Performance analysis of a pipelined backpropagation parallel algorithm

Pipelined Training with Stale Weights in Deep Convolutional Neural Networks

The growth in size and complexity of convolutional neural networks (CNNs) is forcing the partitioning of a network across multiple accelerators during training and pipelining of backpropagation computations over these accelerators. Pipelining results in the use of stale weights. Existing approaches to pipelined training avoid or limit the use of stale weights with techniques that either underutilize accelerators or increase training memory footprint. This paper contributes a pipelined backpropagation scheme that uses stale weights to maximize accelerator utilization and keep memory overhead modest. It explores the impact of stale weights on the statistical efficiency and performance using 4 CNNs (LeNet-5, AlexNet, VGG, and ResNet) and shows that when pipelining is introduced in early layers, training with stale weights converges and results in models with comparable inference accuracies to those resulting from nonpipelined training (a drop in accuracy of 0.4%, 4%, 0.83%, and 1.45% for the 4 networks, respectively). However, when pipelining is deeper in the network, inference accuracies drop significantly (up to 12% for VGG and 8.5% for ResNet-20). The paper also contributes a hybrid training scheme that combines pipelined with nonpipelined training to address this drop. The potential for performance improvement of the proposed scheme is demonstrated with a proof-of-concept pipelined backpropagation implementation in PyTorch on 2 GPUs using ResNet-56/110/224/362, achieving speedups of up to 1.8X over a 1-GPU baseline.

An emergent group mind across a swarm of robots: Collective cognition and distributed sensing via a shared wireless neural network

We pose the “trained-at-runtime heterogeneous swarm response problem,” in which a swarm of robots must do the following three things: (1) Learn to differentiate between multiple classes of environmental feature patterns (where the feature patterns are distributively sensed across all robots in the swarm). (2) Perform the particular collective behavior that is the appropriate response to the feature pattern that the swarm recognizes in the environment at runtime (where a collective behavior is defined by a mapping of robot actions to robots). (3) The data required for both (1) and (2) is uploaded to the swarm after it has been deployed, i.e., also at runtime (the data required for (1) is the specific environmental feature patterns that the swarm should learn to differentiate between, and the data required for (2) is the mapping from feature classes to swarm behaviors). To solve this problem, we propose a new form of emergent distributed neural network that we call an “artificial group mind.” The group mind transforms a robotic swarm into a single meta-computer that can be programmed at runtime. In particular, the swarm-spanning artificial neural network emerges as each robot maintains a slice of neurons and forms wireless neural connections between its neurons and those on nearby robots. The nearby robots are discovered at runtime. Experiments on real swarms containing up to 316 robots demonstrate that the group mind enables collective decision-making based on distributed sensor data, and solves the trained-at-runtime heterogeneous swarm response problem. The group mind is a new tool that can be used to create more complex emergent swarm behaviors. The group mind also enables swarm behaviors to be a function of global patterns observed across the environment—where the patterns are orders of magnitude larger than the robots themselves.

Hardware-Efficient On-line Learning through Pipelined Truncated-Error Backpropagation in Binary-State Networks

Artificial neural networks (ANNs) trained using backpropagation are powerful learning architectures that have achieved state-of-the-art performance in various benchmarks. Significant effort has been devoted to developing custom silicon devices to accelerate inference in ANNs. Accelerating the training phase, however, has attracted relatively little attention. In this paper, we describe a hardware-efficient on-line learning technique for feedforward multi-layer ANNs that is based on pipelined backpropagation. Learning is performed in parallel with inference in the forward pass, removing the need for an explicit backward pass and requiring no extra weight lookup. By using binary state variables in the feedforward network and ternary errors in truncated-error backpropagation, the need for any multiplications in the forward and backward passes is removed, and memory requirements for the pipelining are drastically reduced. Further reduction in addition operations owing to the sparsity in the forward neural and backpropagating error signal paths contributes to highly efficient hardware implementation. For proof-of-concept validation, we demonstrate on-line learning of MNIST handwritten digit classification on a Spartan 6 FPGA interfacing with an external 1Gb DDR2 DRAM, that shows small degradation in test error performance compared to an equivalently sized binary ANN trained off-line using standard back-propagation and exact errors. Our results highlight an attractive synergy between pipelined backpropagation and binary-state networks in substantially reducing computation and memory requirements, making pipelined on-line learning practical in deep networks.