As well known, the huge memory and compute costs of both artificial neural networks (ANNs) and spiking neural networks (SNNs) greatly hinder their deployment on edge devices with high efficiency. Model compression has been proposed as a promising technique to improve the running efficiency via parameter and operation reduction. Whereas, this technique is mainly practiced in ANNs rather than SNNs. It is interesting to answer how much an SNN model can be compressed without compromising its functionality, where two challenges should be addressed: i) the accuracy of SNNs is usually sensitive to model compression, which requires an accurate compression methodology; ii) the computation of SNNs is event-driven rather than static, which produces an extra compression dimension on dynamic spikes. To this end, we realize a comprehensive SNN compression through three steps. First, we formulate the connection pruning and weight quantization as a constrained optimization problem. Second, we combine spatio-temporal backpropagation (STBP) and alternating direction method of multipliers (ADMM) to solve the problem with minimum accuracy loss. Third, we further propose activity regularization to reduce the spike events for fewer active operations. These methods can be applied in either a single way for moderate compression or a joint way for aggressive compression. We define several quantitative metrics to evaluation the compression performance for SNNs. Our methodology is validated in pattern recognition tasks over MNIST, N-MNIST, CIFAR10, and CIFAR100 datasets, where extensive comparisons, analyses, and insights are provided. To our best knowledge, this is the first work that studies SNN compression in a comprehensive manner by exploiting all compressible components and achieves better results.
Time series data compression is emerging as an important problem with the growth in IoT devices and sensors. Due to the presence of noise in these datasets, lossy compression can often provide significant compression gains without impacting the performance of downstream applications. In this work, we propose an error-bounded lossy compressor, LFZip, for multivariate floating-point time series data that provides guaranteed reconstruction up to user-specified maximum absolute error. The compressor is based on the prediction-quantization-entropy coder framework and benefits from improved prediction using linear models and neural networks. We evaluate the compressor on several time series datasets where it outperforms the existing state-of-the-art error-bounded lossy compressors. The code and data are available at https://github.com/shubhamchandak94/LFZip
We present a framework based on neural networks to extract music scores directly from polyphonic audio in an end-to-end fashion. Most previous Automatic Music Transcription (AMT) methods seek a piano-roll representation of the pitches, that can be further transformed into a score by incorporating tempo estimation, beat tracking, key estimation or rhythm quantization. Unlike these methods, our approach generates music notation directly from the input audio in a single stage. For this, we use a Convolutional Recurrent Neural Network (CRNN) with Connectionist Temporal Classification (CTC) loss function which does not require annotated alignments of audio frames with the score rhythmic information. We trained our model using as input Haydn, Mozart, and Beethoven string quartets and Bach chorales synthesized with different tempos and expressive performances. The output is a textual representation of four-voice music scores based on **kern format. Although the proposed approach is evaluated in a simplified scenario, results show that this model can learn to transcribe scores directly from audio signals, opening a promising avenue towards complete AMT.
Quantizing large Neural Networks (NN) while maintaining the performance is highly desirable for resource-limited devices due to reduced memory and time complexity. It is usually formulated as a constrained optimization problem and optimized via a modified version of gradient descent. In this work, by interpreting the continuous parameters (unconstrained) as the dual of the quantized ones, we introduce a Mirror Descent (MD) framework for NN quantization. Specifically, we provide conditions on the projections (i.e., mapping from continuous to quantized ones) which would enable us to derive valid mirror maps and in turn the respective MD updates. Furthermore, we present a numerically stable implementation of MD that requires storing an additional set of auxiliary variables (unconstrained), and show that it is strikingly analogous to the Straight Through Estimator (STE) based method which is typically viewed as a "trick" to avoid vanishing gradients issue. Our experiments on CIFAR-10/100, TinyImageNet, and ImageNet classification datasets with VGG-16, ResNet-18, and MobileNetV2 architectures show that our MD variants obtain quantized networks with state-of-the-art performance. Code is available at https://github.com/kartikgupta-at-anu/md-bnn.
Recently, pre-trained Transformer based language models such as BERT and GPT, have shown great improvement in many Natural Language Processing (NLP) tasks. However, these models contain a large amount of parameters. The emergence of even larger and more accurate models such as GPT2 and Megatron, suggest a trend of large pre-trained Transformer models. However, using these large models in production environments is a complex task requiring a large amount of compute, memory and power resources. In this work we show how to perform quantization-aware training during the fine-tuning phase of BERT in order to compress BERT by $4\times$ with minimal accuracy loss. Furthermore, the produced quantized model can accelerate inference speed if it is optimized for 8bit Integer supporting hardware.
