BriefGPT.xyz
Feb, 2025
在神经嵌入中发现块以实现可解释性
Discovering Chunks in Neural Embeddings for Interpretability
HTML
PDF
Shuchen Wu, Stephan Alaniz, Eric Schulz, Zeynep Akata
TL;DR
本研究针对神经网络难以理解的问题提出了一种新方法,通过借鉴人类认知中的“块”处理原理,以解释人工神经网络的活动。研究表明,神经网络中的嵌入状态能够识别和提取出反映输入概念的规律性块,从而为解读神经网络提供了新的框架,具有潜在的广泛影响。
Abstract
Understanding
Neural Networks
is challenging due to their high-dimensional, interacting components. Inspired by human cognition, which processes complex sensory data by
Chunking
it into recurring entities, we pro
→