AUTOSPARSE: Towards Automated Sparse Training of Deep Neural Networks

Sparse training is emerging as a promising avenue for reducing the computational cost of training neural networks. Several recent studies have proposed pruning methods using learnable thresholds to efficiently explore the non-uniform distribution of sparsity inherent within the models. In this paper, we propose Gradient Annealing (GA), where gradients of masked weights are scaled down in a non-linear manner. GA provides an elegant trade-off between sparsity and accuracy without the need for additional sparsity-inducing regularization. We integrated GA with the latest learnable pruning methods to create an automated sparse training algorithm called AutoSparse, which achieves better accuracy and/or training/inference FLOPS reduction than existing learnable pruning methods for sparse ResNet50 and MobileNetV1 on ImageNet-1K: AutoSparse achieves (2x, 7x) reduction in (training,inference) FLOPS for ResNet50 on ImageNet at 80% sparsity. Finally, AutoSparse outperforms sparse-to-sparse SotA method MEST (uniform sparsity) for 80% sparse ResNet50 with similar accuracy, where MEST uses 12% more training FLOPS and 50% more inference FLOPS.

Tensor Processing Primitives: A Programming Abstraction for Efficiency and Portability in Deep Learning Workloads

During the past decade, novel Deep Learning (DL) algorithms/workloads and hardware have been developed to tackle a wide range of problems. Despite the advances in workload/hardware ecosystems, the programming methodology of DL-systems is stagnant. DL-workloads leverage either highly-optimized, yet platform-specific and inflexible kernels from DL-libraries, or in the case of novel operators, reference implementations are built via DL-framework primitives with underwhelming performance. This work introduces the Tensor Processing Primitives (TPP), a programming abstraction striving for efficient, portable implementation of DL-workloads with high-productivity. TPPs define a compact, yet versatile set of 2D-tensor operators (or a virtual Tensor ISA), which subsequently can be utilized as building-blocks to construct complex operators on high-dimensional tensors. The TPP specification is platform-agnostic, thus code expressed via TPPs is portable, whereas the TPP implementation is highly-optimized and platform-specific. We demonstrate the efficacy of our approach using standalone kernels and end-to-end DL-workloads expressed entirely via TPPs that outperform state-of-the-art implementations on multiple platforms.

K-TanH: Hardware Efficient Activations For Deep Learning

We propose K-TanH, a novel, highly accurate, hardware efficient approximation of popular activation function Tanh for Deep Learning. K-TanH consists of a sequence of parameterized bit/integer operations, such as, masking, shift and add/subtract (no floating point operation needed) where parameters are stored in a very small look-up table (bit-masking step can be eliminated). The design of K-TanH is flexible enough to deal with multiple numerical formats, such as, FP32 and BFloat16. High quality approximations to other activation functions, e.g., Swish and GELU, can be derived from K-TanH. We provide RTL design for K-TanH to demonstrate its area/power/performance efficacy. It is more accurate than existing piecewise approximations for Tanh. For example, K-TanH achieves $\sim 5\times$ speed up and $> 6\times$ reduction in maximum approximation error over software implementation of Hard TanH. Experimental results for low-precision BFloat16 training of language translation model GNMT on WMT16 data sets with approximate Tanh and Sigmoid obtained via K-TanH achieve similar accuracy and convergence as training with exact Tanh and Sigmoid.

A Study of BFLOAT16 for Deep Learning Training

This paper presents the first comprehensive empirical study demonstrating the efficacy of the Brain Floating Point (BFLOAT16) half-precision format for Deep Learning training across image classification, speech recognition, language modeling, generative networks and industrial recommendation systems. BFLOAT16 is attractive for Deep Learning training for two reasons: the range of values it can represent is the same as that of IEEE 754 floating-point format (FP32) and conversion to/from FP32 is simple. Maintaining the same range as FP32 is important to ensure that no hyper-parameter tuning is required for convergence; e.g., IEEE 754 compliant half-precision floating point (FP16) requires hyper-parameter tuning. In this paper, we discuss the flow of tensors and various key operations in mixed precision training, and delve into details of operations, such as the rounding modes for converting FP32 tensors to BFLOAT16. We have implemented a method to emulate BFLOAT16 operations in Tensorflow, Caffe2, IntelCaffe, and Neon for our experiments. Our results show that deep learning training using BFLOAT16 tensors achieves the same state-of-the-art (SOTA) results across domains as FP32 tensors in the same number of iterations and with no changes to hyper-parameters.

Ternary Residual Networks

Sub-8-bit representation of DNNs incur some discernible loss of accuracy despite rigorous (re)training at low-precision. Such loss of accuracy essentially makes them equivalent to a much shallower counterpart, diminishing the power of being deep networks. To address this problem of accuracy drop we introduce the notion of \textit{residual networks} where we add more low-precision edges to sensitive branches of the sub-8-bit network to compensate for the lost accuracy. Further, we present a perturbation theory to identify such sensitive edges. Aided by such an elegant trade-off between accuracy and compute, the 8-2 model (8-bit activations, ternary weights), enhanced by ternary residual edges, turns out to be sophisticated enough to achieve very high accuracy ($\sim 1\%$ drop from our FP-32 baseline), despite $\sim 1.6\times$ reduction in model size, $\sim 26\times$ reduction in number of multiplications, and potentially $\sim 2\times$ power-performance gain comparing to 8-8 representation, on the state-of-the-art deep network ResNet-101 pre-trained on ImageNet dataset. Moreover, depending on the varying accuracy requirements in a dynamic environment, the deployed low-precision model can be upgraded/downgraded on-the-fly by partially enabling/disabling residual connections. For example, disabling the least important residual connections in the above enhanced network, the accuracy drop is $\sim 2\%$ (from FP32), despite $\sim 1.9\times$ reduction in model size, $\sim 32\times$ reduction in number of multiplications, and potentially $\sim 2.3\times$ power-performance gain comparing to 8-8 representation. Finally, all the ternary connections are sparse in nature, and the ternary residual conversion can be done in a resource-constraint setting with no low-precision (re)training.

Ternary Neural Networks with Fine-Grained Quantization

We propose a novel fine-grained quantization (FGQ) method to ternarize pre-trained full precision models, while also constraining activations to 8 and 4-bits. Using this method, we demonstrate a minimal loss in classification accuracy on state-of-the-art topologies without additional training. We provide an improved theoretical formulation that forms the basis for a higher quality solution using FGQ. Our method involves ternarizing the original weight tensor in groups of $N$ weights. Using $N=4$, we achieve Top-1 accuracy within $3.7\%$ and $4.2\%$ of the baseline full precision result for Resnet-101 and Resnet-50 respectively, while eliminating $75\%$ of all multiplications. These results enable a full 8/4-bit inference pipeline, with best-reported accuracy using ternary weights on ImageNet dataset, with a potential of $9\times$ improvement in performance. Also, for smaller networks like AlexNet, FGQ achieves state-of-the-art results. We further study the impact of group size on both performance and accuracy. With a group size of $N=64$, we eliminate $\approx99\%$ of the multiplications; however, this introduces a noticeable drop in accuracy, which necessitates fine tuning the parameters at lower precision. We address this by fine-tuning Resnet-50 with 8-bit activations and ternary weights at $N=64$, improving the Top-1 accuracy to within $4\%$ of the full precision result with $<30\%$ additional training overhead. Our final quantized model can run on a full 8-bit compute pipeline using 2-bit weights and has the potential of up to $15\times$ improvement in performance compared to baseline full-precision models.