Prism-SQA: An Interpretable and Adaptable Neural Framework for Surface Electromyography Quality Assessment
为解决sEMG信号质量评估的可解释性和适应性问题，提出Prism-SQA框架，通过生理感知源分离和验证过程，实现信号分解与质量定制。
0 citationsRead paper
Selected work
Representative Papers
AdapterMoE: A Two-Stage Hard-Routing Mixture-of-Experts Architecture for Multi-Crop Disease Recognition with Calibrated Rejection and Incremental Learning
This work addresses key limitations of conventional soft-routing Mixture-of-Experts (MoE) approaches in multi-crop disease recognition—namely expert collapse, absence of crop-specific semantic alignment, high retraining costs, and unstable out-of-distribution rejection. To overcome these issues, the authors propose a hard-routing MoE architecture featuring a RouterHead that performs crop classification and rejection based on maximum Softmax probability, complemented by an Energy+KNN dual-gating mechanism to detect distribution shifts. Each crop is assigned a dedicated adapter module, enabling disease classification and temperature-scaled calibration while keeping the EfficientNet-B0 backbone frozen. This design eliminates expert collapse, supports crop-wise incremental updates, and reduces training costs to approximately 9% of full-network fine-tuning. On PlantVillage, the method achieves accuracy statistically on par with the best baseline (Macro-F1 gap ≤ 0.24%) while balancing rejection reliability and deployment efficiency.
0 citationsRead paper
Retrieval-Augmented Generation-Based Color Restoration for Low-Light Image Enhancement
This work addresses the prevalent issue of systematic color casts—such as green-tinted skies or yellowish skin tones—in low-light image enhancement, which often arises from insufficient supervision of color channels in end-to-end training. To mitigate this, the study introduces retrieval-augmented generation (RAG) into the task for the first time, proposing a universal post-processing module. This module employs a dual-index FAISS retriever to dynamically fetch reference images from an external high-quality color knowledge base and integrates a GlobalSPHistAdaIN mechanism to enable structure-aware color transfer without requiring pixel-wise alignment. By modeling non-parametric color priors through a residual correction framework, the method significantly improves color fidelity on the LOLv1 and LOLv2 datasets and demonstrates strong compatibility with diverse front-end models—including CPGA-Net++ and LLFormer—thereby validating its effectiveness and cross-model generalization capability.
0 citationsRead paper
Robust Welfare Decentralization under Population Entry
This study addresses the challenge of preserving efficiency in economies with indivisible goods when any new agent joins, while maintaining allocations decentralized by anonymous prices. It introduces the notion of “universal entry robustness” and establishes its equivalence to aggregate welfare additivity: this property holds if and only if the sum of all agents’ valuation functions is additive. Through demand optimality analysis, welfare comparisons, and linear programming duality, the work constructs a canonical entrant to verify robustness and proves the existence of a unique anonymous price vector that simultaneously decentralizes Pareto optimal allocations in both the original economy and all its single-agent extensions—without requiring individual valuations to satisfy gross substitutability or additivity.
0 citationsRead paper
Doctor-Optimal Stability in Unitary Many-to-Many Markets
This study addresses the implementation of the doctor-optimal stable allocation in many-to-many matching markets between doctors and hospitals. By introducing the notion of weakly hospital-quasi-stable allocations, the authors demonstrate that this set forms a finite lattice whose greatest element coincides precisely with the doctor-optimal stable allocation—also the worst outcome for hospitals among all stable allocations. Leveraging choice functions and a cumulative offer mechanism, and invoking standard matching-theoretic conditions such as substitutability and irrelevance of rejected contracts, the paper establishes that under the law of aggregate demand, the proposed procedure converges to the doctor-optimal stable allocation. Furthermore, it guarantees that all stable allocations yield identical numbers of contracts for each participant.
0 citationsRead paper
Recent publications
Latest Papers
Prism-SQA: An Interpretable and Adaptable Neural Framework for Surface Electromyography Quality Assessment
为解决sEMG信号质量评估的可解释性和适应性问题，提出Prism-SQA框架，通过生理感知源分离和验证过程，实现信号分解与质量定制。
0 citationsRead paper
AdapterMoE: A Two-Stage Hard-Routing Mixture-of-Experts Architecture for Multi-Crop Disease Recognition with Calibrated Rejection and Incremental Learning
This work addresses key limitations of conventional soft-routing Mixture-of-Experts (MoE) approaches in multi-crop disease recognition—namely expert collapse, absence of crop-specific semantic alignment, high retraining costs, and unstable out-of-distribution rejection. To overcome these issues, the authors propose a hard-routing MoE architecture featuring a RouterHead that performs crop classification and rejection based on maximum Softmax probability, complemented by an Energy+KNN dual-gating mechanism to detect distribution shifts. Each crop is assigned a dedicated adapter module, enabling disease classification and temperature-scaled calibration while keeping the EfficientNet-B0 backbone frozen. This design eliminates expert collapse, supports crop-wise incremental updates, and reduces training costs to approximately 9% of full-network fine-tuning. On PlantVillage, the method achieves accuracy statistically on par with the best baseline (Macro-F1 gap ≤ 0.24%) while balancing rejection reliability and deployment efficiency.
0 citationsRead paper
Retrieval-Augmented Generation-Based Color Restoration for Low-Light Image Enhancement
This work addresses the prevalent issue of systematic color casts—such as green-tinted skies or yellowish skin tones—in low-light image enhancement, which often arises from insufficient supervision of color channels in end-to-end training. To mitigate this, the study introduces retrieval-augmented generation (RAG) into the task for the first time, proposing a universal post-processing module. This module employs a dual-index FAISS retriever to dynamically fetch reference images from an external high-quality color knowledge base and integrates a GlobalSPHistAdaIN mechanism to enable structure-aware color transfer without requiring pixel-wise alignment. By modeling non-parametric color priors through a residual correction framework, the method significantly improves color fidelity on the LOLv1 and LOLv2 datasets and demonstrates strong compatibility with diverse front-end models—including CPGA-Net++ and LLFormer—thereby validating its effectiveness and cross-model generalization capability.
0 citationsRead paper
Robust Welfare Decentralization under Population Entry
This study addresses the challenge of preserving efficiency in economies with indivisible goods when any new agent joins, while maintaining allocations decentralized by anonymous prices. It introduces the notion of “universal entry robustness” and establishes its equivalence to aggregate welfare additivity: this property holds if and only if the sum of all agents’ valuation functions is additive. Through demand optimality analysis, welfare comparisons, and linear programming duality, the work constructs a canonical entrant to verify robustness and proves the existence of a unique anonymous price vector that simultaneously decentralizes Pareto optimal allocations in both the original economy and all its single-agent extensions—without requiring individual valuations to satisfy gross substitutability or additivity.
0 citationsRead paper
Doctor-Optimal Stability in Unitary Many-to-Many Markets
This study addresses the implementation of the doctor-optimal stable allocation in many-to-many matching markets between doctors and hospitals. By introducing the notion of weakly hospital-quasi-stable allocations, the authors demonstrate that this set forms a finite lattice whose greatest element coincides precisely with the doctor-optimal stable allocation—also the worst outcome for hospitals among all stable allocations. Leveraging choice functions and a cumulative offer mechanism, and invoking standard matching-theoretic conditions such as substitutability and irrelevance of rejected contracts, the paper establishes that under the law of aggregate demand, the proposed procedure converges to the doctor-optimal stable allocation. Furthermore, it guarantees that all stable allocations yield identical numbers of contracts for each participant.
0 citationsRead paper