Selected work

Representative Papers

anyakrakusuma: A Python Library for Entropic Schrödinger Bridges on Idealized Geometries

Jul 20, 2026

This work addresses the numerical instability of discrete static Schrödinger bridge problems—equivalently, entropy-regularized optimal transport—under high cost-to-regularization ratios by introducing a log-domain Sinkhorn–Knopp iteration algorithm. The proposed method achieves high-precision solutions even in extreme parameter regimes where the Gibbs kernel suffers severe underflow. A diagnostic framework based on information entropy, covariance analysis, and geometric metrics is developed to precisely characterize the structural properties of optimal couplings and intermediate distributions. The algorithm maintains marginal residuals as low as 10⁻⁹ and perfect marginal fidelity at cost-to-regularization ratios up to 400, demonstrating convergence across approximately eight orders of magnitude. It accurately recovers geometric transformations such as 90-degree rotations with an error of merely 0.07°, and exhibits empirical convergence rates significantly surpassing theoretical worst-case bounds.

0 citationsRead paper

Sci-Rho: A Multilingual Visually-Grounded Symbolic Benchmark for STEM Problems

Jun 06, 2026

Current evaluations of visual language models (VLMs) for symbolic reasoning are largely confined to English-centric mathematical tasks, lacking multilingual coverage and visual grounding, thereby limiting their ability to comprehensively assess robustness in STEM domains. This work introduces the first dynamic, multilingual, visually grounded symbolic STEM benchmark, spanning five disciplines and seven languages. Leveraging executable templates, it generates semantically equivalent yet syntactically diverse problem instances, each accompanied by reference solutions and step-by-step reasoning traces. We propose worst-case accuracy and step-level F1 metrics to evaluate 17 state-of-the-art VLMs, revealing a substantial performance gap between average and worst-case scenarios. Larger models demonstrate greater multilingual robustness, while attention analyses uncover cross-lingual imbalances in the alignment between visual and textual tokens.

0 citationsRead paper

The Impact of Battery Cell Configuration on Electric Vehicle Performance: An XGBoost-Based Classification with SHAP Interpretability

Mar 01, 2026

This study addresses the insufficient modeling of the nonlinear relationship between battery cell configuration and electric vehicle (EV) acceleration performance in existing literature. Leveraging a dataset of 276 EVs, this work proposes a novel approach that integrates an XGBoost classification model with SHAP interpretability analysis to categorize acceleration performance into high, medium, and low tiers, thereby systematically elucidating the influence mechanism of battery configurations. The model achieves strong predictive performance with 87.5% accuracy, a ROC-AUC of 0.968, and a Matthews Correlation Coefficient (MCC) of 0.812. Key findings reveal a performance gain inflection point associated with the number of battery cells, quantifying the trade-off between enhanced acceleration and increased system complexity. These insights offer data-driven guidance for optimizing battery system design in electric vehicles.

0 citationsRead paper

Time-Continuous Modeling for Temporal Affective Pattern Recognition in LLMs

Jan 18, 2026

This work addresses the challenge that large language models struggle to capture the continuous and evolving nature of emotional dynamics in real-world interactions. To bridge this gap, the authors propose a novel approach that integrates physics-informed neural networks (PINNs) with in-context learning, marking the first application of PINNs to affective modeling. They introduce a new dataset and conceptual framework designed to support temporally continuous emotional evolution. By leveraging the differential structure of PINNs, the method enables differentiable and interpretable temporal modeling of emotional states, significantly enhancing the authenticity and dynamic consistency of large language models in emotionally grounded dialogues.

0 citationsRead paper

Self-MedRAG: a Self-Reflective Hybrid Retrieval-Augmented Generation Framework for Reliable Medical Question Answering

Jan 08, 2026arXiv.org

This work addresses the challenge of hallucination in large language models when tackling complex medical question answering, where conventional single-retrieval retrieval-augmented generation (RAG) struggles to support multi-step reasoning. To this end, the authors propose a self-reflective hybrid RAG framework that emulates the clinical “hypothesis–verification” workflow. The approach integrates BM25 and Contriever retrievers via reciprocal rank fusion (RRF), generates answers grounded in explicit reasoning chains, and incorporates a lightweight self-reflection module—based on either natural language inference (NLI) or a large language model—to iteratively verify and refine responses. Query reformulation is further employed to enhance retrieval quality. Evaluated on MedQA and PubMedQA, the method achieves accuracy rates of 83.33% and 79.82%, respectively, significantly outperforming single-retriever baselines and effectively reducing unsupported answers.

