Selected work

Representative Papers

Deep Learning for Protein-Ligand Docking: Are We There Yet?

May 23, 2024arXiv.org

This work addresses the generalization bottleneck of deep learning (DL) methods for protein–ligand docking in realistic scenarios, focusing on three key challenges: (1) pocket-agnostic docking (i.e., without prior binding-site annotation), (2) multi-ligand cooperative docking (e.g., cofactor binding), and (3) docking into predicted apo-protein structures (critical for novel targets). To rigorously evaluate cross-domain generalization under these conditions, we introduce PoseBench—the first comprehensive, application-oriented benchmark for real-world docking—and publicly release it with support for both single- and multi-ligand evaluation. Methodologically, our approach integrates deep structural modeling, physics-informed loss functions, complex-aware clustering during training, and generative structural refinement. Experiments demonstrate that DL-based methods consistently outperform traditional algorithms overall; however, most exhibit limited generalization to multi-ligand settings. Crucially, incorporating physics-guided constraints significantly enhances robustness—particularly for apo-protein docking and unknown-pocket scenarios.

1 citationsRead paper

Machine Learning Classification Methods and Portfolio Allocation: An Examination of Market Efficiency

Jan 07, 2021Social Science Research Network

This paper frames asset pricing as a multiclass classification task, predicting whether individual stocks will outperform or underperform the market. Method: Supervised learning models generate out-of-sample trading portfolios, and a novel sequential binomial test rigorously evaluates over 3.34 million predictions for statistical significance—constituting the first systematic validation that historical information contains non-random, machine-learnable predictability. Contribution/Results: (1) Multiple classifiers pass stringent statistical tests, confirming short-term market inefficiency; (2) Model uncertainty—quantified by prediction probabilities—significantly affects portfolio performance, with high-confidence predictions yielding superior economic returns; (3) The constructed portfolios consistently outperform benchmarks out-of-sample. Collectively, this work establishes a reproducible, statistically verifiable methodological framework and empirical foundation for machine learning–driven quantitative investing.

1 citationsRead paper

Bayesian Modeling of Gibbs Point Processes via Basis Function Expansions

Aug 12, 2026

This study addresses the challenges in modeling global and local effects within inhomogeneous pairwise interaction Gibbs point processes and the lack of effective methods for testing complete spatial randomness (CSR). To overcome these limitations, the authors propose a hierarchical Bayesian framework that, for the first time, integrates basis function expansions with Bayesian hierarchical modeling to flexibly characterize both the intensity and interaction functions. Building on posterior inference, they develop a Bayesian testing procedure specifically designed for CSR assessment. The approach enables efficient inference via Markov chain Monte Carlo (MCMC) and demonstrates strong empirical performance: when applied to water strider distribution and forest fire data, it successfully uncovers complex spatial dependence structures and provides reliable CSR tests, substantially enhancing the flexibility and inferential power of Gibbs point process modeling.

0 citationsRead paper

A Symmetric Layer-Union Audit of Component Collapse in Hierarchical Procedural Corpora

Aug 09, 2026

This study investigates the conditions and mechanisms under which component collapse—manifested as the emergence of a giant connected component—occurs in hierarchical procedural corpora when content-approximate duplicate layers and shared container layers are used individually or in combination. To this end, the authors propose a symmetric, bounded auditing framework that preserves relational family visibility and integrate graph component analysis, bridge predictors, union density control, threshold sensitivity diagnostics, and lexical cue coverage evaluation. Systematic experiments across six corpora reveal that in MyFixit and Doc2Dial (2 out of 6), single-layer configurations can be effective while their union fails. Although no bridge-specific mechanism is identified, the work rigorously delineates multiple explanatory boundaries, thereby preventing causal overgeneralization and offering empirical constraints and methodological guidance for modeling hierarchical relations.

0 citationsRead paper

AutoCue: Multimodal LLM-Assisted Externalization of Implicit Inputs as Instructional Visual Cues in Screencast Tutorials

Aug 05, 2026

This study addresses the frequent disruptions learners experience when watching screen-recorded software tutorials, often caused by missing implicit inputs such as mouse actions and keyboard modifier keys. To mitigate this issue, the authors propose AutoCue—a novel approach that combines multimodal large language models with human-in-the-loop refinement to infer and explicitly visualize hidden user interactions. By analyzing visual frame transitions, spoken narration, and official documentation, AutoCue introduces a first-of-its-kind visual cue grammar tailored for software learning, generating editable overlay prompts that effectively reduce cognitive load. Experimental results on Maya tutorials demonstrate that AutoCue significantly decreases task completion time, minimizes interaction breakdowns, and enhances overall learning experience.

