Selected work

Representative Papers

Fuzzy-UCS Revisited: Self-Adaptation of Rule Representations in Michigan-Style Learning Fuzzy-Classifier Systems

Jul 12, 2023Annual Conference on Genetic and Evolutionary Computation

Traditional Michigan-style Learning Fuzzy Classifier Systems (LFCS) suffer from limited generalization in continuous domains due to fixed, pre-specified rule representations that cannot adapt to unknown data characteristics. To address this, we propose an adaptive rule representation mechanism featuring evolvable “fuzzy indicators”—parameters that dynamically select between crisp (hyper-rectangular) and fuzzy (triangular) membership functions, enabling online, context-aware rule-shape adaptation. This approach transcends rigid structural assumptions by unifying fuzzy logic, genetic evolution, and supervised learning within a single cohesive framework. Empirical evaluation across multiple continuous-domain benchmark tasks demonstrates that our method achieves significantly higher classification accuracy than the conventional UCS, while exhibiting superior robustness and stability under uncertainty—including noise corruption and missing values.

4 citationsRead paper

Test-Time Logit Prompting for Source-Free Missing Modality Adaptation

Sep 01, 2026

本文提出了一种名为Test-Time Logit Prompting的方法，用于在缺少模态信息的情况下提高视觉识别性能，且无需访问源训练数据。

0 citationsRead paper

Learning from Reliable Latent Prompts for Visual Recognition with Missing Modalities

Jun 29, 2026

This work addresses the significant performance degradation of multimodal visual recognition models under missing input modalities. To mitigate this issue, the authors propose a novel paradigm that introduces input-agnostic, learnable implicit prompts as intrinsic modality priors, replacing reliance on unreliable instance-specific features and establishing stable knowledge anchors for cross-modal compensation. By integrating these implicit prompts with cross-modal knowledge transfer, the method achieves state-of-the-art performance across three benchmark datasets and demonstrates remarkable robustness—even under extreme conditions with up to 90% modality missingness—while maintaining high recognition accuracy.

0 citationsRead paper

FATE: Pillar Encoding and Frequency-Aware Training for Event-Based Object Detection

Jun 15, 2026

This work addresses the challenge of adapting sparse, asynchronous event streams from neuromorphic cameras to mainstream deep learning models, which is hindered by existing approaches that rely on fixed temporal sub-binning and thereby disrupt fine-grained temporal structures, limiting high-frequency inference performance. To overcome this, the authors propose the FATE framework, which abandons conventional sub-binning and instead employs Pillar Encoding combined with continuous orthogonal polynomial bases within a macro accumulation window to approximate event dynamics, yielding dense pseudo-images that preserve rich temporal information. Furthermore, they introduce a frequency-aware training strategy augmented with a soft mean-teacher mechanism to generate high-density pseudo-labels, effectively decoupling supervision frequency during training from inference frequency. The method significantly outperforms strong baselines across multiple architectures, enabling robust object detection at up to 200 Hz with negligible increases in model parameters or inference latency.

0 citationsRead paper

MM++: Unsupervised Scale-Invariant Multilayer OOD Detection via Top-K Gated Feature Fusion

Jun 15, 2026

This work addresses the trade-off between scale invariance and hierarchical representational capacity in unsupervised out-of-distribution (OOD) detection by proposing a rigorous post-hoc framework that requires neither additional data nor model modifications. The method automatically selects discriminative intermediate layers exhibiting significant semantic compression via an entropy density descent criterion and integrates them through Ledoit-Wolf regularized covariance estimation, achieving scale-invariant and robust multi-layer feature fusion. A Top-K gating mechanism combined with Mahalanobis distance scoring is then employed within a unified feature space to produce stable OOD scores. Evaluated across diverse architectures, the approach demonstrates consistently strong performance on both near- and far-OOD samples, significantly outperforming existing unsupervised methods.

0 citationsRead paper

Recent publications

Latest Papers

Test-Time Logit Prompting for Source-Free Missing Modality Adaptation

Sep 01, 2026

本文提出了一种名为Test-Time Logit Prompting的方法，用于在缺少模态信息的情况下提高视觉识别性能，且无需访问源训练数据。

0 citationsRead paper

Learning from Reliable Latent Prompts for Visual Recognition with Missing Modalities

Jun 29, 2026

This work addresses the significant performance degradation of multimodal visual recognition models under missing input modalities. To mitigate this issue, the authors propose a novel paradigm that introduces input-agnostic, learnable implicit prompts as intrinsic modality priors, replacing reliance on unreliable instance-specific features and establishing stable knowledge anchors for cross-modal compensation. By integrating these implicit prompts with cross-modal knowledge transfer, the method achieves state-of-the-art performance across three benchmark datasets and demonstrates remarkable robustness—even under extreme conditions with up to 90% modality missingness—while maintaining high recognition accuracy.

0 citationsRead paper

FATE: Pillar Encoding and Frequency-Aware Training for Event-Based Object Detection

Jun 15, 2026

This work addresses the challenge of adapting sparse, asynchronous event streams from neuromorphic cameras to mainstream deep learning models, which is hindered by existing approaches that rely on fixed temporal sub-binning and thereby disrupt fine-grained temporal structures, limiting high-frequency inference performance. To overcome this, the authors propose the FATE framework, which abandons conventional sub-binning and instead employs Pillar Encoding combined with continuous orthogonal polynomial bases within a macro accumulation window to approximate event dynamics, yielding dense pseudo-images that preserve rich temporal information. Furthermore, they introduce a frequency-aware training strategy augmented with a soft mean-teacher mechanism to generate high-density pseudo-labels, effectively decoupling supervision frequency during training from inference frequency. The method significantly outperforms strong baselines across multiple architectures, enabling robust object detection at up to 200 Hz with negligible increases in model parameters or inference latency.

0 citationsRead paper

MM++: Unsupervised Scale-Invariant Multilayer OOD Detection via Top-K Gated Feature Fusion

Jun 15, 2026

This work addresses the trade-off between scale invariance and hierarchical representational capacity in unsupervised out-of-distribution (OOD) detection by proposing a rigorous post-hoc framework that requires neither additional data nor model modifications. The method automatically selects discriminative intermediate layers exhibiting significant semantic compression via an entropy density descent criterion and integrates them through Ledoit-Wolf regularized covariance estimation, achieving scale-invariant and robust multi-layer feature fusion. A Top-K gating mechanism combined with Mahalanobis distance scoring is then employed within a unified feature space to produce stable OOD scores. Evaluated across diverse architectures, the approach demonstrates consistently strong performance on both near- and far-OOD samples, significantly outperforming existing unsupervised methods.

0 citationsRead paper

Dominant-Layer ZO: A Single Layer Dominates Zeroth-Order Fine-Tuning of LLMs

Jun 03, 2026

This work addresses the inefficiency in zeroth-order (ZO) fine-tuning of large language models caused by unclear layer-wise contributions, which leads to unnecessary computational overhead. The study identifies a dominant decoder layer whose exclusive fine-tuning achieves performance on par with or even surpassing full-model ZO tuning. This dominant layer is determined by model architecture rather than task specifics and can be reliably identified in advance through anomalous activation patterns during forward inference. Its high sensitivity, combined with its early position in the residual stream, generates a strong optimization signal. Leveraging this insight, the authors propose an efficient single-layer ZO fine-tuning strategy that outperforms both full-model MeZO and LoRA-based ZO methods across nine benchmarks on LLaMA2-7B and Qwen3-8B, achieving up to a 4.52× speedup in training.

0 citationsRead paper