Selected work

Representative Papers

LoMETab: Beyond Rank-1 Ensembles for Tabular Deep Learning

May 14, 2026

Existing implicit ensemble methods for tabular data are constrained by rank-1 structures, limiting their ability to enhance predictive diversity and performance. This work proposes LoMETab, the first rank-$r$ implicit ensemble framework that strictly expands the hypothesis space of BatchEnsemble. By parameterizing member weights through low-rank factors, LoMETab introduces two controllable dimensions—adapter rank $r$ and initialization scale $\sigma_{\text{init}}$—enabling fine-grained control over inter-member diversity. The method integrates a rank-$r$ identity residual structure based on the Hadamard product, low-rank decomposition, and end-to-end training, with diversity quantified via KL divergence and decision disagreement. Experiments demonstrate that LoMETab significantly outperforms additive low-rank baselines, with $(r, \sigma_{\text{init}})$ configurations tuning member differences across several orders of magnitude; optimal settings vary by dataset, effectively overcoming current performance bottlenecks.

0 citationsRead paper

NexusSum: Hierarchical LLM Agents for Long-Form Narrative Summarization

May 30, 2025

Existing large language models (LLMs) struggle to accurately model plot progression, character relationships, and thematic coherence in long-form narrative texts (e.g., novels, films, TV series). To address this, we propose a fine-tuning-free hierarchical multi-agent LLM framework. Our approach introduces a novel “dialogue-to-description” preprocessing mechanism, integrated with narrative-driven text normalization, hierarchical summarization scheduling, and length-controllable generation. By orchestrating multiple LLMs, the framework enables chunk-level optimization and cross-granularity summary synthesis. Evaluated on book, film, and television series datasets, our method achieves a +30.0% improvement in BERTScore (F1) over state-of-the-art methods, substantially advancing the performance ceiling for narrative summarization. Moreover, it demonstrates strong cross-domain generalization capability.

0 citationsRead paper

Recent publications

Latest Papers

LoMETab: Beyond Rank-1 Ensembles for Tabular Deep Learning

May 14, 2026

Existing implicit ensemble methods for tabular data are constrained by rank-1 structures, limiting their ability to enhance predictive diversity and performance. This work proposes LoMETab, the first rank-$r$ implicit ensemble framework that strictly expands the hypothesis space of BatchEnsemble. By parameterizing member weights through low-rank factors, LoMETab introduces two controllable dimensions—adapter rank $r$ and initialization scale $\sigma_{\text{init}}$—enabling fine-grained control over inter-member diversity. The method integrates a rank-$r$ identity residual structure based on the Hadamard product, low-rank decomposition, and end-to-end training, with diversity quantified via KL divergence and decision disagreement. Experiments demonstrate that LoMETab significantly outperforms additive low-rank baselines, with $(r, \sigma_{\text{init}})$ configurations tuning member differences across several orders of magnitude; optimal settings vary by dataset, effectively overcoming current performance bottlenecks.

0 citationsRead paper

NexusSum: Hierarchical LLM Agents for Long-Form Narrative Summarization

May 30, 2025

Existing large language models (LLMs) struggle to accurately model plot progression, character relationships, and thematic coherence in long-form narrative texts (e.g., novels, films, TV series). To address this, we propose a fine-tuning-free hierarchical multi-agent LLM framework. Our approach introduces a novel “dialogue-to-description” preprocessing mechanism, integrated with narrative-driven text normalization, hierarchical summarization scheduling, and length-controllable generation. By orchestrating multiple LLMs, the framework enables chunk-level optimization and cross-granularity summary synthesis. Evaluated on book, film, and television series datasets, our method achieves a +30.0% improvement in BERTScore (F1) over state-of-the-art methods, substantially advancing the performance ceiling for narrative summarization. Moreover, it demonstrates strong cross-domain generalization capability.

0 citationsRead paper