Selected work

Representative Papers

How Reliable Are Multimodal Signals of Conversational State? Evidence from Remote Dyadic Collaborative Tasks

Jul 19, 2026

This study addresses the challenge of reliably measuring conversational states—such as cognitive load and conversational dominance—from multimodal behavioral signals, balancing predictive power, cross-task generalizability, and test–retest reliability. Leveraging the AVCAffe dataset (53 dyads across nine remote collaborative tasks), the authors construct a three-dimensional assessment framework integrating interactional, acoustic, and linguistic features, enhanced by speaker normalization to improve comparability. Findings reveal that while linguistic features exhibit the strongest predictive performance for cognitive load, they generalize poorly across tasks; acoustic features show high reliability but are strongly speaker-dependent; only interactional features—such as speaking dominance duration—robustly capture within-dyad asymmetries in cognitive load. Notably, classification of conversational dominance roles remains near chance-level across conditions. This work provides both methodological guidance and empirical grounding for selecting reliable multimodal features in social signal processing.

0 citationsRead paper

Any Proof of Polynomial Hirsch Must be Completely Incoherent

Jul 13, 2026

This study addresses a coherent-path variant of the polynomial Hirsch conjecture: whether every polytope oriented by a linear objective function admits a monotone path of polynomial length that is consistent with the induced orientation. By integrating fiber polytope theory, directed graph structural analysis, and combinatorial geometric constructions, we explicitly exhibit a family of polytopes together with associated linear functions for which all coherent monotone paths are of exponential length. This result constitutes the first disproof of the conjecture within the framework of coherent monotone paths, demonstrating that approaches relying on coherence cannot resolve the original Hirsch conjecture. Furthermore, it strengthens known lower bounds in the context of the shadow simplex method, geometric transversal problems, and parametric linear optimization.

0 citationsRead paper

Predicting Cognitive Load from Speech and Interaction Dynamics in Dyadic Conversations

Jun 11, 2026

Existing approaches struggle to reliably estimate cognitive load in natural collaborative dialogue, often being confined to controlled laboratory settings. This study leverages audio recordings from 53 participant pairs engaged in nine collaborative tasks and integrates static acoustic features with dynamic and interactive cues—such as turn-taking overlap and speaking imbalance—to model multidimensional cognitive load. We propose a dual-head Gated Recurrent Unit (GRU) encoder that, for the first time, reveals robust associations between interactional dynamics in natural conversation and dimensions of cognitive load, including time pressure and mental demand. Results demonstrate that dialogic interaction signals effectively predict overall cognitive load and its key components, while also exhibiting significant links to specific interaction patterns, thereby underscoring the critical role of task structure and conversational dynamics in cognitive load modeling.

0 citationsRead paper

Efficiency-Performance Trade-offs in Neural Speaker Diarization via Structured Pruning and Low-Bit Quantization

Jun 11, 2026

Deploying streaming speaker diarization systems on resource-constrained devices requires careful trade-offs among model size, latency, and performance, particularly in time-sensitive healthcare dispatch scenarios. This work presents the first systematic evaluation of buffering strategies, latency constraints, and model compression techniques for speaker diarization on real-world streaming speech from the SIMSAMU dataset. The authors apply structured pruning and FP16/low-bit quantization to compress neural segmentation models and analyze their real-time factor and diarization error rate (DER) within a streaming inference framework. Experimental results show that FP16 quantization halves model size with negligible impact on real-time performance but increases DER by 40% relative to the baseline. The study further reveals that additional buffering does not necessarily benefit low-latency operation and delineates practical deployment boundaries for time-critical applications.

0 citationsRead paper

Data-driven Lake Water Quality Forecasting for Time Series with Missing Data using Machine Learning

Jan 21, 2026

This study addresses temporal gaps in volunteer-led lake monitoring caused by ice cover, weather, and human factors by leveraging three decades of in situ observations from 30 lakes in Maine. Missing values are handled via multiple imputation by chained equations (MICE), and a ridge regression model is developed to predict Secchi disk depth. The work innovatively introduces a joint feasibility function that simultaneously optimizes the length of recent historical data and feature selection. Remarkably, using only approximately 64 recent samples and a single predictor variable, the approach achieves 95% of the accuracy (measured by normalized mean absolute error, nMAE) of a full-history, full-feature model, substantially reducing monitoring costs. These findings demonstrate the viability of lightweight, highly efficient sampling strategies for early warning of harmful algal blooms.