0 citationsRead paper

Recent publications

Latest Papers

anyakrakusuma: A Python Library for Entropic Schrödinger Bridges on Idealized Geometries

Jul 20, 2026

This work addresses the numerical instability of discrete static Schrödinger bridge problems—equivalently, entropy-regularized optimal transport—under high cost-to-regularization ratios by introducing a log-domain Sinkhorn–Knopp iteration algorithm. The proposed method achieves high-precision solutions even in extreme parameter regimes where the Gibbs kernel suffers severe underflow. A diagnostic framework based on information entropy, covariance analysis, and geometric metrics is developed to precisely characterize the structural properties of optimal couplings and intermediate distributions. The algorithm maintains marginal residuals as low as 10⁻⁹ and perfect marginal fidelity at cost-to-regularization ratios up to 400, demonstrating convergence across approximately eight orders of magnitude. It accurately recovers geometric transformations such as 90-degree rotations with an error of merely 0.07°, and exhibits empirical convergence rates significantly surpassing theoretical worst-case bounds.

0 citationsRead paper

Sci-Rho: A Multilingual Visually-Grounded Symbolic Benchmark for STEM Problems

Jun 06, 2026

Current evaluations of visual language models (VLMs) for symbolic reasoning are largely confined to English-centric mathematical tasks, lacking multilingual coverage and visual grounding, thereby limiting their ability to comprehensively assess robustness in STEM domains. This work introduces the first dynamic, multilingual, visually grounded symbolic STEM benchmark, spanning five disciplines and seven languages. Leveraging executable templates, it generates semantically equivalent yet syntactically diverse problem instances, each accompanied by reference solutions and step-by-step reasoning traces. We propose worst-case accuracy and step-level F1 metrics to evaluate 17 state-of-the-art VLMs, revealing a substantial performance gap between average and worst-case scenarios. Larger models demonstrate greater multilingual robustness, while attention analyses uncover cross-lingual imbalances in the alignment between visual and textual tokens.

0 citationsRead paper

The Impact of Battery Cell Configuration on Electric Vehicle Performance: An XGBoost-Based Classification with SHAP Interpretability

Mar 01, 2026

This study addresses the insufficient modeling of the nonlinear relationship between battery cell configuration and electric vehicle (EV) acceleration performance in existing literature. Leveraging a dataset of 276 EVs, this work proposes a novel approach that integrates an XGBoost classification model with SHAP interpretability analysis to categorize acceleration performance into high, medium, and low tiers, thereby systematically elucidating the influence mechanism of battery configurations. The model achieves strong predictive performance with 87.5% accuracy, a ROC-AUC of 0.968, and a Matthews Correlation Coefficient (MCC) of 0.812. Key findings reveal a performance gain inflection point associated with the number of battery cells, quantifying the trade-off between enhanced acceleration and increased system complexity. These insights offer data-driven guidance for optimizing battery system design in electric vehicles.

0 citationsRead paper

Time-Continuous Modeling for Temporal Affective Pattern Recognition in LLMs

Jan 18, 2026

This work addresses the challenge that large language models struggle to capture the continuous and evolving nature of emotional dynamics in real-world interactions. To bridge this gap, the authors propose a novel approach that integrates physics-informed neural networks (PINNs) with in-context learning, marking the first application of PINNs to affective modeling. They introduce a new dataset and conceptual framework designed to support temporally continuous emotional evolution. By leveraging the differential structure of PINNs, the method enables differentiable and interpretable temporal modeling of emotional states, significantly enhancing the authenticity and dynamic consistency of large language models in emotionally grounded dialogues.

0 citationsRead paper

Self-MedRAG: a Self-Reflective Hybrid Retrieval-Augmented Generation Framework for Reliable Medical Question Answering

Jan 08, 2026arXiv.org

This work addresses the challenge of hallucination in large language models when tackling complex medical question answering, where conventional single-retrieval retrieval-augmented generation (RAG) struggles to support multi-step reasoning. To this end, the authors propose a self-reflective hybrid RAG framework that emulates the clinical “hypothesis–verification” workflow. The approach integrates BM25 and Contriever retrievers via reciprocal rank fusion (RRF), generates answers grounded in explicit reasoning chains, and incorporates a lightweight self-reflection module—based on either natural language inference (NLI) or a large language model—to iteratively verify and refine responses. Query reformulation is further employed to enhance retrieval quality. Evaluated on MedQA and PubMedQA, the method achieves accuracy rates of 83.33% and 79.82%, respectively, significantly outperforming single-retriever baselines and effectively reducing unsupported answers.

0 citationsRead paper