Skip to content

SanJoao/Eyes-RL

Repository files navigation

Eyes RL: Learning to Recognize Digits with a Focus-Based Approach

Description:

Eyes RL introduces a novel reinforcement learning approach to image recognition inspired by human vision. By mimicking the way our eyes selectively focus on areas of interest, this project demonstrates accurate digit recognition using a limited 4x4 input window, challenging the traditional reliance on large inputs and convolutional neural networks (CNNs).

Key Features:

  • Focus-Based Learning: Simulates human-like vision by using a small, moving window to analyze images.
  • Reinforcement Learning Agent: An LSTM-based agent learns optimal movement and zoom actions to maximize recognition accuracy.
  • Efficient Architecture: Achieves competitive performance without the use of CNNs.
  • MNIST Benchmark: Demonstrates the effectiveness of Eyes RL on the MNIST dataset, achieving a 69% accuracy with only a 4x4 input.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published