Interpretable features to identify neural representations

Project Overview

Understanding how neural networks represent and process information is crucial for both neuroscience and artificial intelligence. This project focuses on developing interpretable features to identify and analyze neural representations across different systems. We aim to bridge the gap between biological and artificial neural networks by creating tools that can reveal the underlying computational principles and organizational patterns in both domains.

Research Goals

• Develop interpretable features for analyzing neural representations
• Create tools for comparing representations across different neural systems
• Understand the relationship between neural representations and computational function
• Apply these methods to both biological and artificial neural networks
• Identify common principles in neural representation across different systems

Research Approach

We combine techniques from machine learning, neuroscience, and computational modeling to develop interpretable features that can reveal the structure and organization of neural representations. Our approach involves analyzing both biological neural data and artificial neural networks to identify common patterns and principles. We focus on creating tools that are both theoretically grounded and practically useful for understanding neural computation.

Key Components

• Development of interpretable feature extraction methods
• Analysis of neural representations in biological systems
• Comparison of representations across different neural networks
• Validation of interpretability methods through controlled experiments
• Application to both neuroscience and AI research questions

Related Publications

Invited Talks