information flow
The way data and signals travel through different layers or components of a neural network, affecting how information is processed and transformed.
- CoRe: Benchmarking LLMs’ Code Reasoning Capabilities through Static Analysis Tasks
- DePass: Unified Feature Attributing by Simple Decomposed Forward Pass
- FlowCut: Rethinking Redundancy via Information Flow for Efficient Vision-Language Models
- LaX: Boosting Low-Rank Training of Foundation Models via Latent Crossing
- Learning to Flow from Generative Pretext Tasks for Neural Architecture Encoding
- SONAR: Long-Range Graph Propagation Through Information Waves
- The Narrow Gate: Localized Image-Text Communication in Native Multimodal Models