The organization of behavior: A neuropsychological theory

DNN training is extremely time-consuming, necessitating efficient multi-accelerator parallelization. Current approaches to parallelizing training primarily use intra-batch parallelization, where a single iteration of training is split over the available workers, but suffer from diminishing returns at higher worker counts. We present PipeDream, a system that adds inter-batch pipelining to intra-batch parallelism to further improve parallel training throughput, helping to better overlap computation with communication and reduce the amount of communication when possible. Unlike traditional pipelining, DNN training is bi-directional, where a forward pass through the computation graph is followed by a backward pass that uses state and intermediate data computed during the forward pass. Naive pipelining can thus result in mismatches in state versions used in the forward and backward passes, or excessive pipeline flushes and lower hardware efficiency. To address these challenges, PipeDream versions model parameters for numerically correct gradient computations, and schedules forward and backward passes of different minibatches concurrently on different workers with minimal pipeline stalls. PipeDream also automatically partitions DNN layers among workers to balance work and minimize communication. Extensive experimentation with a range of DNN tasks, models, and hardware configurations shows that PipeDream trains models to high accuracy up to 5.3X faster than commonly used intra-batch parallelism techniques.

PipeDream: generalized pipeline parallelism for DNN training

Memristors and memristor crossbar arrays have been widely studied for neuromorphic and other in-memory computing applications. To achieve optimal system performance, however, it is essential to integrate memristor crossbars with peripheral and control circuitry. Here, we report a fully functional, hybrid memristor chip in which a passive crossbar array is directly integrated with custom-designed circuits, including a full set of mixed-signal interface blocks and a digital processor for reprogrammable computing. The memristor crossbar array enables online learning and forward and backward vector-matrix operations, while the integrated interface and control circuitry allow mapping of different algorithms on chip. The system supports charge-domain operation to overcome the nonlinear I–V characteristics of memristor devices through pulse width modulation and custom analogue-to-digital converters. The integrated chip offers all the functions required for operational neuromorphic computing hardware. Accordingly, we demonstrate a perceptron network, sparse coding algorithm and principal component analysis with an integrated classification layer using the system. A programmable neuromorphic computing chip based on passive memristor crossbar arrays integrated with analogue and digital components and an on-chip processor enables the implementation of neuromorphic and machine learning algorithms.

A fully integrated reprogrammable memristor–CMOS system for efficient multiply–accumulate operations

We propose a self-supervised spatiotemporal learning technique which leverages the chronological order of videos. Our method can learn the spatiotemporal representation of the video by predicting the order of shuffled clips from the video. The category of the video is not required, which gives our technique the potential to take advantage of infinite unannotated videos. There exist related works which use frames, while compared to frames, clips are more consistent with the video dynamics. Clips can help to reduce the uncertainty of orders and are more appropriate to learn a video representation. The 3D convolutional neural networks are utilized to extract features for clips, and these features are processed to predict the actual order. The learned representations are evaluated via nearest neighbor retrieval experiments. We also use the learned networks as the pre-trained models and finetune them on the action recognition task. Three types of 3D convolutional neural networks are tested in experiments, and we gain large improvements compared to existing self-supervised methods.

/pdf/self-supervised-spatiotemporal-learning-via-video-clip-order-5et6ze0a54.pdf

Self-Supervised Spatiotemporal Learning via Video Clip Order Prediction

Batch normalization (BN) is a technique to normalize activations in intermediate layers of deep neural networks. Its tendency to improve accuracy and speed up training have established BN as a favorite technique in deep learning. Yet, despite its enormous success, there remains little consensus on the exact reason and mechanism behind these improvements. In this paper we take a step towards a better understanding of BN, following an empirical approach. We conduct several experiments, and show that BN primarily enables training with larger learning rates, which is the cause for faster convergence and better generalization. For networks without BN we demonstrate how large gradient updates can result in diverging loss and activations growing uncontrollably with network depth, which limits possible learning rates. BN avoids this problem by constantly correcting activations to be zero-mean and of unit standard deviation, which enables larger gradient steps, yields faster convergence and may help bypass sharp local minima. We further show various ways in which gradients and activations of deep unnormalized networks are ill-behaved. We contrast our results against recent findings in random matrix theory, shedding new light on classical initialization schemes and their consequences.

