VISTA: A Panoramic View of Neural Representations

TL;DR

VISTA is a new visualization pipeline that maps neural network internal representations into a 2D space, enabling intuitive exploration and interpretation of complex model behaviors.

Contribution

It introduces a novel method for visualizing high-dimensional neural representations in a semantic 2D space, aiding interpretability and analysis.

Findings

Uncovered new properties of autoencoder latents

Demonstrated effective visualization of neural representations

Facilitated interpretability across machine learning domains

Abstract

We present VISTA (Visualization of Internal States and Their Associations), a novel pipeline for visually exploring and interpreting neural network representations. VISTA addresses the challenge of analyzing vast multidimensional spaces in modern machine learning models by mapping representations into a semantic 2D space. The resulting collages visually reveal patterns and relationships within internal representations. We demonstrate VISTA's utility by applying it to sparse autoencoder latents uncovering new properties and interpretations. We review the VISTA methodology, present findings from our case study ( https://got.drib.net/latents/ ), and discuss implications for neural network interpretability across various domains of machine learning.