← heapsort-ai

information theory

15 items

RESEARCHarXiv CS.CL·4/9/2026

The Stepwise Informativeness Assumption: Why are Entropy Dynamics and Reasoning Correlated in LLMs?

Este artigo investiga a correlação entre a dinâmica interna de entropia e o raciocínio correto em Large Language Models (LLMs), um enigma ainda sem solução. Propõe a Hipótese de Informatividade Gradual (SIA), que afirma que os modelos raciocinam corretamente ao acumular informações relevantes sobre a resposta por meio de prefixos informativos, um processo reforçado por métodos de treinamento padrão.

28
RESEARCHarXiv CS.AI·12d ago

On the Origin of Synthetic Information by Means of Steganographic Inheritance

This research paper posits the origin of synthetic information as a core mystery in information science, drawing an analogy to the origin of species. It introduces a steganographic inheritance mechanism to help trace the evolutionary lineage of AI-generated synthetic information, acknowledging the moral implications and technical challenges.

28
RESEARCHarXiv CS.LG·19d ago

Neural Estimation of Pairwise Mutual Information in Masked Discrete Sequence Models

The paper proposes a neural framework to estimate pairwise conditional mutual information (MI) directly from the hidden states of pretrained masked diffusion models (MDMs). This method captures dependency structures and enables MI-guided parallel decoding, showing utility in Sudoku and protein sequence generation by recovering known structural constraints.

27
RESEARCHarXiv CS.LG·5/4/2026

Information-Theoretic Generalization Bounds for Stochastic Gradient Descent with Predictable Virtual Noise

This paper introduces predictable history-adaptive virtual perturbations to enhance information-theoretic generalization bounds for Stochastic Gradient Descent. This new approach allows perturbation covariances to dynamically depend on past SGD history, addressing limitations of existing methods that require fixed covariances.

27