Low-precision training reduces computational cost and produces efficient models. Recent research in developing new low-precision training algorithms often relies on simulation to empirically evaluate the statistical effects of quantization while avoiding the substantial overhead of building specific hardware. To support this empirical research, we introduce QPyTorch, a low-precision arithmetic simulation framework. Built natively in PyTorch, QPyTorch provides a convenient interface that minimizes the efforts needed to reliably convert existing codes to study low-precision training. QPyTorch is general, and supports a variety of combinations of precisions, number formats, and rounding options. Additionally, it leverages an efficient fused-kernel approach to reduce simulator overhead, which enables simulation of large-scale, realistic problems. QPyTorch is publicly available at https://github.com/Tiiiger/QPyTorch.
The homogeneous transformation between a LiDAR and monocular camera is required for sensor fusion tasks, such as SLAM. While determining such a transformation is not considered glamorous in any sense of the word, it is nonetheless crucial for many modern autonomous systems. Indeed, an error of a few degrees in rotation or a few percent in translation can lead to 20 cm translation errors at a distance of 5 m when overlaying a LiDAR image on a camera image. The biggest impediments to determining the transformation accurately are the relative sparsity of LiDAR point clouds and systematic errors in their distance measurements. This paper proposes (1) the use of targets of known dimension and geometry to ameliorate target pose estimation in face of the quantization and systematic errors inherent in a LiDAR image of a target, and (2) a fitting method for the LiDAR to monocular camera transformation that fundamentally assumes the camera image data is the most accurate information in one's possession.
People learn throughout life. However, incrementally updating conventional neural networks leads to catastrophic forgetting. A common remedy is replay, which is inspired by how the brain consolidates memory. Replay involves fine-tuning a network on a mixture of new and old instances. While there is neuroscientific evidence that the brain replays compressed memories, existing methods for convolutional networks replay raw images. Here, we propose REMIND, a brain-inspired approach that enables efficient replay with compressed representations. REMIND is trained in an online manner, meaning it learns one example at a time, which is closer to how humans learn. Under the same constraints, REMIND outperforms other methods for incremental class learning on the ImageNet ILSVRC-2012 dataset. We probe REMIND's robustness to data ordering schemes known to induce catastrophic forgetting. We demonstrate REMIND's generality by pioneering online learning for Visual Question Answering (VQA).
Deep reinforcement learning continues to show tremendous potential in achieving task-level autonomy, however, its computational and energy demands remain prohibitively high. In this paper, we tackle this problem by applying quantization to reinforcement learning. To that end, we introduce a novel Reinforcement Learning (RL) training paradigm, \textit{ActorQ}, to speed up actor-learner distributed RL training. \textit{ActorQ} leverages 8-bit quantized actors to speed up data collection without affecting learning convergence. Our quantized distributed RL training system, \textit{ActorQ}, demonstrates end-to-end speedups \blue{between 1.5 $\times$ and 5.41$\times$}, and faster convergence over full precision training on a range of tasks (Deepmind Control Suite) and different RL algorithms (D4PG, DQN). Furthermore, we compare the carbon emissions (Kgs of CO2) of \textit{ActorQ} versus standard reinforcement learning \blue{algorithms} on various tasks. Across various settings, we show that \textit{ActorQ} enables more environmentally friendly reinforcement learning by achieving \blue{carbon emission improvements between 1.9$\times$ and 3.76$\times$} compared to training RL-agents in full-precision. We believe that this is the first of many future works on enabling computationally energy-efficient and sustainable reinforcement learning. The source code is available here for the public to use: \url{https://github.com/harvard-edge/QuaRL}.
We propose Additive Powers-of-Two~(APoT) quantization, an efficient non-uniform quantization scheme for the bell-shaped and long-tailed distribution of weights and activations in neural networks. By constraining all quantization levels as the sum of Powers-of-Two terms, APoT quantization enjoys high computational efficiency and a good match with the distribution of weights. A simple reparameterization of the clipping function is applied to generate a better-defined gradient for learning the clipping threshold. Moreover, weight normalization is presented to refine the distribution of weights to make the training more stable and consistent. Experimental results show that our proposed method outperforms state-of-the-art methods, and is even competitive with the full-precision models, demonstrating the effectiveness of our proposed APoT quantization. For example, our 4-bit quantized ResNet-50 on ImageNet achieves 76.6% top-1 accuracy without bells and whistles; meanwhile, our model reduces 22% computational cost compared with the uniformly quantized counterpart. The code is available at https://github.com/yhhhli/APoT_Quantization.
