}trieval-\textbf{A}ugmented \textbf{L}LM-based \textbf{M}achine \textbf{T}ranslation (REAL-MT) shows promise for knowledge-intensive tasks like idiomatic translation, but its reliability under noisy retrieval contexts remains poorly understood despite this being a common challenge in real-world deployment. To address this gap, we propose a noise synthesis framework and new metrics to evaluate the robustness of REAL-MT systematically. Using this framework, we instantiate REAL-MT with Qwen-series models, including standard LLMs and large reasoning models (LRMs) with enhanced reasoning, and evaluate their performance on idiomatic translation across high-, medium-, and low-resource language pairs under synthesized noise. Our results show that low-resource language pairs, which rely more heavily on retrieved context, degrade more severely under noise than high-resource ones and often produce nonsensical translations. Although LRMs possess enhanced reasoning capabilities, they show no improvement in error correction and are even more susceptible to noise, tending to rationalize incorrect contexts. We find that this stems from an attention shift away from the source idiom to noisy content, while confidence increases despite declining accuracy, indicating poor calibration. To mitigate these issues, we investigate training-free and fine-tuning strategies, which improve robustness at the cost of performance in clean contexts, revealing a fundamental trade-off. Our findings highlight the limitations of current approaches, underscoring the need for self-verifying integration mechanisms. △ Less

Submitted 17 November, 2025; v1 submitted 1 October, 2025; originally announced October 2025.

Comments: Accepted by AAAI 2026

Why Alignment Must Precede Distillation: A Minimal Working Explanation

Authors: Sungmin Cha, Kyunghyun Cho

Abstract: For efficiency, preference alignment is often performed on compact, knowledge-distilled (KD) models. We argue this common practice introduces a significant limitation by overlooking a key property of the alignment's reference model: its distributional recall. We show that the standard KD -> Align workflow diminishes the model's capacity to align rare yet desirable behaviors, even under strong pref… ▽ More For efficiency, preference alignment is often performed on compact, knowledge-distilled (KD) models. We argue this common practice introduces a significant limitation by overlooking a key property of the alignment's reference model: its distributional recall. We show that the standard KD -> Align workflow diminishes the model's capacity to align rare yet desirable behaviors, even under strong preference signals. We instead demonstrate that reversing the pipeline (i.e., Align -> KD) is essential: alignment must first be performed on a high-recall reference before distillation. Our contributions are threefold. First, we provide a minimal working explanation of how the reference model constrains preference alignment objectives at a fundamental level. Second, we validate this theory in a controllable Mixture-of-Gaussians experiment, where low-recall anchoring consistently results in suboptimal model performance. Finally, we demonstrate that the same phenomenon holds in LLM alignment with the SmolLM2 family: models aligned after KD fail to effectively align target behaviors, resulting in substantially lower reward and target precision. In contrast, our proposed Align -> KD pipeline robustly aligns these behaviors, yielding models with superior target-oriented metrics and lower variance. Together, these results establish reference-model recall as a first-order design choice in alignment, offering a clear principle: alignment must precede distillation. △ Less

Submitted 28 September, 2025; originally announced September 2025.

Comments: Preprint

Toward a Holistic Approach to Continual Model Merging

Authors: Hoang Phan, Sungmin Cha, Tung Lam Tran, Qi Lei

Abstract: We present a holistic framework for Continual Model Merging (CMM) that intervenes at three critical stages: pre-merging, during merging, and post-merging-to address two fundamental challenges in continual learning. In particular, conventional approaches either maintain a growing list of per-domain task vectors, leading to scalability issues or rely solely on weight-space merging when old data is i… ▽ More We present a holistic framework for Continual Model Merging (CMM) that intervenes at three critical stages: pre-merging, during merging, and post-merging-to address two fundamental challenges in continual learning. In particular, conventional approaches either maintain a growing list of per-domain task vectors, leading to scalability issues or rely solely on weight-space merging when old data is inaccessible, thereby losing crucial functional information. Our method overcomes these limitations by first fine-tuning the main model within its tangent space on domain-specific data; this linearization amplifies per-task weight disentanglement, effectively mitigating across-task interference. During merging, we leverage functional information from available optimizer states beyond mere parameter averages to avoid the need to revisit old data. Finally, a post-merging correction aligns the representation discrepancy between pre- and post-merged models, reducing bias and enhancing overall performance-all while operating under constant memory constraints without accessing historical data. Extensive experiments on standard class-incremental and domain-incremental benchmarks demonstrate that our approach not only achieves competitive performance but also provides a scalable and efficient solution to the catastrophic forgetting problem. △ Less

