Learning to Train a Binary Neural Network

09/27/2018
by   Joseph Bethge, et al.
0

Convolutional neural networks have achieved astonishing results in different application areas. Various methods which allow us to use these models on mobile and embedded devices have been proposed. Especially binary neural networks seem to be a promising approach for these devices with low computational power. However, understanding binary neural networks and training accurate models for practical applications remains a challenge. In our work, we focus on increasing our understanding of the training process and making it accessible to everyone. We publish our code and models based on BMXNet for everyone to use. Within this framework, we systematically evaluated different network architectures and hyperparameters to provide useful insights on how to train a binary neural network. Further, we present how we improved accuracy by increasing the number of connections in the network.

READ FULL TEXT

page 13

page 14

research
12/05/2018

Training Competitive Binary Neural Networks from Scratch

Convolutional neural networks have achieved astonishing results in diffe...
research
11/23/2022

Join the High Accuracy Club on ImageNet with A Binary Neural Network Ticket

Binary neural networks are the extreme case of network quantization, whi...
research
11/19/2015

Adjustable Bounded Rectifiers: Towards Deep Binary Representations

Binary representation is desirable for its memory efficiency, computatio...
research
07/30/2021

Dependable Neural Networks Through Redundancy, A Comparison of Redundant Architectures

With edge-AI finding an increasing number of real-world applications, es...
research
03/08/2022

Bayesian Optimisation-Assisted Neural Network Training Technique for Radio Localisation

Radio signal-based (indoor) localisation technique is important for IoT ...
research
01/18/2017

On the Performance of Network Parallel Training in Artificial Neural Networks

Artificial Neural Networks (ANNs) have received increasing attention in ...
research
08/04/2020

Controlling Information Capacity of Binary Neural Network

Despite the growing popularity of deep learning technologies, high memor...

Please sign up or login with your details

Forgot password? Click here to reset