Gradient Forward-Propagation for Large-Scale Temporal Video Modelling

Abstract

How can neural networks be trained on large-volume temporal data efficiently? To compute the gradients required to update parameters, backpropagation blocks computations until the forward and backward passes are completed. For temporal signals, this introduces high latency and hinders real-time learning. It also creates a coupling between consecutive layers, which limits model parallelism and increases memory consumption. In this paper, we build upon Sideways, which avoids blocking by propagating approximate gradients forward in time, and we propose mechanisms for temporal integration of information based on different variants of skip connections. We also show how to decouple computation and delegate individual neural modules to different devices, allowing distributed and parallel training. The proposed Skip-Sideways achieves low latency training, model parallelism, and, importantly, is capable of extracting temporal features, leading to more stable training and improved performance on real-world action recognition video datasets such as HMDB51, UCF101, and the large-scale Kinetics-600. Finally, we also show that models trained with Skip-Sideways generate better future frames than Sideways models, and hence they can better utilize motion cues.

Stochastic Backpropagation: A Memory Efficient Strategy for Training Video Models

“…Sparse Network [14] is applied to image recognition models but can only save the memory theoretically. Sideways [35,36] reduce the memory cost by overwriting activations whenever new ones become available but can only be applied to causal models. Regarding video specific methods, a popular paradigm for training temporal action detectors is to build the model upon pre-extracted features for temporal modeling and reasoning ("Freeze Backbone" in Fig.…”