Low precision training can significantly reduce the computational overhead of training deep neural networks (DNNs). Though many such techniques exist, cyclic precision training (CPT), which dynamically adjusts precision throughout training according to a cyclic schedule, achieves particularly impressive improvements in training efficiency, while actually improving DNN performance. Existing CPT implementations take common learning rate schedules (e.g., cyclical cosine schedules) and use them for low precision training without adequate comparisons to alternative scheduling options. We define a diverse suite of CPT schedules and analyze their performance across a variety of DNN training regimes, some of which are unexplored in the low precision training literature (e.g., node classification with graph neural networks). From these experiments, we discover alternative CPT schedules that offer further improvements in training efficiency and model performance, as well as derive a set of best practices for choosing CPT schedules. Going further, we find that a correlation exists between model performance and training cost, and that changing the underlying CPT schedule can control the tradeoff between these two variables. To explain the direct correlation between model performance and training cost, we draw a connection between quantized training and critical learning periods, suggesting that aggressive quantization is a form of learning impairment that can permanently damage model performance.

Decentralized distributed learning is the key to enabling large-scale machine learning (training) on edge devices utilizing private user-generated local data, without relying on the cloud. However, the practical realization of such on-device training is limited by the communication and compute bottleneck. In this paper, we propose and show the convergence of low precision decentralized training that aims to reduce the computational complexity and communication cost of decentralized training. Many feedback-based compression techniques have been proposed in the literature to reduce communication costs. To the best of our knowledge, there is no work that applies and shows compute efficient training techniques such as quantization, pruning, etc., for peer-to-peer decentralized learning setups. Since real-world applications have a significant skew in the data distribution, we design "Range-EvoNorm" as the normalization activation layer which is better suited for low precision training over non-IID data. Moreover, we show that the proposed low precision training can be used in synergy with other communication compression methods decreasing the communication cost further. Our experiments indicate that 8-bit decentralized training has minimal accuracy loss compared to its full precision counterpart even with non-IID data. However, when low precision training is accompanied by communication compression through sparsification we observe a 1-2% drop in accuracy. The proposed low precision decentralized training decreases computational complexity, memory usage, and communication cost by 4x and compute energy by a factor of ~20x, while trading off less than a $1\%$ accuracy for both IID and non-IID data. In particular, with higher skew values, we observe an increase in accuracy (by ~ 0.5%) with low precision training, indicating the regularization effect of the quantization.

Low precision operations can provide scalability, memory savings, portability, and energy efficiency. This paper proposes SWALP, an approach to low precision training that averages low-precision SGD iterates with a modified learning rate schedule. SWALP is easy to implement and can match the performance of full-precision SGD even with all numbers quantized down to 8 bits, including the gradient accumulators. Additionally, we show that SWALP converges arbitrarily close to the optimal solution for quadratic objectives, and to a noise ball asymptotically smaller than low precision SGD in strongly convex settings.

Almost two years after the acquisition by Intel, the deep learning chip architecture from startup Nervana Systems will finally be moving from its codenamed "Lake Crest" status to an actual product. In that time, Nvidia, which owns the deep learning training market by a long shot, has had time to firm up its commitment to this expanding (if not overhyped in terms of overall industry dollar figures) market with new deep learning-tuned GPUs and appliances on the horizon as well as software tweaks to make training at scale more robust. In other words, even with solid technology at a reasonable price point, for Intel to bring Nervana to the fore of the training market–and push its other products for inference at scale along with that current, it will take a herculean effort–one that Intel seems willing to invest in given its aggressive roadmap for the Nervana-based lineup. The difference now is that at least we have some insight into how (and by how much) this architecture differs from GPUs–and where it might carve out a performance advantage and more certainly, a power efficiency one. The Nervana Intel chip will be very similar to the first generation of chips Nervana was set to bring to market pre-acquisition but with the added benefit of more expertise and technology from Intel feeding developments that put the deep learning chip on a yearly cadence schedule, according to Nervana's first non-founder employee four years ago and now head of AI hardware within Intel, Carey Kloss.