Submitted 19 February, 2026; v1 submitted 27 September, 2025; originally announced September 2025.

Comments: Accepted to Workshop on Continual Learning in Computer Vision, ICCV 2025

Task-Agnostic Federated Continual Learning via Replay-Free Gradient Projection

Authors: Seohyeon Cha, Huancheng Chen, Haris Vikalo

Abstract: Federated continual learning (FCL) enables distributed client devices to learn from streaming data across diverse and evolving tasks. A major challenge to continual learning, catastrophic forgetting, is exacerbated in decentralized settings by the data heterogeneity, constrained communication and privacy concerns. We propose Federated gradient Projection-based Continual Learning with Task Identity… ▽ More Federated continual learning (FCL) enables distributed client devices to learn from streaming data across diverse and evolving tasks. A major challenge to continual learning, catastrophic forgetting, is exacerbated in decentralized settings by the data heterogeneity, constrained communication and privacy concerns. We propose Federated gradient Projection-based Continual Learning with Task Identity Prediction (FedProTIP), a novel FCL framework that mitigates forgetting by projecting client updates onto the orthogonal complement of the subspace spanned by previously learned representations of the global model. This projection reduces interference with earlier tasks and preserves performance across the task sequence. To further address the challenge of task-agnostic inference, we incorporate a lightweight mechanism that leverages core bases from prior tasks to predict task identity and dynamically adjust the global model's outputs. Extensive experiments across standard FCL benchmarks demonstrate that FedProTIP significantly outperforms state-of-the-art methods in average accuracy, particularly in settings where task identities are a priori unknown. △ Less

Submitted 10 November, 2025; v1 submitted 25 September, 2025; originally announced September 2025.

Learning Majority-to-Minority Transformations with MMD and Triplet Loss for Imbalanced Classification

Authors: Suman Cha, Hyunjoong Kim

Abstract: Class imbalance in supervised classification often degrades model performance by biasing predictions toward the majority class, particularly in critical applications such as medical diagnosis and fraud detection. Traditional oversampling techniques, including SMOTE and its variants, generate synthetic minority samples via local interpolation but fail to capture global data distributions in high-di… ▽ More Class imbalance in supervised classification often degrades model performance by biasing predictions toward the majority class, particularly in critical applications such as medical diagnosis and fraud detection. Traditional oversampling techniques, including SMOTE and its variants, generate synthetic minority samples via local interpolation but fail to capture global data distributions in high-dimensional spaces. Deep generative models based on GANs offer richer distribution modeling yet suffer from training instability and mode collapse under severe imbalance. To overcome these limitations, we introduce an oversampling framework that learns a parametric transformation to map majority samples into the minority distribution. Our approach minimizes the maximum mean discrepancy (MMD) between transformed and true minority samples for global alignment, and incorporates a triplet loss regularizer to enforce boundary awareness by guiding synthesized samples toward challenging borderline regions. We evaluate our method on 29 synthetic and real-world datasets, demonstrating consistent improvements over classical and generative baselines in AUROC, G-mean, F1-score, and MCC. These results confirm the robustness, computational efficiency, and practical utility of the proposed framework for imbalanced classification tasks. △ Less

Submitted 14 September, 2025; originally announced September 2025.

