-
Wrong but Useful: Trajectory Value Beyond Answer Correctness in Multi-Agent Messages
Authors:
Chih-Hsuan Yang,
Anjir Ahmed Chowdhury,
Cheng-Hau Yang,
Weijian Zheng,
Fernando Llorente,
Xiaolong Ma,
Xinyang Li,
Eliu A. Huerta,
Ian T. Foster,
Rajeev Thakur
Abstract:
Multi-agent reasoning systems often use agreement, confidence, or automated scores to decide which messages should shape a final answer. Such filtering assumes that a message likely to be correct is also worth keeping. Yet a wrong answer can contain a useful decomposition, constraint, or scientific principle. We test this distinction with Diverse Hypothesis Deliberation (DHD), a controlled measure…
▽ More
Multi-agent reasoning systems often use agreement, confidence, or automated scores to decide which messages should shape a final answer. Such filtering assumes that a message likely to be correct is also worth keeping. Yet a wrong answer can contain a useful decomposition, constraint, or scientific principle. We test this distinction with Diverse Hypothesis Deliberation (DHD), a controlled measurement protocol that caches five independently generated messages and replays the same downstream solver, called the integrator, with each message available or hidden. The replay comparison measures a message's trajectory value: whether making the message available helps or harms subsequent reasoning. Across five mathematics and science benchmarks and two openly available model families, gpt-oss-120b and gemma-4-31B-it, wrong-helpful messages appear in every benchmark-model combination. Among wrong-answer messages that change final correctness, more than four in ten changes are helpful in each model. Controlled repeats show that the number of repeatable message effects is unlikely to arise from replay variation alone (p=0.0002). A focused intervention on repeatable wrong-helpful messages finds that the complete message works best, while retaining its reasoning preserves more success than retaining only its answer; the source of the complete-message advantage remains open. Within the same problem, repeated trajectory-value evidence also identifies a better keep-or-remove choice than answer correctness alone. Answer correctness is therefore informative but does not determine trajectory value. DHD measures this missing property and produces reusable labels for learning when agents should listen.
△ Less
Submitted 14 August, 2026;
originally announced August 2026.
-
Know When To Fold 'Em: Token-Efficient LLM Synthetic Data Generation via Multi-Stage In-Flight Rejection
Authors:
Anjir Ahmed Chowdhury,
Syed Zawad,
Feng Yan
Abstract:
While synthetic data generation with large language models (LLMs) is widely used in post-training pipelines, existing approaches typically generate full outputs before applying quality filters, leading to substantial token waste on samples that are ultimately discarded. To address this, we propose Multi-Stage In-Flight Rejection (MSIFR), a lightweight, training-free framework that detects and term…
▽ More
While synthetic data generation with large language models (LLMs) is widely used in post-training pipelines, existing approaches typically generate full outputs before applying quality filters, leading to substantial token waste on samples that are ultimately discarded. To address this, we propose Multi-Stage In-Flight Rejection (MSIFR), a lightweight, training-free framework that detects and terminates low-quality generation trajectories at intermediate checkpoints before they reach full completion. MSIFR decomposes the generation process into sequential stages and applies fast rule-based validators to identify arithmetic inconsistencies, hallucination patterns, and formatting violations, enabling early rejection of faulty samples. We formalize in-flight rejection as a sequential decision process and show that any non-trivial discard policy reduces expected token consumption, with stage-wise savings increasing when rejection occurs earlier in the generation pipeline. We further demonstrate that conditional utility estimates form a martingale, ensuring that early, in-flight rejection does not bias the expected utility of retained samples. Across five instruction-tuned models and seven reasoning benchmarks, MSIFR reduces token consumption by 11%-77% as a standalone method, and up to 78.2% when combined with early-exit methods, while preserving or improving evaluation accuracy. These results confirm that MSIFR provides a practical mechanism for improving the efficiency of LLM-based synthetic data generation without additional training or architectural changes.
△ Less
Submitted 13 May, 2026;
originally announced May 2026.
