search

NVIDIA / DIGITS

Deep Learning GPU Training System
https://developer.nvidia.com/digits
BSD 3-Clause "New" or "Revised" License
4.12k stars 1.38k forks source link

Deep compression #1096

Open naranjuelo opened 7 years ago

naranjuelo commented 7 years ago

Hi! I was wondering if there is any possibility in Digits to compress a neural network.

In the paper "Deep compression: Compressing Deep Neural Networks with Pruning, Trained Quantization and Huffman Coding" (https://arxiv.org/abs/1510.00149), they could reduce for example the size of VGG-16 by 49x without affecting its accuracy (that's very interesting in networks that require so much memory).

Any way to remove the connections with weights below a threshold? Thank you very much!

ShervinAr commented 7 years ago

@lukeyeager Hi, any progress in this direction yet?

shoemakerlevy9 commented 6 years ago

I would also be interested in a pruning feature being added. Here is an additional source talking about how pruning can lead to smaller and faster networks without decreasing accuracy substantially. https://jacobgil.github.io/deeplearning/pruning-deep-learning