This paper proposes an improved training algorithm for binary neural networks in which both weights and activations are binary numbers. A key but fairly overlooked feature of the current state-of-the-art method of XNOR-Net is the use of analytically calculated real-valued scaling factors for re-weighting the output of binary convolutions. We argue that analytic calculation of these factors is sub-optimal. Instead, in this work, we make the following contributions: (a) we propose to fuse the activation and weight scaling factors into a single one that is learned discriminatively via backpropagation. (b) More importantly, we explore several ways of constructing the shape of the scale factors while keeping the computational budget fixed. (c) We empirically measure the accuracy of our approximations and show that they are significantly more accurate than the analytically calculated one. (d) We show that our approach significantly outperforms XNOR-Net within the same computational budget when tested on the challenging task of ImageNet classification, offering up to 6\% accuracy gain.
Convolutional neural networks (CNNs) have become the dominant neural network architecture for solving visual processing tasks. One of the major obstacles hindering the ubiquitous use of CNNs for inference is their relatively high memory bandwidth requirements, which can be a main energy consumer and throughput bottleneck in hardware accelerators. Accordingly, an efficient feature map compression method can result in substantial performance gains. Inspired by quantization-aware training approaches, we propose a compression-aware training (CAT) method that involves training the model in a way that allows better compression of feature maps during inference. Our method trains the model to achieve low-entropy feature maps, which enables efficient compression at inference time using classical transform coding methods. CAT significantly improves the state-of-the-art results reported for quantization. For example, on ResNet-34 we achieve 73.1% accuracy (0.2% degradation from the baseline) with an average representation of only 1.79 bits per value. Reference implementation accompanies the paper at https://github.com/CAT-teams/CAT
Weight and activation binarization is an effective approach to deep neural network compression and can accelerate the inference by leveraging bitwise operations. Although many binarization methods have improved the accuracy of the model by minimizing the quantization error in forward propagation, there remains a noticeable performance gap between the binarized model and the full-precision one. Our empirical study indicates that the quantization brings information loss in both forward and backward propagation, which is the bottleneck of training accurate binary neural networks. To address these issues, we propose an Information Retention Network (IR-Net) to retain the information that consists in the forward activations and backward gradients. IR-Net mainly relies on two technical contributions: (1) Libra Parameter Binarization (Libra-PB): simultaneously minimizing both quantization error and information loss of parameters by balanced and standardized weights in forward propagation; (2) Error Decay Estimator (EDE): minimizing the information loss of gradients by gradually approximating the sign function in backward propagation, jointly considering the updating ability and accurate gradients. We are the first to investigate both forward and backward processes of binary networks from the unified information perspective, which provides new insight into the mechanism of network binarization. Comprehensive experiments with various network structures on CIFAR-10 and ImageNet datasets manifest that the proposed IR-Net can consistently outperform state-of-the-art quantization methods.
Applications of Fully Convolutional Networks (FCN) in iris segmentation have shown promising advances. For mobile and embedded systems, a significant challenge is that the proposed FCN architectures are extremely computationally demanding. In this article, we propose a resource-efficient, end-to-end iris recognition flow, which consists of FCN-based segmentation, contour fitting, followed by Daugman normalization and encoding. To attain accurate and efficient FCN models, we propose a three-step SW/HW co-design methodology consisting of FCN architectural exploration, precision quantization, and hardware acceleration. In our exploration, we propose multiple FCN models, and in comparison to previous works, our best-performing model requires 50X less FLOPs per inference while achieving a new state-of-the-art segmentation accuracy. Next, we select the most efficient set of models and further reduce their computational complexity through weights and activations quantization using 8-bit dynamic fixed-point (DFP) format. Each model is then incorporated into an end-to-end flow for true recognition performance evaluation. A few of our end-to-end pipelines outperform the previous state-of-the-art on two datasets evaluated. Finally, we propose a novel DFP accelerator and fully demonstrate the SW/HW co-design realization of our flow on an embedded FPGA platform. In comparison with the embedded CPU, our hardware acceleration achieves up to 8.3X speedup for the overall pipeline while using less than 15% of the available FPGA resources. We also provide comparisons between the FPGA system and an embedded GPU showing different benefits and drawbacks for the two platforms.
