Selected work

Representative Papers

An approach to melodic segmentation and classification based on filtering with the Haar-wavelet

Dec 01, 2013

This work addresses automatic segmentation and classification of symbolic melodies, specifically targeting two tasks: attribution of excerpts from Bach’s Two-Part Inventions (BWV 772–786) to their parent works, and assignment of 360 Dutch folk songs to one of 26 tune families. We propose a single-scale continuous Haar wavelet filtering method that models pitch sequences as time-series signals; segmentation is driven by detection of local extrema and zero crossings, followed by k-nearest neighbors classification using Euclidean or Manhattan distance. This is the first application of Haar wavelets directly for melodic structural representation and segmentation—bypassing heuristic Gestalt-based rules and enabling cross-modal integration of signal processing and music cognition. Experiments show our method significantly outperforms both unfiltered pitch sequences and Gestalt-based segmentation on Bach excerpt attribution, and achieves performance close to the pitch-based baseline on folk tune family classification—slightly below state-of-the-art multi-feature string-matching approaches.

32 citationsRead paper

Wavelet-filtering of symbolic music representations for folk tune segmentation and classification

Jun 05, 2013

This work addresses the automatic segmentation and tune-family classification of folk-song symbolic scores. We propose a multi-scale time-frequency analysis method based on the Haar continuous wavelet transform (CWT). Symbolic scores are modeled as discrete pitch–time sequences; structural features are extracted via Haar wavelet filtering, and phrase boundaries are precisely localized using local maxima of the wavelet coefficients. Tune-family identification is performed using a k-nearest neighbors classifier with either Euclidean or Manhattan distance. To our knowledge, this is the first application of Haar wavelet filtering to symbolic music segmentation and classification, significantly enhancing melodic structural perception and classification robustness. Cross-validation results demonstrate that the optimized scale-parameter configuration achieves substantially higher classification accuracy than conventional Gestalt-based baselines.

3 citationsRead paper

Neurosymbolic Embodied Agents

Aug 17, 2026

This study addresses the inexecutability and environmental inconsistency of language model-based embodied planning by proposing a neuro-symbolic agent. The approach decomposes long-horizon tasks into visual exploration and constrained symbolic planning, leveraging PDDL-constrained decoding integrated with Monte Carlo Tree Search to ensure construct-level executability without additional training. Our analysis confirms that symbolic constraints and search are complementary rather than interchangeable. Experiments demonstrate that models ranging from 4B to 27B parameters achieve over 90% success rates in household tasks. Notably, smaller models outperform larger counterparts while significantly reducing token consumption, thereby validating the method’s efficiency and practicality for embodied AI applications.

0 citationsRead paper

A Cognitively Grounded Bayesian Framework for Misinformation Susceptibility

May 10, 2026

This work addresses the susceptibility of human judgment and annotator disagreement in discerning false, misleading, and malicious information by proposing the Bounded Pragmatic Listener (BPL) model. Integrating Rational Speech Act theory with three cognitive constraints—working memory limitations, information bottlenecks, and salience-based sampling—within a Bayesian framework, BPL formally characterizes the “depth-mismatch paradox,” elucidating distinct mechanisms underlying cognitive vulnerability across types of informational disorder. By incorporating cognitive heuristics such as bounded recursion depth, prior compression, and availability-based sample size, the model achieves competitive performance on truthfulness classification benchmarks LIAR and MultiFC, while providing empirical support for the proposed paradox.

0 citationsRead paper

Learning and Enforcing Context-Sensitive Control for LLMs

Apr 12, 2026

Current approaches to controlling the outputs of large language models rely on manually defined context-sensitive constraints, which are difficult to scale and require significant expertise. This work proposes a novel two-stage framework that integrates context-sensitive grammar learning with large language model generation for the first time: it first automatically collects diverse model outputs through syntactic exploration to learn constraints, then enforces the learned rules during text generation. The method guarantees output validity without any human intervention and achieves a 100% constraint compliance rate using only a 1B-parameter model—outperforming both larger-scale models and state-of-the-art reasoning systems in constrained generation tasks.

