Selected work

Representative Papers

Motion-Aware Reinforcement Learning For Object Localization

Jun 19, 2026

This work addresses the challenges of overshooting and instability in bounding box refinement for object detection, particularly the lack of motion consistency in temporal scenarios. To this end, the authors propose MARLNet, a proximal policy optimization (PPO)-based reinforcement learning framework for refinement. MARLNet incorporates a constant-velocity motion prior into its state representation and introduces an action smoothness penalty in the reward function, effectively decoupling the conventional IoU-based reward from physical deviation penalties to prevent reward collapse. The study also reveals an inherent representational ceiling in refinement strategies relying on cropped features. Experimental results demonstrate that MARLNet improves detection success rates by 0.011 and 0.007 (at IoU ≥ 0.5) on Pascal VOC 2012 and VisDrone 2019, respectively, while achieving stable training and significantly suppressing overshoot artifacts.

0 citationsRead paper

A generative machine learning model for designing metal hydrides applied to hydrogen storage

Jan 28, 2026International journal of hydrogen energy

This study addresses the limited scale of existing metal hydride databases, which hinders the discovery of high-performance hydrogen storage materials. To overcome this challenge, we propose a novel approach that integrates causal discovery with a lightweight generative machine learning model to generate structurally plausible and chemically novel metal hydride candidates from scarce data. By synergistically combining materials database mining, causal inference, generative modeling, and density functional theory (DFT) validation, we successfully constructed 1,000 candidate structures and identified six previously unreported chemical compositions with unique crystal structures. DFT calculations confirmed that four of these exhibit promising hydrogen storage properties, thereby significantly expanding the design space for hydrogen storage materials and accelerating the discovery of new candidates.

0 citationsRead paper

Deep Feed-Forward Neural Network for Bangla Isolated Speech Recognition

Jul 08, 2025

To address the scarcity of research on isolated-word speech recognition for Bangla and the limited modeling capacity for low-resource languages, this paper introduces a speaker-independent bilingual speech dataset comprising Bangla and English, and proposes an end-to-end classification framework based on MFCC features and a 7-layer deep feedforward neural network (DFFNN). The method bypasses complex acoustic modeling and external linguistic resources, achieving 93.42% accuracy on a moderately sized multi-class dataset—significantly outperforming existing approaches. Key contributions include: (1) the first open-source bilingual (Bangla–English) benchmark dataset for isolated-word recognition; (2) empirical validation of lightweight DFFNNs for effective isolated-word recognition in low-resource languages; and (3) a reproducible, deployable technical pathway for speech recognition in resource-constrained languages.

0 citationsRead paper

Recent publications

Latest Papers

Motion-Aware Reinforcement Learning For Object Localization

Jun 19, 2026

This work addresses the challenges of overshooting and instability in bounding box refinement for object detection, particularly the lack of motion consistency in temporal scenarios. To this end, the authors propose MARLNet, a proximal policy optimization (PPO)-based reinforcement learning framework for refinement. MARLNet incorporates a constant-velocity motion prior into its state representation and introduces an action smoothness penalty in the reward function, effectively decoupling the conventional IoU-based reward from physical deviation penalties to prevent reward collapse. The study also reveals an inherent representational ceiling in refinement strategies relying on cropped features. Experimental results demonstrate that MARLNet improves detection success rates by 0.011 and 0.007 (at IoU ≥ 0.5) on Pascal VOC 2012 and VisDrone 2019, respectively, while achieving stable training and significantly suppressing overshoot artifacts.

0 citationsRead paper

A generative machine learning model for designing metal hydrides applied to hydrogen storage

Jan 28, 2026International journal of hydrogen energy

This study addresses the limited scale of existing metal hydride databases, which hinders the discovery of high-performance hydrogen storage materials. To overcome this challenge, we propose a novel approach that integrates causal discovery with a lightweight generative machine learning model to generate structurally plausible and chemically novel metal hydride candidates from scarce data. By synergistically combining materials database mining, causal inference, generative modeling, and density functional theory (DFT) validation, we successfully constructed 1,000 candidate structures and identified six previously unreported chemical compositions with unique crystal structures. DFT calculations confirmed that four of these exhibit promising hydrogen storage properties, thereby significantly expanding the design space for hydrogen storage materials and accelerating the discovery of new candidates.

0 citationsRead paper

Deep Feed-Forward Neural Network for Bangla Isolated Speech Recognition

Jul 08, 2025

To address the scarcity of research on isolated-word speech recognition for Bangla and the limited modeling capacity for low-resource languages, this paper introduces a speaker-independent bilingual speech dataset comprising Bangla and English, and proposes an end-to-end classification framework based on MFCC features and a 7-layer deep feedforward neural network (DFFNN). The method bypasses complex acoustic modeling and external linguistic resources, achieving 93.42% accuracy on a moderately sized multi-class dataset—significantly outperforming existing approaches. Key contributions include: (1) the first open-source bilingual (Bangla–English) benchmark dataset for isolated-word recognition; (2) empirical validation of lightweight DFFNNs for effective isolated-word recognition in low-resource languages; and (3) a reproducible, deployable technical pathway for speech recognition in resource-constrained languages.

0 citationsRead paper