Neural Network Compression – Dmitri Puzyrev

Neural networks are growing in size almost exponentially. At the same time, we want to use neural network models in devices like smartphones, tablets, and wearables. To do that, one of the solutions is to compress them.

In this video, we cover commonly used methods to reduce the size of neural networks: weight sharing, pruning, decomposition, distillation, and quantization.

Follow us on social media:
  / serokell  

Learn more about us:
https://serokell.io/
https://github.com/serokell

Contact us:
[email protected]

0:00 Intro
0:19 Neural network sizes
1:44 Neural networks for smartphones
2:25 Deep learning research
3:00 Weight sharing
5:50 Pruning
10:07 Decomposition
11:34 Knowledge distillation
13:35 Quantization
15:07 Summary