Comments: .19 pages, 6 figures

Benchmarking the Detection of LLMs-Generated Modern Chinese Poetry

Authors: Shanshan Wang, Junchao Wu, Fengying Ye, Jingming Yao, Lidia S. Chao, Derek F. Wong

Abstract: The rapid development of advanced large language models (LLMs) has made AI-generated text indistinguishable from human-written text. Previous work on detecting AI-generated text has made effective progress, but has not involved modern Chinese poetry. Due to the distinctive characteristics of modern Chinese poetry, it is difficult to identify whether a poem originated from humans or AI. The prolife… ▽ More The rapid development of advanced large language models (LLMs) has made AI-generated text indistinguishable from human-written text. Previous work on detecting AI-generated text has made effective progress, but has not involved modern Chinese poetry. Due to the distinctive characteristics of modern Chinese poetry, it is difficult to identify whether a poem originated from humans or AI. The proliferation of AI-generated modern Chinese poetry has significantly disrupted the poetry ecosystem. Based on the urgency of identifying AI-generated poetry in the real Chinese world, this paper proposes a novel benchmark for detecting LLMs-generated modern Chinese poetry. We first construct a high-quality dataset, which includes both 800 poems written by six professional poets and 41,600 poems generated by four mainstream LLMs. Subsequently, we conduct systematic performance assessments of six detectors on this dataset. Experimental results demonstrate that current detectors cannot be used as reliable tools to detect modern Chinese poems generated by LLMs. The most difficult poetic features to detect are intrinsic qualities, especially style. The detection results verify the effectiveness and necessity of our proposed benchmark. Our work lays a foundation for future detection of AI-generated poetry. △ Less

Submitted 1 September, 2025; originally announced September 2025.

Comments: Accepted by EMNLP 2025

Aligning Reasoning LLMs for Materials Discovery with Physics-aware Rejection Sampling

Authors: Lee Hyun, Sohee Yoon, Jinwoo Park, Sue In Chae, Seongeon Park, Jooyeon Ahn, Yebin Jung, Youjung Chung, Hogeun Chang, Sujin Park, Myeonginn Kang, Jina Kim, Ho-Gyeong Kim, Myeonghun Jeong

Abstract: AI-driven materials discovery that couples automated experimentation with algorithmic decision-making requires process aware recipe to property predictors that are accurate, calibrated, and physically admissible. We approach this as a reasoning problem with large reasoning models (LRMs). To instill reasoning capability into language models, we curate reasoning traces from a teacher model to train… ▽ More AI-driven materials discovery that couples automated experimentation with algorithmic decision-making requires process aware recipe to property predictors that are accurate, calibrated, and physically admissible. We approach this as a reasoning problem with large reasoning models (LRMs). To instill reasoning capability into language models, we curate reasoning traces from a teacher model to train a student model. However, most training pipelines select reasoning traces using binary correctness or learned preference signals that poorly reflect physical admissibility. We introduce Physics-aware Rejection Sampling (PaRS), a training-time trace selection scheme that favors traces consistent with fundamental physics and numerically close to targets, with lightweight halting to control compute. We instantiate our framework with a large student model fine-tuned on traces synthesized by a larger teacher model, and evaluate under matched token budgets against various rejection sampling baselines. Our method improves accuracy and calibration, reduces physics-violation rates, and lowers sampling cost relative to baselines. These results indicate that modest, domain-aware constraints combined with trace-level selection provide a practical path toward reliable, efficient LRMs for process-aware property prediction and closed-loop materials design. △ Less

Submitted 2 October, 2025; v1 submitted 31 August, 2025; originally announced September 2025.

Comments: 16 pages, 6 figures

Batching-Aware Joint Model Onloading and Offloading for Hierarchical Multi-Task Inference

