Improved Projection Learning for Lower Dimensional Feature Maps
This addresses efficiency problems for on-chip inference in CNNs, but it is incremental as it builds on existing projection learning methods.
The paper tackled the high energy and time costs of moving large feature maps during CNN inference by compressing all feature maps below a specified limit using learned projections with end-to-end finetuning, achieving compression that can be folded into the pre-trained network.
The requirement to repeatedly move large feature maps off- and on-chip during inference with convolutional neural networks (CNNs) imposes high costs in terms of both energy and time. In this work we explore an improved method for compressing all feature maps of pre-trained CNNs to below a specified limit. This is done by means of learned projections trained via end-to-end finetuning, which can then be folded and fused into the pre-trained network. We also introduce a new `ceiling compression' framework in which evaluate such techniques in view of the future goal of performing inference fully on-chip.