This paper develops a communication-efficient algorithm to solve the stochastic optimization problem defined over a distributed network, aiming at reducing the burdensome communication in applications such as distributed machine learning.Different from the existing works based on quantization and sparsification, we introduce a communication-censoring technique to reduce the transmissions of variables, which leads to our communication-Censored distributed Stochastic Gradient Descent (CSGD) algorithm. Specifically, in CSGD, the latest mini-batch stochastic gradient at a worker will be transmitted to the server if and only if it is sufficiently informative. When the latest gradient is not available, the stale one will be reused at the server. To implement this communication-censoring strategy, the batch-size is increasing in order to alleviate the effect of stochastic gradient noise. Theoretically, CSGD enjoys the same order of convergence rate as that of SGD, but effectively reduces communication. Numerical experiments demonstrate the sizable communication saving of CSGD.
This paper proposes a new subspace learning method, named Quantized Fisher Discriminant Analysis (QFDA), which makes use of both machine learning and information theory. There is a lack of literature for combination of machine learning and information theory and this paper tries to tackle this gap. QFDA finds a subspace which discriminates the uniformly quantized images in the Discrete Cosine Transform (DCT) domain at least as well as discrimination of non-quantized images by Fisher Discriminant Analysis (FDA) while the images have been compressed. This helps the user to throw away the original images and keep the compressed images instead without noticeable loss of classification accuracy. We propose a cost function whose minimization can be interpreted as rate-distortion optimization in information theory. We also propose quantized Fisherfaces for facial analysis in QFDA. Our experiments on AT&T face dataset and Fashion MNIST dataset show the effectiveness of this subspace learning method.
Compressing word embeddings is important for deploying NLP models in memory-constrained settings. However, understanding what makes compressed embeddings perform well on downstream tasks is challenging---existing measures of compression quality often fail to distinguish between embeddings that perform well and those that do not. We thus propose the eigenspace overlap score as a new measure. We relate the eigenspace overlap score to downstream performance by developing generalization bounds for the compressed embeddings in terms of this score, in the context of linear and logistic regression. We then show that we can lower bound the eigenspace overlap score for a simple uniform quantization compression method, helping to explain the strong empirical performance of this method. Finally, we show that by using the eigenspace overlap score as a selection criterion between embeddings drawn from a representative set we compressed, we can efficiently identify the better performing embedding with up to $2\times$ lower selection error rates than the next best measure of compression quality, and avoid the cost of training a model for each task of interest.
We present PULP-NN, an optimized computing library for a parallel ultra-low-power tightly coupled cluster of RISC-V processors. The key innovation in PULP-NN is a set of kernels for Quantized Neural Network (QNN) inference, targeting byte and sub-byte data types, down to INT-1, tuned for the recent trend toward aggressive quantization in deep neural network inference. The proposed library exploits both the digital signal processing (DSP) extensions available in the PULP RISC-V processors and the cluster's parallelism, achieving up to 15.5 MACs/cycle on INT-8 and improving performance by up to 63x with respect to a sequential implementation on a single RISC-V core implementing the baseline RV32IMC ISA. Using PULP-NN, a CIFAR-10 network on an octa-core cluster runs in 30x and 19.6x less clock cycles than the current state-of-the-art ARM CMSIS-NN library, running on STM32L4 and STM32H7 MCUs, respectively. The proposed library, when running on GAP-8 processor, outperforms by 36.8x and by 7.45x the execution on energy efficient MCUs such as STM32L4 and high-end MCUs such as STM32H7 respectively, when operating at the maximum frequency. The energy efficiency on GAP-8 is 14.1x higher than STM32L4 and 39.5x higher than STM32H7, at the maximum efficiency operating point.
Quantization based techniques are the current state-of-the-art for scaling maximum inner product search to massive databases. Traditional approaches to quantization aim to minimize the reconstruction error of the database points. Based on the observation that for a given query, the database points that have the largest inner products are more relevant, we develop a family of anisotropic quantization loss functions. Under natural statistical assumptions, we show that quantization with these loss functions leads to a new variant of vector quantization that more greatly penalizes the parallel component of a datapoint's residual relative to its orthogonal component. The proposed approach achieves state-of-the-art results on the public benchmarks available at \url{ann-benchmarks.com}.
Embedding layers are commonly used to map discrete symbols into continuous embedding vectors that reflect their semantic meanings. Despite their effectiveness, the number of parameters in an embedding layer increases linearly with the number of symbols and poses a critical challenge on memory and storage constraints. In this work, we propose a generic and end-to-end learnable compression framework termed differentiable product quantization (DPQ). We present two instantiations of DPQ that leverage different approximation techniques to enable differentiability in end-to-end learning. Our method can readily serve as a drop-in alternative for any existing embedding layer. Empirically, DPQ offers significant compression ratios (14-238$\times$) at negligible or no performance cost on 10 datasets across three different language tasks.