Authors: Seohyeon Cha, Kevin Chan, Gustavo de Veciana, Haris Vikalo

Abstract: The growing demand for intelligent services on resource-constrained edge devices has spurred the development of collaborative inference systems that distribute workloads across end devices, edge servers, and the cloud. While most existing frameworks focus on single-task, single-model scenarios, many real-world applications (e.g., autonomous driving and augmented reality) require concurrent executi… ▽ More The growing demand for intelligent services on resource-constrained edge devices has spurred the development of collaborative inference systems that distribute workloads across end devices, edge servers, and the cloud. While most existing frameworks focus on single-task, single-model scenarios, many real-world applications (e.g., autonomous driving and augmented reality) require concurrent execution of diverse tasks including detection, segmentation, and depth estimation. In this work, we propose a unified framework to jointly decide which multi-task models to deploy (onload) at clients and edge servers, and how to route queries across the hierarchy (offload) to maximize overall inference accuracy under memory, compute, and communication constraints. We formulate this as a mixed-integer program and introduce J3O (Joint Optimization of Onloading and Offloading), an alternating algorithm that (i) greedily selects models to onload via Lagrangian-relaxed submodular optimization and (ii) determines optimal offloading via constrained linear programming. We further extend J3O to account for batching at the edge, maintaining scalability under heterogeneous task loads. Experiments show J3O consistently achieves over $97\%$ of the optimal accuracy while incurring less than $15\%$ of the runtime required by the optimal solver across multi-task benchmarks. △ Less

Submitted 18 August, 2025; originally announced August 2025.

RepreGuard: Detecting LLM-Generated Text by Revealing Hidden Representation Patterns

Authors: Xin Chen, Junchao Wu, Shu Yang, Runzhe Zhan, Zeyu Wu, Ziyang Luo, Di Wang, Min Yang, Lidia S. Chao, Derek F. Wong

Abstract: Detecting content generated by large language models (LLMs) is crucial for preventing misuse and building trustworthy AI systems. Although existing detection methods perform well, their robustness in out-of-distribution (OOD) scenarios is still lacking. In this paper, we hypothesize that, compared to features used by existing detection methods, the internal representations of LLMs contain more com… ▽ More Detecting content generated by large language models (LLMs) is crucial for preventing misuse and building trustworthy AI systems. Although existing detection methods perform well, their robustness in out-of-distribution (OOD) scenarios is still lacking. In this paper, we hypothesize that, compared to features used by existing detection methods, the internal representations of LLMs contain more comprehensive and raw features that can more effectively capture and distinguish the statistical pattern differences between LLM-generated texts (LGT) and human-written texts (HWT). We validated this hypothesis across different LLMs and observed significant differences in neural activation patterns when processing these two types of texts. Based on this, we propose RepreGuard, an efficient statistics-based detection method. Specifically, we first employ a surrogate model to collect representation of LGT and HWT, and extract the distinct activation feature that can better identify LGT. We can classify the text by calculating the projection score of the text representations along this feature direction and comparing with a precomputed threshold. Experimental results show that RepreGuard outperforms all baselines with average 94.92% AUROC on both in-distribution (ID) and OOD scenarios, while also demonstrating robust resilience to various text sizes and mainstream attacks. Data and code are publicly available at: https://github.com/NLP2CT/RepreGuard △ Less

Submitted 18 August, 2025; originally announced August 2025.

Comments: Accepted to TACL 2025. This version is a pre-MIT Press publication version

Annotation-Free Reinforcement Learning Query Rewriting via Verifiable Search Reward

Authors: Sungguk Cha, DongWook Kim, Taeseung Hahn, Mintae Kim, Youngsub Han, Byoung-Ki Jeon

Abstract: Optimizing queries for Retrieval-Augmented Generation (RAG) systems poses a significant challenge, particularly across diverse modal indices. We introduce RL-QR, a novel annotation-free reinforcement learning framework for query rewriting that eliminates the need for costly human-annotated data. By leveraging verifiable search rewards derived from index-aligned synthetic queries, RL-QR overcomes h… ▽ More Optimizing queries for Retrieval-Augmented Generation (RAG) systems poses a significant challenge, particularly across diverse modal indices. We introduce RL-QR, a novel annotation-free reinforcement learning framework for query rewriting that eliminates the need for costly human-annotated data. By leveraging verifiable search rewards derived from index-aligned synthetic queries, RL-QR overcomes human-annotation dependencies, extending its applicability to various modalities and index domains. Experimental results demonstrate the framework's robustness, achieving substantial retrieval performance gains of up to 3.9$\times$ on lexical retrievers and 3.5$\times$ on semantic retrievers on the MTEB VIDORE V2 benchmark for unstructured visual documents, along with consistent 5\% to 10\% improvements on MS MARCO v2.1 and internal industrial datasets. △ Less

Submitted 12 December, 2025; v1 submitted 31 July, 2025; originally announced July 2025.

