ML Paper Challenge Day 24 — Binarized Neural Networks: Training Deep Neural Networks with Weights and Activations Constrained to +1 or -1

Papers with Code - Binarized Neural Networks: Training Deep Neural Networks with Weights and…

Day 24: 2020.05.05
Paper: Binarized Neural Networks: Training Deep Neural Networks with Weights and Activations Constrained to +1 or -1
Category: Model/Deep Learning/Binarized Neural Network

This is an interesting paper, showing the possibility of having a neural network which all the weights and activations are binaries. The idea should have been there quite a while, but the breakthrough here is that they discovered a feasible way to train it and with very low error rate, close to the state of art at that time.

Binarized Neural Network

• Neural networks with binary weights and activations at run-time
• At training-time the binary weights and activations are used for computing the parameters gradients