Timezone: »

 
Poster
Backprop with Approximate Activations for Memory-efficient Network Training
Ayan Chakrabarti · Benjamin Moseley

Thu Dec 12 10:45 AM -- 12:45 PM (PST) @ East Exhibition Hall B + C #156
in Deep Learning -- Efficient Training Methods »

Training convolutional neural network models is memory intensive since back-propagation requires storing activations of all intermediate layers. This presents a practical concern when seeking to deploy very deep architectures in production, especially when models need to be frequently re-trained on updated datasets. In this paper, we propose a new implementation for back-propagation that significantly reduces memory usage, by enabling the use of approximations with negligible computational cost and minimal effect on training performance. The algorithm reuses common buffers to temporarily store full activations and compute the forward pass exactly. It also stores approximate per-layer copies of activations, at significant memory savings, that are used in the backward pass. Compared to simply approximating activations within standard back-propagation, our method limits accumulation of errors across layers. This allows the use of much lower-precision approximations without affecting training accuracy. Experiments on CIFAR-10, CIFAR-100, and ImageNet show that our method yields performance close to exact training, while storing activations compactly with as low as 4-bit precision.

Keywords: Efficient Training Methods · CNN Architectures · Deep Learning · Deep Learning; Deep Learning
[ Paper]  [ Poster] 

Author Information

Ayan Chakrabarti (Washington University in St. Louis)
Benjamin Moseley (Carnegie Mellon University)

More from the Same Authors