Subscribe to the gold package and get unlimited access to Shamra Academy
Register a new userA Study of BFLOAT16 for Deep Learning Training
339
0
0.0
(
0
)
Ask ChatGPT about the research
No Arabic abstract
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.
rate research
Read More
State-of-the-art generic low-precision training algorithms use a mix of 16-bit and 32-bit precision, creating the folklore that 16-bit hardware compute units alone are not enough to maximize model accuracy. As a result, deep learning accelerators are forced to support both 16-bit and 32-bit floating-point units (FPUs), which is more costly than only using 16-bit FPUs for hardware design. We ask: can we train deep learning models only with 16-bit floating-point units, while still matching the model accuracy attained by 32-bit training? Towards this end, we study 16-bit-FPU training on the widely adopted BFloat16 unit. While these units conventionally use nearest rounding to cast output to 16-bit precision, we show that nearest rounding for model weight updates often cancels small updates, which degrades the convergence and model accuracy. Motivated by this, we study two simple techniques well-established in numerical analysis, stochastic rounding and Kahan summation, to remedy the model accuracy degradation in 16-bit-FPU training. We demonstrate that these two techniques can enable up to 7% absolute validation accuracy gain in 16-bit-FPU training. This leads to 0.1% lower to 0.2% higher validation accuracy compared to 32-bit training across seven deep learning applications.
In spite of advances in understanding lazy training, recent work attributes the practical success of deep learning to the rich regime with complex inductive bias. In this paper, we study rich regime training empirically with benchmark datasets, and find that while most parameters are lazy, there is always a small number of active parameters which change quite a bit during training. We show that re-initializing (resetting to their initial random values) the active parameters leads to worse generalization. Further, we show that most of the active parameters are in the bottom layers, close to the input, especially as the networks become wider. Based on such observations, we study static Layer-Wise Sparse (LWS) SGD, which only updates some subsets of layers. We find that only updating the top and bottom layers have good generalization and, as expected, only updating the top layers yields a fast algorithm. Inspired by this, we investigate probabilistic LWS-SGD, which mostly updates the top layers and occasionally updates the full network. We show that probabilistic LWS-SGD matches the generalization performance of vanilla SGD and the back-propagation time can be 2-5 times more efficient.
Learning Rate (LR) is an important hyper-parameter to tune for effective training of deep neural networks (DNNs). Even for the baseline of a constant learning rate, it is non-trivial to choose a good constant value for training a DNN. Dynamic learning rates involve multi-step tuning of LR values at various stages of the training process and offer high accuracy and fast convergence. However, they are much harder to tune. In this paper, we present a comprehensive study of 13 learning rate functions and their associated LR policies by examining their range parameters, step parameters, and value update parameters. We propose a set of metrics for evaluating and selecting LR policies, including the classification confidence, variance, cost, and robustness, and implement them in LRBench, an LR benchmarking system. LRBench can assist end-users and DNN developers to select good LR policies and avoid bad LR policies for training their DNNs. We tested LRBench on Caffe, an open source deep learning framework, to showcase the tuning optimization of LR policies. Evaluated through extensive experiments, we attempt to demystify the tuning of LR policies by identifying good LR policies with effective LR value ranges and step sizes for LR update schedules.
Unsupervised contrastive learning has gained increasing attention in the latest research and has proven to be a powerful method for learning representations from unlabeled data. However, little theoretical analysis was known for this framework. In this paper, we study the optimization of deep unsupervised contrastive learning. We prove that, by applying end-to-end training that simultaneously updates two deep over-parameterized neural networks, one can find an approximate stationary solution for the non-convex contrastive loss. This result is inherently different from the existing over-parameterized analysis in the supervised setting because, in contrast to learning a specific target function, unsupervised contrastive learning tries to encode the unlabeled data distribution into the neural networks, which generally has no optimal solution. Our analysis provides theoretical insights into the practical success of these unsupervised pretraining methods.
Fully quantized training (FQT), which uses low-bitwidth hardware by quantizing the activations, weights, and gradients of a neural network model, is a promising approach to accelerate the training of deep neural networks. One major challenge with FQT is the lack of theoretical understanding, in particular of how gradient quantization impacts convergence properties. In this paper, we address this problem by presenting a statistical framework for analyzing FQT algorithms. We view the quantized gradient of FQT as a stochastic estimator of its full precision counterpart, a procedure known as quantization-aware training (QAT). We show that the FQT gradient is an unbiased estimator of the QAT gradient, and we discuss the impact of gradient quantization on its variance. Inspired by these theoretical results, we develop two novel gradient quantizers, and we show that these have smaller variance than the existing per-tensor quantizer. For training ResNet-50 on ImageNet, our 5-bit block Householder quantizer achieves only 0.5% validation accuracy loss relative to QAT, comparable to the existing INT8 baseline.
suggested questions
Log in to be able to interact and post comments
comments
Fetching comments
Sign in to be able to follow your search criteria