0 citationsRead paper

Recent publications

Latest Papers

Neurosymbolic Embodied Agents

Aug 17, 2026

This study addresses the inexecutability and environmental inconsistency of language model-based embodied planning by proposing a neuro-symbolic agent. The approach decomposes long-horizon tasks into visual exploration and constrained symbolic planning, leveraging PDDL-constrained decoding integrated with Monte Carlo Tree Search to ensure construct-level executability without additional training. Our analysis confirms that symbolic constraints and search are complementary rather than interchangeable. Experiments demonstrate that models ranging from 4B to 27B parameters achieve over 90% success rates in household tasks. Notably, smaller models outperform larger counterparts while significantly reducing token consumption, thereby validating the method’s efficiency and practicality for embodied AI applications.

0 citationsRead paper

A Cognitively Grounded Bayesian Framework for Misinformation Susceptibility

May 10, 2026

This work addresses the susceptibility of human judgment and annotator disagreement in discerning false, misleading, and malicious information by proposing the Bounded Pragmatic Listener (BPL) model. Integrating Rational Speech Act theory with three cognitive constraints—working memory limitations, information bottlenecks, and salience-based sampling—within a Bayesian framework, BPL formally characterizes the “depth-mismatch paradox,” elucidating distinct mechanisms underlying cognitive vulnerability across types of informational disorder. By incorporating cognitive heuristics such as bounded recursion depth, prior compression, and availability-based sample size, the model achieves competitive performance on truthfulness classification benchmarks LIAR and MultiFC, while providing empirical support for the proposed paradox.

0 citationsRead paper

Learning and Enforcing Context-Sensitive Control for LLMs

Apr 12, 2026

Current approaches to controlling the outputs of large language models rely on manually defined context-sensitive constraints, which are difficult to scale and require significant expertise. This work proposes a novel two-stage framework that integrates context-sensitive grammar learning with large language model generation for the first time: it first automatically collects diverse model outputs through syntactic exploration to learn constraints, then enforces the learned rules during text generation. The method guarantees output validity without any human intervention and achieves a 100% constraint compliance rate using only a 1B-parameter model—outperforming both larger-scale models and state-of-the-art reasoning systems in constrained generation tasks.

0 citationsRead paper

Evolving Excellence: Automated Optimization of LLM-based Agents

Dec 09, 2025

LLM-based agents suffer from suboptimal performance due to poorly engineered prompts, ambiguous tool descriptions, and misconfigured parameters; existing optimization methods are either overly complex or neglect inter-component dependencies. This paper introduces ARTEMIS—the first end-to-end, code-free, semantics-driven framework for joint agent configuration optimization. It employs semantic-aware genetic operators to automatically evolve complete agent configurations—including prompts, tool schemas, and parameters—without architectural modifications and with full compatibility across commercial and open-source LLMs. Its core innovation lies in a multimodal evolutionary paradigm integrating log semantic parsing, automatic component discovery, and execution-signal extraction to enable cross-component co-optimization. Evaluated on four representative tasks, ARTEMIS achieves +13.6% acceptance rate, +10.1% overall performance gain, +22% improvement in mathematical accuracy, and −36.9% reduction in inference token consumption.

0 citationsRead paper

Exploring Spectral Characteristics for Single Image Reflection Removal

Sep 15, 2025

Image reflection removal is an ill-posed inverse problem, primarily due to the strong coupling between reflection and transmission layers in a single image and the lack of effective modeling of their distinct spectral responses. To address this, we propose a novel spectrum-aware reflection removal paradigm. First, we construct a spectral codebook to explicitly model the wavelength-selective nature of reflected light. Second, we design a dual-spectral prior module to enforce physical consistency and introduce a spectrum-aware Transformer that jointly models spatial and spectral (wavelength) dimensions for effective layer decoupling. Our method integrates spectral reconstruction, adaptive spectral enhancement, and cross-domain feature co-optimization. Extensive experiments on three benchmark datasets demonstrate that our approach significantly outperforms existing state-of-the-art methods, achieving superior reflection removal accuracy and enhanced generalization capability.

0 citationsRead paper