DNNBrain

DNNBrain enables exploration of internal representations in deep neural networks (DNNs) and their mapping to biological brains for comparative analyses between DNNs and cognitive neuroscience.

Key Features:

• Integration of tools: Integrates deep learning frameworks (e.g., PyTorch, TensorFlow) with brain imaging tools to support cross-disciplinary analyses.
• Activation extraction: Extracts DNN activations for analysis of internal representations.
• Representation probing: Probes and analyzes representations at multiple levels of abstraction within DNNs.
• Cross-modal mapping: Maps DNN representations onto brain structures to relate artificial and biological representations.
• Visualization: Provides visualization capabilities for interpreting patterns in DNN and brain representations.

Scientific Applications:

• Modeling biological neural systems: Uses DNN representations to model and compare computations in biological neural systems.
• Comparative analysis: Enables drawing parallels between internal operations of DNNs and biological neural systems.
• Decoding brain representations: Supports mapping and interpretation of brain imaging data through correspondence with DNN features.
• Cognitive-neuroscience alignment: Applies cognitive neuroscience paradigms to interpret DNN internal representations.

Methodology:

Extraction of DNN activations and mapping of internal representations onto brain structures, combined with application of cognitive neuroscience paradigms.