/pdf/understanding-batch-normalization-2apfudyrup.pdf

Understanding Batch Normalization

Modern deep neural network training is typically based on mini-batch stochastic gradient optimization. While the use of large mini-batches increases the available computational parallelism, small batch training has been shown to provide improved generalization performance and allows a significantly smaller memory footprint, which might also be exploited to improve machine throughput. 
In this paper, we review common assumptions on learning rate scaling and training duration, as a basis for an experimental comparison of test performance for different mini-batch sizes. We adopt a learning rate that corresponds to a constant average weight update per gradient calculation (i.e., per unit cost of computation), and point out that this results in a variance of the weight updates that increases linearly with the mini-batch size $m$. 
The collected experimental results for the CIFAR-10, CIFAR-100 and ImageNet datasets show that increasing the mini-batch size progressively reduces the range of learning rates that provide stable convergence and acceptable test performance. On the other hand, small mini-batch sizes provide more up-to-date gradient calculations, which yields more stable and reliable training. The best performance has been consistently obtained for mini-batch sizes between $m = 2$ and $m = 32$, which contrasts with recent work advocating the use of mini-batch sizes in the thousands.

Revisiting Small Batch Training for Deep Neural Networks

Stochastic Gradient Descent (SGD) has proven to be remarkably effective in optimizing deep neural networks that employ ever-larger numbers of parameters. Yet, improving the efficiency of large-scale optimization remains a vital and highly active area of research. Recent work has shown that deep neural networks can be optimized in randomly-projected subspaces of much smaller dimensionality than their native parameter space. While such training is promising for more efficient and scalable optimization schemes, its practical application is limited by inferior optimization performance. Here, we improve on recent random subspace approaches as follows: Firstly, we show that keeping the random projection fixed throughout training is detrimental to optimization. We propose re-drawing the random subspace at each step, which yields significantly better performance. We realize further improvements by applying independent projections to different parts of the network, making the approximation more efficient as network dimensionality grows. To implement these experiments, we leverage hardware-accelerated pseudo-random number generation to construct the random projections on-demand at every optimization step, allowing us to distribute the computation of independent random directions across multiple workers with shared random seeds. This yields significant reductions in memory and is up to 10 times faster for the workloads in question.

Improving Neural Network Training in Low Dimensional Random Bases

Deep learning models trained on large data sets have been widely successful in both vision and language domains. As state-of-the-art deep learning architectures have continued to grow in parameter count so have the compute budgets and times required to train them, increasing the need for compute-efficient methods that parallelize training. Two common approaches to parallelize the training of deep networks have been data and model parallelism. While useful, data and model parallelism suffer from diminishing returns in terms of compute efficiency for large batch sizes. In this paper, we investigate how to continue scaling compute efficiently beyond the point of diminishing returns for large batches through local parallelism, a framework which parallelizes training of individual layers in deep networks by replacing global backpropagation with truncated layer-wise backpropagation. Local parallelism enables fully asynchronous layer-wise parallelism with a low memory footprint, and requires little communication overhead compared with model parallelism. We show results in both vision and language domains across a diverse set of architectures, and find that local parallelism is particularly effective in the high-compute regime.

/pdf/parallel-training-of-deep-networks-with-local-updates-42e6qsoos0.pdf

Parallel Training of Deep Networks with Local Updates

We investigate the reasons for the performance degradation incurred with batch-independent normalization. We find that the prototypical techniques of layer normalization and instance normalization both induce the appearance of failure modes in the neural network's pre-activations: (i) layer normalization induces a collapse towards channel-wise constant functions; (ii) instance normalization induces a lack of variability in instance statistics, symptomatic of an alteration of the expressivity. To alleviate failure mode (i) without aggravating failure mode (ii), we introduce the technique "Proxy Normalization" that normalizes post-activations using a proxy distribution. When combined with layer normalization or group normalization, this batch-independent normalization emulates batch normalization's behavior and consistently matches or exceeds its performance.

Proxy-Normalizing Activations to Match Batch Normalization while Removing Batch Dependence

Much recent research has been dedicated to improving the efficiency of training and inference for image classification. This effort has commonly focused on explicitly improving theoretical efficiency, often measured as ImageNet validation accuracy per FLOP. These theoretical savings have, however, proven challenging to achieve in practice, particularly on high-performance training accelerators. 
In this work, we focus on improving the practical efficiency of the state-of-the-art EfficientNet models on a new class of accelerator, the Graphcore IPU. We do this by extending this family of models in the following ways: (i) generalising depthwise convolutions to group convolutions; (ii) adding proxy-normalized activations to match batch normalization performance with batch-independent statistics; (iii) reducing compute by lowering the training resolution and inexpensively fine-tuning at higher resolution. We find that these three methods improve the practical efficiency for both training and inference. Our code will be made available online.

Carlo Luschi

Papers

Revisiting Small Batch Training for Deep Neural Networks

Improving Neural Network Training in Low Dimensional Random Bases

Parallel Training of Deep Networks with Local Updates

Proxy-Normalizing Activations to Match Batch Normalization while Removing Batch Dependence

Making EfficientNet More Efficient: Exploring Batch-Independent Normalization, Group Convolutions and Reduced Resolution Training.