Neural Information Processing Systems http://nips.cc/

Multiplications are responsible for most of the computational cost involved in neural network training and inference. Recent research has thus looked for ways to reduce the cost associated with them. Inspired by Mogami (2020), we replace multiplication with a cheap piecewise affine approximation that is achieved by adding the bit representation of the floating point numbers together as integers. We show that transformers can be trained with the resulting modified matrix multiplications on both vision and language tasks with little to no performance impact, and without changes to the training hyperparameters. We further replace all non-linearities in the networks making them fully and jointly piecewise affine in both inputs and weights. Finally, we show that we can eliminate all multiplications in the entire training process, including operations in the forward pass, backward pass and optimizer update, demonstrating the first successful training of modern neural network architectures in a fully multiplication-free fashion.

Transformer architectures have become very popular yet the original implementation requires O(L2) in serial time and memory as functions of input length L. Recent works proposed various linear self-attention mechanisms, scaling only as O(L) for serial computation. We conduct a thorough complexity analysis of Performers, a class which includes most recent linear Transformer mechanisms. We note a remarkable computational flexibility: the gradient computation can be performed with no approximations using sublinear memory as a function of L (in addition to negligible storage for the input sequence), at a cost of greater time complexity in the parallel setting. In the extreme case, a Performer consumes only O(1) memory, and still requires O(L) time. Due to complete backwardcompatibility, this discovered time-memory tradeoff can be used for fine-tuning on low-memory devices in a decentralized fashion without any server computations.

It witnesses that the collaborative learning (CL) systems often face the performance bottleneck of limited bandwidth, where multiple low-end devices continuously generate data and transmit intermediate features to the cloud for incremental training. To this end, improving the communication efficiency by reducing traffic size is one of the most crucial issues for realistic deployment. Existing systems mostly compress features at pixel level and ignore the characteristics of feature structure, which could be further exploited for more efficient compression. In this paper, we take new insights into implementing scalable CL systems through a hierarchical compression on features, termed Stripe-wise Group Quantization (SGQ). Different from previous unstructured quantization methods, SGQ captures both channel and spatial similarity in pixels, and simultaneously encodes features in these two levels to gain a much higher compression ratio. In particular, we refactor feature structure based on inter-channel similarity and bound the gradient deviation caused by quantization, in forward and backward passes, respectively. Such a double-stage pipeline makes SGQ hold a sublinear convergence order as the vanilla SGD-based optimization. Extensive experiments show that SGQ achieves a higher traffic reduction ratio by up to 15.97 and provides 9.22 image processing speedup over the uniform quantized training, while preserving adequate model accuracy as FP32 does, even using 4-bit quantization. This verifies that SGQ can be applied to a wide spectrum of edge intelligence applications.

It witnesses that the collaborative learning (CL) systems often face the performance bottleneck of limited bandwidth, where multiple low-end devices continuously generate data and transmit intermediate features to the cloud for incremental training. To this end, improving the communication efficiency by reducing traffic size is one of the most crucial issues for realistic deployment. Existing systems mostly compress features at pixel level and ignore the characteristics of feature structure, which could be further exploited for more efficient compression. In this paper, we take new insights into implementing scalable CL systems through a hierarchical compression on features, termed Stripe-wise Group Quantization (SGQ). Different from previous unstructured quantization methods, SGQ captures both channel and spatial similarity in pixels, and simultaneously encodes features in these two levels to gain a much higher compression ratio. In particular, we refactor feature structure based on inter-channel similarity and bound the gradient deviation caused by quantization, in forward and backward passes, respectively. Such a double-stage pipeline makes SGQ hold a sublinear convergence order as the vanilla SGD-based optimization. Extensive experiments show that SGQ achieves a higher traffic reduction ratio by up to 15.97 and provides 9.22 image processing speedup over the uniform quantized training, while preserving adequate model accuracy as FP32 does, even using 4-bit quantization. This verifies that SGQ can be applied to a wide spectrum of edge intelligence applications.

Policy gradient computes a backward pass for every sample, even though the backward pass is expensive and most samples carry little learning value. The Delightful Policy Gradient (DG) provides a forward-pass signal of learning value: \emph{delight}, the product of advantage and surprisal (negative log-probability). We introduce the \emph{Kondo gate}, which compares delight against a compute price and pays for a backward pass only when the sample is worth it, thereby tracing a quality--cost Pareto frontier. In bandits, zero-price gating preserves useful gradient signal while removing perpendicular noise, and delight is a more reliable screening signal than additive combinations of value and surprise. On MNIST and transformer token reversal, the Kondo gate skips most backward passes while retaining nearly all of DG's learning quality, with gains that grow as problems get harder and backward passes become more expensive. Because the gate tolerates approximate delight, a cheap forward pass can screen samples before expensive backpropagation, suggesting a speculative-decoding-for-training paradigm.