information-bottleneck

Tag

Cards List
#information-bottleneck

Emergent Sparsity in Frozen Random CNN Feature Extractors for Deep Reinforcement Learning

arXiv cs.LG · 2026-07-30 Cached

This paper reports that deep reinforcement learning agents using frozen, randomly initialized CNN feature extractors spontaneously develop extremely sparse fully-connected representations, compressing task-relevant information through very few neurons without any sparsity-inducing objective.

0 favorites 0 likes
#information-bottleneck

State Compression in Two-Agent LLM Relays: A Closed-World Study of Constraint Preservation

arXiv cs.AI · 2026-07-22 Cached

This paper evaluates hand-off compression in a two-agent LLM relay for travel planning, comparing methods like JSON extraction and narrative summarization, finding that structured representations preserve constraints better.

0 favorites 0 likes
#information-bottleneck

The SIGReg Objective as Variational Free Energy: A Theoretical Active-Inference Account of JEPA World Models

arXiv cs.LG · 2026-07-16 Cached

This paper establishes a formal correspondence between the SIGReg objective used in Joint-Embedding Predictive Architectures (JEPAs) and the Active Inference framework, showing that under certain conditions the training objective becomes an exact variational free energy. It organizes non-contrastive regularisers into an entropy-estimator hierarchy and proves theoretical results including exact information bottleneck and preserved surprise bounds for SIGReg.

0 favorites 0 likes
#information-bottleneck

Learning Generalizable Skill Policy with Data-Efficient Unsupervised RL

arXiv cs.LG · 2026-07-02 Cached

Proposes GenDa, a unified framework for unsupervised reinforcement learning that addresses non-stationary skill semantics and brittle generalization via skill relabeling and a complementary information bottleneck, significantly improving data efficiency and generalizability.

0 favorites 0 likes
#information-bottleneck

Linguistic Bias Mitigation for Spoofing Detection via Gradient Reversal and A Variational Information Bottleneck

arXiv cs.CL · 2026-07-01 Cached

This paper proposes a linguistic-invariant spoofing detection framework that uses teacher-student adversarial learning and a variational information bottleneck to mitigate linguistic bias, achieving up to a 36.2% relative reduction in equal error rate across nine datasets.

0 favorites 0 likes
#information-bottleneck

SynIB: Informational Bottleneck for Maximizing Synergy in Multimodal Learning

arXiv cs.LG · 2026-06-10 Cached

The article introduces SynIB, a scalable objective based on the information bottleneck principle that targets synergistic information in multimodal learning by penalizing confident predictions when a modality is masked, improving performance on tasks requiring cross-modal reasoning.

0 favorites 0 likes
#information-bottleneck

Cosine Misleads: Auxiliary Losses Reshape Vision Language Models, Not Their Latents

Hugging Face Daily Papers · 2026-06-04 Cached

The paper challenges the assumption that cosine alignment between supervised latents and visual targets improves accuracy in vision-language models, finding a strong negative correlation. It introduces PRISM diagnostics revealing that answers are decoded downstream from latents, not within them, and that the auxiliary loss reshapes the language model via shared parameters.

0 favorites 0 likes
#information-bottleneck

Diagnosing Multi-step Reasoning Failures in Black-box LLMs via Stepwise Confidence Attribution

arXiv cs.CL · 2026-05-20 Cached

Introduces Stepwise Confidence Attribution (SCA), a framework for assigning step-level confidence to reasoning traces from black-box LLMs without internal access, using the Information Bottleneck principle to distinguish legitimate variability from errors. Experiments show SCA reliably identifies low-confidence steps and improves self-correction success rates by up to 13.5% over answer-level feedback.

0 favorites 0 likes
#information-bottleneck

StableVLA: Towards Robust Vision-Language-Action Models without Extra Data

Hugging Face Daily Papers · 2026-05-18 Cached

This paper introduces an Information Bottleneck Adapter (IB-Adapter) for Vision-Language-Action (VLA) models to improve robustness against unseen visual disturbances without requiring extra data, achieving up to 30% improvement with minimal parameter overhead.

0 favorites 0 likes
← Back to home

Submit Feedback