CatchPhrase: EXPrompt-Guided Encoder Adaptation for Audio-to-Image Generation

Authors: Hyunwoo Oh, SeungJu Cha, Kwanyoung Lee, Si-Woo Kim, Dong-Jin Kim

Abstract: We propose CatchPhrase, a novel audio-to-image generation framework designed to mitigate semantic misalignment between audio inputs and generated images. While recent advances in multi-modal encoders have enabled progress in cross-modal generation, ambiguity stemming from homographs and auditory illusions continues to hinder accurate alignment. To address this issue, CatchPhrase generates enriched… ▽ More We propose CatchPhrase, a novel audio-to-image generation framework designed to mitigate semantic misalignment between audio inputs and generated images. While recent advances in multi-modal encoders have enabled progress in cross-modal generation, ambiguity stemming from homographs and auditory illusions continues to hinder accurate alignment. To address this issue, CatchPhrase generates enriched cross-modal semantic prompts (EXPrompt Mining) from weak class labels by leveraging large language models (LLMs) and audio captioning models (ACMs). To address both class-level and instance-level misalignment, we apply multi-modal filtering and retrieval to select the most semantically aligned prompt for each audio sample (EXPrompt Selector). A lightweight mapping network is then trained to adapt pre-trained text-to-image generation models to audio input. Extensive experiments on multiple audio classification datasets demonstrate that CatchPhrase improves audio-to-image alignment and consistently enhances generation quality by mitigating semantic misalignment. △ Less

Submitted 24 July, 2025; originally announced July 2025.

SIDA: Synthetic Image Driven Zero-shot Domain Adaptation

Authors: Ye-Chan Kim, SeungJu Cha, Si-Woo Kim, Taewhan Kim, Dong-Jin Kim

Abstract: Zero-shot domain adaptation is a method for adapting a model to a target domain without utilizing target domain image data. To enable adaptation without target images, existing studies utilize CLIP's embedding space and text description to simulate target-like style features. Despite the previous achievements in zero-shot domain adaptation, we observe that these text-driven methods struggle to cap… ▽ More Zero-shot domain adaptation is a method for adapting a model to a target domain without utilizing target domain image data. To enable adaptation without target images, existing studies utilize CLIP's embedding space and text description to simulate target-like style features. Despite the previous achievements in zero-shot domain adaptation, we observe that these text-driven methods struggle to capture complex real-world variations and significantly increase adaptation time due to their alignment process. Instead of relying on text descriptions, we explore solutions leveraging image data, which provides diverse and more fine-grained style cues. In this work, we propose SIDA, a novel and efficient zero-shot domain adaptation method leveraging synthetic images. To generate synthetic images, we first create detailed, source-like images and apply image translation to reflect the style of the target domain. We then utilize the style features of these synthetic images as a proxy for the target domain. Based on these features, we introduce Domain Mix and Patch Style Transfer modules, which enable effective modeling of real-world variations. In particular, Domain Mix blends multiple styles to expand the intra-domain representations, and Patch Style Transfer assigns different styles to individual patches. We demonstrate the effectiveness of our method by showing state-of-the-art performance in diverse zero-shot adaptation scenarios, particularly in challenging domains. Moreover, our approach achieves high efficiency by significantly reducing the overall adaptation time. △ Less

Submitted 24 July, 2025; originally announced July 2025.