-
PEML: Parameter-efficient Multi-Task Learning with Optimized Continuous Prompts
Authors:
Anjir Ahmed Chowdhury,
Syed Zawad,
Xiaolong Ma,
Xu Dong,
Feng Yan
Abstract:
Parameter-Efficient Fine-Tuning (PEFT) is widely used for adapting Large Language Models (LLMs) for various tasks. Recently, there has been an increasing demand for fine-tuning a single LLM for multiple tasks because it requires overall less data for fine-tuning thanks to the common features shared among tasks. More importantly, LLMs are resource demanding and deploying a single model for multiple…
▽ More
Parameter-Efficient Fine-Tuning (PEFT) is widely used for adapting Large Language Models (LLMs) for various tasks. Recently, there has been an increasing demand for fine-tuning a single LLM for multiple tasks because it requires overall less data for fine-tuning thanks to the common features shared among tasks. More importantly, LLMs are resource demanding and deploying a single model for multiple tasks facilitates resource consolidation and consumes significantly less resources compared to deploying individual large model for each task. Existing PEFT methods like LoRA and Prefix Tuning are designed to adapt LLMs to a specific task. LoRA and its variation focus on aligning the model itself for tasks, overlooking the importance of prompt tuning in multi-task learning while Prefix Tuning only adopts a simple architecture to optimize prompts, which limits the adaption capabilities for multi-task. To enable efficient fine-tuning for multi-task learning, it is important to co-optimize prompt optimization and model adaptation. In this work, we propose a Parameter-Efficient Multi-task Learning (\PM), which employs a neural architecture engineering method for optimizing the continuous prompts while also performing low-rank adaption for model weights. We prototype PEML by creating an automated framework for optimizing the continuous prompts and adapting model weights. We evaluate PEML against state-of-the-arts multi-task learning methods MTL-LoRA, MultiLoRa, C-Poly, and MoE, on the GLUE, SuperGLUE, Massive Multitask Language Understanding, and commonsense reasoning benchmarks. The evaluation results present an average accuracy improvement of up to 6.67%, with individual tasks showing peak gains of up to 10.75%.
△ Less
Submitted 13 May, 2026;
originally announced May 2026.
-
Performance Analysis of Machine Learning Algorithms in Chronic Kidney Disease Prediction
Authors:
Iftekhar Ahmed,
Tanzil Ebad Chowdhury,
Biggo Bushon Routh,
Nafisa Tasmiya,
Shadman Sakib,
Adil Ahmed Chowdhury
Abstract:
Kidneys are the filter of the human body. About 10% of the global population is thought to be affected by Chronic Kidney Disease (CKD), which causes kidney function to decline. To protect in danger patients from additional kidney damage, effective risk evaluation of CKD and appropriate CKD monitoring are crucial. Due to quick and precise detection capabilities, Machine Learning models can help pra…
▽ More
Kidneys are the filter of the human body. About 10% of the global population is thought to be affected by Chronic Kidney Disease (CKD), which causes kidney function to decline. To protect in danger patients from additional kidney damage, effective risk evaluation of CKD and appropriate CKD monitoring are crucial. Due to quick and precise detection capabilities, Machine Learning models can help practitioners accomplish this goal efficiently; therefore, an enormous number of diagnosis systems and processes in the healthcare sector nowadays are relying on machine learning due to its disease prediction capability. In this study, we designed and suggested disease predictive computer-aided designs for the diagnosis of CKD. The dataset for CKD is attained from the repository of machine learning of UCL, with a few missing values; those are filled in using "mean-mode" and "Random sampling method" strategies. After successfully achieving the missing data, eight ML techniques (Random Forest, SVM, Naive Bayes, Logistic Regression, KNN, XGBoost, Decision Tree, and AdaBoost) were used to establish models, and the performance evaluation comparisons among the result accuracies are measured by the techniques to find the machine learning models with the highest accuracy. Among them, Random Forest as well as Logistic Regression showed an outstanding 99% accuracy, followed by the Ada Boost, XGBoost, Naive Bayes, Decision Tree, and SVM, whereas the KNN classifier model stands last with an accuracy of 73%.
△ Less
Submitted 10 October, 2025;
originally announced October 2025.
-
Role of AI Innovation, Clean Energy and Digital Economy towards Net Zero Emission in the United States: An ARDL Approach
Authors:
Adita Sultana,
Abdullah Al Abrar Chowdhury,
Azizul Hakim Rafi,
Abdulla All Noman
Abstract:
The current paper investigates the influences of AI innovation, GDP growth, renewable energy utilization, the digital economy, and industrialization on CO2 emissions in the USA from 1990 to 2022, incorporating the ARDL methodology. The outcomes observe that AI innovation, renewable energy usage, and the digital economy reduce CO2 emissions, while GDP expansion and industrialization intensify ecosy…
▽ More
The current paper investigates the influences of AI innovation, GDP growth, renewable energy utilization, the digital economy, and industrialization on CO2 emissions in the USA from 1990 to 2022, incorporating the ARDL methodology. The outcomes observe that AI innovation, renewable energy usage, and the digital economy reduce CO2 emissions, while GDP expansion and industrialization intensify ecosystem damage. Unit root tests (ADF, PP, and DF-GLS) reveal heterogeneous integration levels amongst components, ensuring robustness in the ARDL analysis. Complementary methods (FMOLS, DOLS, and CCR) validate the results, enhancing their reliability. Pairwise Granger causality assessments identify strong unidirectional connections within CO2 emissions and AI innovation, as well as the digital economy, underscoring their significant roles in ecological sustainability. This research highlights the requirement for strategic actions to nurture equitable growth, including advancements in AI technology, green energy adoption, and environmentally conscious industrial development, to improve environmental quality in the United States.
