/Interpretable-part-whole-hierarchies-and-conceptual-semantic-relationships-in-neural-networks

This repository contains the code for the Implementation of the research paper "Interpretable part-whole hierarchies and conceptual-semantic relationships in neural networks" (CVPR 2022) done as a coursework in the course Deep Learning under the supervision of Dr. Mayank Vatsa

Primary LanguageJupyter Notebook

Interpretable part whole hierarchies and conceptual semantic relationships in neural networks

This repository contains the code for the Implementation of the research paper "Interpretable part-whole hierarchies and conceptual-semantic relationships in neural networks" (CVPR 2022) done as a coursework in the course Deep Learning under the supervision of Dr. Mayank Vatsa.

Authors of University of Trento - Department of Information Engineering and Computer Science - DISIVia Sommarive, 9, 38123 Povo, Trento TN:

- Nicola Garau
- Niccol ́o Bisagno
- Zeno Sambugaro
- Nicola Conci

Abstract

  • The paper discusses theinterpretability and understanding of the network's response to a given input limitations of current neural network topologies.
  • The paper introduces Agglomerator, a framework that organizes input distribution.

Datasets Used

CIFAR10

EMNIST

Architecture Used

Model Architecture

Video Demontration Link

Video Demo