Comments: Accepted to ACM MM 2025

Cross-Modal Watermarking for Authentic Audio Recovery and Tamper Localization in Synthesized Audiovisual Forgeries

Authors: Minyoung Kim, Sehwan Park, Sungmin Cha, Paul Hongsuck Seo

Abstract: Recent advances in voice cloning and lip synchronization models have enabled Synthesized Audiovisual Forgeries (SAVFs), where both audio and visuals are manipulated to mimic a target speaker. This significantly increases the risk of misinformation by making fake content seem real. To address this issue, existing methods detect or localize manipulations but cannot recover the authentic audio that c… ▽ More Recent advances in voice cloning and lip synchronization models have enabled Synthesized Audiovisual Forgeries (SAVFs), where both audio and visuals are manipulated to mimic a target speaker. This significantly increases the risk of misinformation by making fake content seem real. To address this issue, existing methods detect or localize manipulations but cannot recover the authentic audio that conveys the semantic content of the message. This limitation reduces their effectiveness in combating audiovisual misinformation. In this work, we introduce the task of Authentic Audio Recovery (AAR) and Tamper Localization in Audio (TLA) from SAVFs and propose a cross-modal watermarking framework to embed authentic audio into visuals before manipulation. This enables AAR, TLA, and a robust defense against misinformation. Extensive experiments demonstrate the strong performance of our method in AAR and TLA against various manipulations, including voice cloning and lip synchronization. △ Less

Submitted 16 July, 2025; originally announced July 2025.

Comments: 5 pages, 2 figures, Interspeech 2025

The Role of Explanation Styles and Perceived Accuracy on Decision Making in Predictive Process Monitoring

Authors: Soobin Chae, Suhwan Lee, Hanna Hauptmann, Hajo A. Reijers, Xixi Lu

Abstract: Predictive Process Monitoring (PPM) often uses deep learning models to predict the future behavior of ongoing processes, such as predicting process outcomes. While these models achieve high accuracy, their lack of interpretability undermines user trust and adoption. Explainable AI (XAI) aims to address this challenge by providing the reasoning behind the predictions. However, current evaluations o… ▽ More Predictive Process Monitoring (PPM) often uses deep learning models to predict the future behavior of ongoing processes, such as predicting process outcomes. While these models achieve high accuracy, their lack of interpretability undermines user trust and adoption. Explainable AI (XAI) aims to address this challenge by providing the reasoning behind the predictions. However, current evaluations of XAI in PPM focus primarily on functional metrics (such as fidelity), overlooking user-centered aspects such as their effect on task performance and decision-making. This study investigates the effects of explanation styles (feature importance, rule-based, and counterfactual) and perceived AI accuracy (low or high) on decision-making in PPM. We conducted a decision-making experiment, where users were presented with the AI predictions, perceived accuracy levels, and explanations of different styles. Users' decisions were measured both before and after receiving explanations, allowing the assessment of objective metrics (Task Performance and Agreement) and subjective metrics (Decision Confidence). Our findings show that perceived accuracy and explanation style have a significant effect. △ Less

Submitted 19 June, 2025; originally announced June 2025.

Comments: Accepted at CAiSE'25

SGIC: A Self-Guided Iterative Calibration Framework for RAG

Authors: Guanhua Chen, Yutong Yao, Lidia S. Chao, Xuebo Liu, Derek F. Wong

Abstract: Recent research in retrieval-augmented generation (RAG) has concentrated on retrieving useful information from candidate documents. However, numerous methodologies frequently neglect the calibration capabilities of large language models (LLMs), which capitalize on their robust in-context reasoning prowess. This work illustrates that providing LLMs with specific cues substantially improves their ca… ▽ More Recent research in retrieval-augmented generation (RAG) has concentrated on retrieving useful information from candidate documents. However, numerous methodologies frequently neglect the calibration capabilities of large language models (LLMs), which capitalize on their robust in-context reasoning prowess. This work illustrates that providing LLMs with specific cues substantially improves their calibration efficacy, especially in multi-round calibrations. We present a new SGIC: Self-Guided Iterative Calibration Framework that employs uncertainty scores as a tool. Initially, this framework calculates uncertainty scores to determine both the relevance of each document to the query and the confidence level in the responses produced by the LLMs. Subsequently, it reevaluates these scores iteratively, amalgamating them with prior responses to refine calibration. Furthermore, we introduce an innovative approach for constructing an iterative self-calibration training set, which optimizes LLMs to efficiently harness uncertainty scores for capturing critical information and enhancing response accuracy. Our proposed framework significantly improves performance on both closed-source and open-weight LLMs. △ Less

Submitted 19 June, 2025; originally announced June 2025.