△ Less
Submitted 24 March, 2025;
originally announced March 2025.
-
Enhancing Green Economy with Artificial Intelligence: Role of Energy Use and FDI in the United States
Authors:
Abdullah Al Abrar Chowdhury,
Azizul Hakim Rafi,
Adita Sultana,
Abdulla All Noman
Abstract:
The escalating challenge of climate change necessitates an urgent exploration of factors influencing carbon emissions. This study contributes to the discourse by examining the interplay of technological, economic, and demographic factors on environmental sustainability. This study investigates the impact of artificial intelligence (AI) innovation, economic growth, foreign direct investment (FDI),…
▽ More
The escalating challenge of climate change necessitates an urgent exploration of factors influencing carbon emissions. This study contributes to the discourse by examining the interplay of technological, economic, and demographic factors on environmental sustainability. This study investigates the impact of artificial intelligence (AI) innovation, economic growth, foreign direct investment (FDI), energy consumption, and urbanization on CO2 emissions in the United States from 1990 to 2022. Employing the ARDL framework integrated with the STIRPAT model, the findings reveal a dual narrative: while AI innovation mitigates environmental stress, economic growth, energy use, FDI, and urbanization exacerbate environmental degradation. Unit root tests (ADF, PP, and DF-GLS) confirm mixed integration levels among variables, and the ARDL bounds test establishes long-term co-integration. The analysis highlights that AI innovation positively correlates with CO2 reduction when environmental safeguards are in place, whereas GDP growth, energy consumption, FDI, and urbanization intensify CO2 emissions. Robustness checks using FMOLS, DOLS, and CCR validate the ARDL findings. Additionally, Pairwise Granger causality tests reveal significant one-way causal links between CO2 emissions and economic growth, AI innovation, energy use, FDI, and urbanization. These relationships emphasize the critical role of AI-driven technological advancements, sustainable investments, and green energy in fostering ecological sustainability. The study suggests policy measures such as encouraging green FDI, advancing AI technologies, adopting sustainable energy practices, and implementing eco-friendly urban development to promote sustainable growth in the USA.
△ Less
Submitted 19 December, 2024;
originally announced January 2025.
-
Unveiling the Role of Artificial Intelligence and Stock Market Growth in Achieving Carbon Neutrality in the United States: An ARDL Model Analysis
Authors:
Azizul Hakim Rafi,
Abdullah Al Abrar Chowdhury,
Adita Sultana,
Abdulla All Noman
Abstract:
Given the fact that climate change has become one of the most pressing problems in many countries in recent years, specialized research on how to mitigate climate change has been adopted by many countries. Within this discussion, the influence of advanced technologies in achieving carbon neutrality has been discussed. While several studies investigated how AI and Digital innovations could be used…
▽ More
Given the fact that climate change has become one of the most pressing problems in many countries in recent years, specialized research on how to mitigate climate change has been adopted by many countries. Within this discussion, the influence of advanced technologies in achieving carbon neutrality has been discussed. While several studies investigated how AI and Digital innovations could be used to reduce the environmental footprint, the actual influence of AI in reducing CO2 emissions (a proxy measuring carbon footprint) has yet to be investigated. This paper studies the role of advanced technologies in general, and Artificial Intelligence (AI) and ICT use in particular, in advancing carbon neutrality in the United States, between 2021. Secondly, this paper examines how Stock Market Growth, ICT use, Gross Domestic Product (GDP), and Population affect CO2 emissions using the STIRPAT model. After examining stationarity among the variables using a variety of unit root tests, this study concluded that there are no unit root problems across all the variables, with a mixed order of integration. The ARDL bounds test for cointegration revealed that variables in this study have a long-run relationship. Moreover, the estimates revealed from the ARDL model in the short- and long-run indicated that economic growth, stock market capitalization, and population significantly contributed to the carbon emissions in both the short-run and long-run. Conversely, AI and ICT use significantly reduced carbon emissions over both periods. Furthermore, findings were confirmed to be robust using FMOLS, DOLS, and CCR estimations. Furthermore, diagnostic tests indicated the absence of serial correlation, heteroscedasticity, and specification errors and, thus, the model was robust.
△ Less
Submitted 4 December, 2024;
originally announced December 2024.
