Abstract:For a long time, the vision community tries to learn the spatio-temporal representation by combining convolutional neural network together with various temporal models, such as the families of Markov chain, optical flow, RNN and temporal convolution. However, these pipelines consume enormous computing resources due to the alternately learning process for spatial and temporal information. One natural question is whether we can embed the temporal information into the spatial one so the information in the two domains can be jointly learned once-only. In this work, we answer this question by presenting a simple yet powerful operator – temporal interlacing network (TIN). Instead of learning the temporal features, TIN fuses the two kinds of information by interlacing spatial representations from the past to the future, and vice versa. A differentiable interlacing target can be learned to control the interlacing process. In this way, a heavy temporal model is replaced by a simple interlacing operator. We theoretically prove that with a learnable interlacing target, TIN performs equivalently to the regularized temporal convolution network (r-TCN), but gains 4% more accuracy with 6x less latency on 6 challenging benchmarks. These results push the state-of-the-art performances of video understanding by a considerable margin. Not surprising, the ensemble model of the proposed TIN won the 1st place in the ICCV19 - Multi Moments in Time challenge. Code is made available to facilitate further research.1

Forward and Inverse Approximation Theory for Linear Temporal Convolutional Networks

Inverse Approximation Theory for Nonlinear Recurrent Neural Networks

Approximation and Optimization Theory for Linear Continuous-Time Recurrent Neural Networks

Approximation Analysis of Convolutional Neural Networks

On the Curse of Memory in Recurrent Neural Networks: Approximation and Optimization Analysis

Approximation analysis of CNNs from a feature extraction view

Approximation Rate of the Transformer Architecture for Sequence Modeling

Efficient and Accurate Approximations of Nonlinear Convolutional Networks

On the Approximation and Complexity of Deep Neural Networks to Invariant Functions

Approximated Bilinear Modules for Temporal Modeling

Approximation Performance Analysis of Recurrent Neural Networks.

Temporal Interlacing Network

On the approximation of bi-Lipschitz maps by invertible neural networks

Almost Sure Convergence of Linear Temporal Difference Learning with Arbitrary Features

Finite-Time Analysis of Adaptive Temporal Difference Learning with Deep Neural Networks

Best Approximation and Inverse Results for Neural Network Operators

A discretization-invariant extension and analysis of some deep operator networks

Approximation of neural network operators based on the Joukowski transformation

Multivariate Jackson-type inequality for a new type neural network approximation