CIFAR-10 Preprocessing using Binarized and Grayscale images
Zoom In
Hebbian & Storkey Learning Reconstruction for Binarized images
Zoom In
Hebbian & Storkey Learning Reconstruction for Grayscaled images
Zoom In
Phase-Space Learning for Binarized images
Zoom In
Spiking Network Image Reconstruction for Binarized images
Zoom In

Project information

  • Category: Neural Networks / Associative Memory
  • Project duration: Feb '25 - Mar '25
  • Team size: 5
  • Dataset: CIFAR-10 (Binarized & Grayscale)
  • Github URL: Forgetful-Memory-Machines
  • Report URL: Detailed Report

Forgetful Memory Machines: Hopfield Networks

Technology(s) Used: PyTorch, NumPy, Hebbian Learning, Storkey Learning, Spiking Neural Networks (SNN), Izhikevich Models

  • Advanced Hopfield Variants: Implemented and compared Hebbian, Storkey, and Phase-Space Learning models to enhance storage capacity and robustness against masked or flipped inputs.
  • Phase-Space Learning: Developed models with iterative feedback loops that successfully stored and reconstructed up to 600 grayscale images with high precision.
  • Biologically Inspired AI: Designed Spiking Hopfield Networks using Izhikevich neuron models to explore error-correcting abilities in a temporal, spike-based dynamics environment.
  • Grayscale Encoding: Experimented with bit-level grayscale encoding and binary thresholding to adapt traditional binarized networks for complex CIFAR-10 data.
  • Evaluation: Quantitatively assessed performance using Mean Squared Error (MSE) and network capacity analysis under various corruption levels.