Tag
This paper proposes SIGMA, a hierarchical collaboration framework for cooperative multi-agent reinforcement learning that learns robust representations under noisy observations by exploiting cooperation structures through density-based grouping and aggregation methods.
The study finds that neural language models degrade similarly under word-level noise but differently under character-level noise, with tokenization identified as the key hidden variable. It provides a method to predict model robustness without noisy evaluation and suggests noise-augmented training for install robustness.
ER-KAN is a new variant of Kolmogorov-Arnold Networks designed for data-scarce and noisy scientific machine learning, showing improved robustness and efficiency over existing KAN variants.
The paper introduces Weak-Pareto, a method that uses adjoint-consistent weak formulations and Pareto-based subset selection to discover fractional differential equations from noisy data, recovering parsimonious structures robustly across benchmarks.
This paper identifies a 'representation confidence gap' in diffusion language models: internal states detect input noise accurately but reported confidence stays high and answer ranking degrades under noise. It introduces a lightweight, training-free extraction tool that leverages hidden states to improve ranking without modifying the base model.
This paper investigates Rank-Order N-of-M codes for sparse distributed memory, disentangling representation and learning effects to evaluate noise robustness compared to contemporary neuromorphic architectures.
Introduces energy conservation as a hard physical constraint on inter-module information flow in modular neural networks, enforcing exact preservation of activation energy at module boundaries to attenuate error propagation. Experiments on CIFAR-10 and a robotic pipeline show significant improvements in noise robustness.
EchoDistill is an alignment-based noisy-to-clean self-distillation framework that improves the robustness of Audio Large Language Models (ALLMs) against real-world noise by using a frozen clean-audio teacher to guide the student model via group-relative policy optimization (GRPO). Experiments show significant improvements in semantic reliability and task performance under strong noise without additional inference costs.