-
GPU-parallelisation of wavelet-based grid adaptation for fast finite volume modelling: application to shallow water flows
Authors:
Alovya Ahmed Chowdhury,
Georges Kesserwani,
Charles Rougé,
Paul Richmond
Abstract:
Wavelet-based grid adaptation driven by the "multiresolution analysis" (MRA) of the Haar wavelet (HW) allows to devise an adaptive first-order finite volume (FV1) model (HWFV1) that can readily preserve the modelling fidelity of its reference uniform-grid FV1 counterpart. However, the MRA incurs a high computational cost as it involves "encoding" (coarsening), "decoding" (refining), analysing and…
▽ More
Wavelet-based grid adaptation driven by the "multiresolution analysis" (MRA) of the Haar wavelet (HW) allows to devise an adaptive first-order finite volume (FV1) model (HWFV1) that can readily preserve the modelling fidelity of its reference uniform-grid FV1 counterpart. However, the MRA incurs a high computational cost as it involves "encoding" (coarsening), "decoding" (refining), analysing and traversing modelled data across a deep hierarchy of nested, uniform grids. GPU-parallelisation of the MRA is needed to reduce its computational cost, but its algorithmic structure (1) hinders coalesced memory access on the GPU, and (2) involves an inherently sequential tree traversal problem. This work redesigns the algorithmic structure of the MRA in order to parallelise it on the GPU, addressing (1) by applying Z-order space-filling curves and addressing (2) by adopting a parallel tree traversal algorithm. This results in a GPU-parallelised HWFV1 model (GPU-HWFV1). GPU-HWFV1 is verified against its CPU predecessor (CPU-HWFV1) and its GPU-parallelised reference uniform-grid counterpart (GPU-FV1) over five shallow water flow test cases. GPU-HWFV1 preserves the modelling fidelity of GPU-FV1 while being up to 30 times faster. Compared to CPU-HWFV1, it is up to 200 times faster, suggesting the GPU-parallelised MRA could be used to speed up other FV1 models.
△ Less
Submitted 16 May, 2023; v1 submitted 12 June, 2022;
originally announced June 2022.
-
A Study of Password Security Factors among Bangladeshi Government Websites
Authors:
Adil Ahmed Chowdhury,
Farida Chowdhury,
Md Sadek Ferdous
Abstract:
The Government of Bangladesh is aggressively transforming its public service landscape by transforming public services into online services via a number of websites. The motivation is that this would be a catalyst for a transformative change in every aspect of citizen life. Some web services must be protected from any unauthorised usages and passwords remain the most widely used credential mechani…
▽ More
The Government of Bangladesh is aggressively transforming its public service landscape by transforming public services into online services via a number of websites. The motivation is that this would be a catalyst for a transformative change in every aspect of citizen life. Some web services must be protected from any unauthorised usages and passwords remain the most widely used credential mechanism for this purpose. However, if passwords are not adopted properly, they can be a cause for security breach. That is why it is important to study different aspects of password security on different websites. In this paper, we present a study of password security among 36 different Bangladeshi government websites against six carefully chosen password security heuristics. This study is the first of its kind in this domain and offers interesting insights. For example, many websites have not adopted proper security measures with respect to security. There is no password construction guideline adopted by many websites, thus creating a barrier for users to select a strong password. Some of them allow supposedly weak passwords and still do not utilise a secure HTTPS channel to transmit information over the Internet.
△ Less
Submitted 3 December, 2020;
originally announced December 2020.
-
Performance Evaluation of SCM-WDM System Using Different Linecoding
Authors:
Md. Shamim Reza,
Md. Maruf Hossain,
Adnan Ahmed Chowdhury,
S. M. Shamim Reza,
Md. Moshiur Rahman
Abstract:
This paper investigates the theoretical performance analysis for a subcarrier multiplexed (SCM) wavelength division multiplexing (WDM) optical transmission system in presence of optical beat interference (OBI) which occurs during the photo detection process. We have presented a comparison for improving the performance of SCM-WDM system in presence of OBI. Non-return-to zero (NRZ), Manchester and M…
▽ More
This paper investigates the theoretical performance analysis for a subcarrier multiplexed (SCM) wavelength division multiplexing (WDM) optical transmission system in presence of optical beat interference (OBI) which occurs during the photo detection process. We have presented a comparison for improving the performance of SCM-WDM system in presence of OBI. Non-return-to zero (NRZ), Manchester and Miller code (MC) line coding are used for performance investigation of SCM-WDM system. A suitable signal bandwidth is selected and 200 KHz is considered as channel bandwidth. Power spectrum of signal and cross component for those line coding are analyzed. Comparison results are evaluated in terms of signal to OBI ratio for the three linecoding schemes which is called signal to interference ratio (SIR). It is found that there is a significant increase in the SIR by employing Miller code compared to NRZ and Manchester for the same data rate. For example, for a number of subcarriers of 10, the achievable SIR is about -24 dB for Miller coded system compared to -46 dB for NRZ coded system and -49 dB for Manchester coded system. The results are found to be satisfactorily agreed with the expected results.
△ Less
Submitted 26 April, 2010;
originally announced April 2010.