0 citationsRead paper

Recent publications

Latest Papers

Bayesian Modeling of Gibbs Point Processes via Basis Function Expansions

Aug 12, 2026

This study addresses the challenges in modeling global and local effects within inhomogeneous pairwise interaction Gibbs point processes and the lack of effective methods for testing complete spatial randomness (CSR). To overcome these limitations, the authors propose a hierarchical Bayesian framework that, for the first time, integrates basis function expansions with Bayesian hierarchical modeling to flexibly characterize both the intensity and interaction functions. Building on posterior inference, they develop a Bayesian testing procedure specifically designed for CSR assessment. The approach enables efficient inference via Markov chain Monte Carlo (MCMC) and demonstrates strong empirical performance: when applied to water strider distribution and forest fire data, it successfully uncovers complex spatial dependence structures and provides reliable CSR tests, substantially enhancing the flexibility and inferential power of Gibbs point process modeling.

0 citationsRead paper

A Symmetric Layer-Union Audit of Component Collapse in Hierarchical Procedural Corpora

Aug 09, 2026

This study investigates the conditions and mechanisms under which component collapse—manifested as the emergence of a giant connected component—occurs in hierarchical procedural corpora when content-approximate duplicate layers and shared container layers are used individually or in combination. To this end, the authors propose a symmetric, bounded auditing framework that preserves relational family visibility and integrate graph component analysis, bridge predictors, union density control, threshold sensitivity diagnostics, and lexical cue coverage evaluation. Systematic experiments across six corpora reveal that in MyFixit and Doc2Dial (2 out of 6), single-layer configurations can be effective while their union fails. Although no bridge-specific mechanism is identified, the work rigorously delineates multiple explanatory boundaries, thereby preventing causal overgeneralization and offering empirical constraints and methodological guidance for modeling hierarchical relations.

0 citationsRead paper

AutoCue: Multimodal LLM-Assisted Externalization of Implicit Inputs as Instructional Visual Cues in Screencast Tutorials

Aug 05, 2026

This study addresses the frequent disruptions learners experience when watching screen-recorded software tutorials, often caused by missing implicit inputs such as mouse actions and keyboard modifier keys. To mitigate this issue, the authors propose AutoCue—a novel approach that combines multimodal large language models with human-in-the-loop refinement to infer and explicitly visualize hidden user interactions. By analyzing visual frame transitions, spoken narration, and official documentation, AutoCue introduces a first-of-its-kind visual cue grammar tailored for software learning, generating editable overlay prompts that effectively reduce cognitive load. Experimental results on Maya tutorials demonstrate that AutoCue significantly decreases task completion time, minimizes interaction breakdowns, and enhances overall learning experience.

0 citationsRead paper

metasignal: A Python Package for Comprehensive Metacognitive Analysis and Decision-Making

Jul 31, 2026

This study addresses the lack of a unified, open-source analytical framework in metacognition research, which has hindered methodological standardization and reproducibility. To bridge this gap, the authors introduce an open-source Python toolkit that, for the first time, integrates 17 distinct metacognitive metrics—spanning parametric, non-parametric, and model-based approaches—within a signal detection theory framework. The toolkit enables automatic computation of perceptual sensitivity, response bias, confidence, and multiple metacognitive indices directly from single-trial data, while also offering bootstrap confidence intervals, permutation tests, hierarchical Bayesian modeling, and information-theoretic analyses. Designed with both command-line accessibility and built-in visualization capabilities, the package substantially lowers the barrier to entry, thereby promoting standardized and widespread application of metacognitive measurement across cognitive science and decision-making research.

0 citationsRead paper

What Carries the Signal in Pathology Foundation-Model Atlases? A Patient-Level Controlled Benchmark in Breast Cancer

Jul 31, 2026

This study aims to identify components within foundation models of histopathology that genuinely encode molecular pathway signals, rather than relying on cohort-level statistical associations. Evaluating eleven frozen backbone models on the TCGA-BRCA dataset at the patient level, the authors employ rigorous patient-wise GroupKFold cross-validation and permutation testing to assess predictive performance for four predefined gene pathways. Results demonstrate that embedding representations significantly outperform tissue composition features (maximum Δρ = +0.479, p ≤ 0.003), except in basal-like cases; geometric graph structures confer no additional benefit (Δρ ≈ 0); and a compact, interpretable set of 54 cellular count features nearly matches full model performance. UNI2 emerges as the top-performing model (Spearman ρ = 0.556 for immune pathways), providing the first patient-level validation of the true signal source.

0 citationsRead paper