0 citationsRead paper

Recent publications

Latest Papers

How Reliable Are Multimodal Signals of Conversational State? Evidence from Remote Dyadic Collaborative Tasks

Jul 19, 2026

This study addresses the challenge of reliably measuring conversational states—such as cognitive load and conversational dominance—from multimodal behavioral signals, balancing predictive power, cross-task generalizability, and test–retest reliability. Leveraging the AVCAffe dataset (53 dyads across nine remote collaborative tasks), the authors construct a three-dimensional assessment framework integrating interactional, acoustic, and linguistic features, enhanced by speaker normalization to improve comparability. Findings reveal that while linguistic features exhibit the strongest predictive performance for cognitive load, they generalize poorly across tasks; acoustic features show high reliability but are strongly speaker-dependent; only interactional features—such as speaking dominance duration—robustly capture within-dyad asymmetries in cognitive load. Notably, classification of conversational dominance roles remains near chance-level across conditions. This work provides both methodological guidance and empirical grounding for selecting reliable multimodal features in social signal processing.

0 citationsRead paper

Any Proof of Polynomial Hirsch Must be Completely Incoherent

Jul 13, 2026

This study addresses a coherent-path variant of the polynomial Hirsch conjecture: whether every polytope oriented by a linear objective function admits a monotone path of polynomial length that is consistent with the induced orientation. By integrating fiber polytope theory, directed graph structural analysis, and combinatorial geometric constructions, we explicitly exhibit a family of polytopes together with associated linear functions for which all coherent monotone paths are of exponential length. This result constitutes the first disproof of the conjecture within the framework of coherent monotone paths, demonstrating that approaches relying on coherence cannot resolve the original Hirsch conjecture. Furthermore, it strengthens known lower bounds in the context of the shadow simplex method, geometric transversal problems, and parametric linear optimization.

0 citationsRead paper

Predicting Cognitive Load from Speech and Interaction Dynamics in Dyadic Conversations

Jun 11, 2026

Existing approaches struggle to reliably estimate cognitive load in natural collaborative dialogue, often being confined to controlled laboratory settings. This study leverages audio recordings from 53 participant pairs engaged in nine collaborative tasks and integrates static acoustic features with dynamic and interactive cues—such as turn-taking overlap and speaking imbalance—to model multidimensional cognitive load. We propose a dual-head Gated Recurrent Unit (GRU) encoder that, for the first time, reveals robust associations between interactional dynamics in natural conversation and dimensions of cognitive load, including time pressure and mental demand. Results demonstrate that dialogic interaction signals effectively predict overall cognitive load and its key components, while also exhibiting significant links to specific interaction patterns, thereby underscoring the critical role of task structure and conversational dynamics in cognitive load modeling.

0 citationsRead paper

Efficiency-Performance Trade-offs in Neural Speaker Diarization via Structured Pruning and Low-Bit Quantization

Jun 11, 2026

Deploying streaming speaker diarization systems on resource-constrained devices requires careful trade-offs among model size, latency, and performance, particularly in time-sensitive healthcare dispatch scenarios. This work presents the first systematic evaluation of buffering strategies, latency constraints, and model compression techniques for speaker diarization on real-world streaming speech from the SIMSAMU dataset. The authors apply structured pruning and FP16/low-bit quantization to compress neural segmentation models and analyze their real-time factor and diarization error rate (DER) within a streaming inference framework. Experimental results show that FP16 quantization halves model size with negligible impact on real-time performance but increases DER by 40% relative to the baseline. The study further reveals that additional buffering does not necessarily benefit low-latency operation and delineates practical deployment boundaries for time-critical applications.

0 citationsRead paper

Data-driven Lake Water Quality Forecasting for Time Series with Missing Data using Machine Learning

Jan 21, 2026

This study addresses temporal gaps in volunteer-led lake monitoring caused by ice cover, weather, and human factors by leveraging three decades of in situ observations from 30 lakes in Maine. Missing values are handled via multiple imputation by chained equations (MICE), and a ridge regression model is developed to predict Secchi disk depth. The work innovatively introduces a joint feasibility function that simultaneously optimizes the length of recent historical data and feature selection. Remarkably, using only approximately 64 recent samples and a single predictor variable, the approach achieves 95% of the accuracy (measured by normalized mean absolute error, nMAE) of a full-history, full-feature model, substantially reducing monitoring costs. These findings demonstrate the viability of lightweight, highly efficient sampling strategies for early warning of harmful algal blooms.

0 citationsRead paper