Several current ultra-wide band applications, such as millimeter wave radar and communication systems, require high sampling rates and therefore expensive and energy-hungry analogto-digital converters (ADCs). In applications where cost and power constraints exist, the use of high-precision ADCs is not feasible and the designer must resort to ADCs with coarse quantization. Consequently the interest in the topic of signal parameter estimation from quantized data has increased significantly in recent years. The Cramer-Rao bound (CRB) is an important yardstick in any parameter estimation problem. Indeed it lower bounds the variance of any unbiased parameter estimator. Moreover, the CRB is an achievable limit, for instance it is asymptotically attained by the maximum likelihood estimator (under regularity conditions), and thus it is a useful benchmark to which the accuracy of any parameter estimator can and should be compared. A formula for the CRB for signal parameter estimation from real-valued quantized data has been presented in but its derivation was somewhat sketchy. The said CRB formula has been extended for instance in to complex-valued quantized data, but again its derivation was rather sketchy. The special case of binary (1-bit) ADCs and a signal consisting of one sinusoid has been thoroughly analyzed in . The CRB formula for a binary ADC and a general real-valued signal has been derived.
Transformer architecture has become the fundamental element of the widespread natural language processing~(NLP) models. With the trends of large NLP models, the increasing memory and computation costs hinder their efficient deployment on resource-limited devices. Therefore, transformer quantization attracts wide research interest. Recent work recognizes that structured outliers are the critical bottleneck for quantization performance. However, their proposed methods increase the computation overhead and still leave the outliers there. To fundamentally address this problem, this paper delves into the inherent inducement and importance of the outliers. We discover that $\boldsymbol \gamma$ in LayerNorm (LN) acts as a sinful amplifier for the outliers, and the importance of outliers varies greatly where some outliers provided by a few tokens cover a large area but can be clipped sharply without negative impacts. Motivated by these findings, we propose an outlier suppression framework including two components: Gamma Migration and Token-Wise Clipping. The Gamma Migration migrates the outlier amplifier to subsequent modules in an equivalent transformation, contributing to a more quantization-friendly model without any extra burden. The Token-Wise Clipping takes advantage of the large variance of token range and designs a token-wise coarse-to-fine pipeline, obtaining a clipping range with minimal final quantization loss in an efficient way. This framework effectively suppresses the outliers and can be used in a plug-and-play mode. Extensive experiments prove that our framework surpasses the existing works and, for the first time, pushes the 6-bit post-training BERT quantization to the full-precision (FP) level. Our code is available at https://github.com/wimh966/outlier_suppression.
Most of today's computer vision pipelines are built around deep neural networks, where convolution operations require most of the generally high compute effort. The Winograd convolution algorithm computes convolutions with fewer MACs compared to the standard algorithm, reducing the operation count by a factor of 2.25x for 3x3 convolutions when using the version with 2x2-sized tiles $F_2$. Even though the gain is significant, the Winograd algorithm with larger tile sizes, i.e., $F_4$, offers even more potential in improving throughput and energy efficiency, as it reduces the required MACs by 4x. Unfortunately, the Winograd algorithm with larger tile sizes introduces numerical issues that prevent its use on integer domain-specific accelerators and higher computational overhead to transform input and output data between spatial and Winograd domains. To unlock the full potential of Winograd $F_4$, we propose a novel tap-wise quantization method that overcomes the numerical issues of using larger tiles, enabling integer-only inference. Moreover, we present custom hardware units that process the Winograd transformations in a power- and area-efficient way, and we show how to integrate such custom modules in an industrial-grade, programmable DSA. An extensive experimental evaluation on a large set of state-of-the-art computer vision benchmarks reveals that the tap-wise quantization algorithm makes the quantized Winograd $F_4$ network almost as accurate as the FP32 baseline. The Winograd-enhanced DSA achieves up to 1.85x gain in energy efficiency and up to 1.83x end-to-end speed-up for state-of-the-art segmentation and detection networks.
While research in the field of transformer models has primarily focused on enhancing performance metrics such as accuracy and perplexity, practical applications in industry often necessitate a rigorous consideration of inference latency constraints. Addressing this challenge, we introduce SpeedLimit, a novel Neural Architecture Search (NAS) technique that optimizes accuracy whilst adhering to an upper-bound latency constraint. Our method incorporates 8-bit integer quantization in the search process to outperform the current state-of-the-art technique. Our results underline the feasibility and efficacy of seeking an optimal balance between performance and latency, providing new avenues for deploying state-of-the-art transformer models in latency-sensitive environments.
This work presents a deep learning-based plant disease diagnostic system using images of fruits and leaves. Five state-of-the-art convolutional neural networks (CNN) have been employed for implementing the system. Hitherto model accuracy has been the focus for such applications and model optimization has not been accounted for the model to be applicable to end-user devices. Two model quantization techniques such as float16 and dynamic range quantization have been applied to the five state-of-the-art CNN architectures. The study shows that the quantized GoogleNet model achieved the size of 0.143 MB with an accuracy of 97%, which is the best candidate model considering the size criterion. The EfficientNet model achieved the size of 4.2MB with an accuracy of 99%, which is the best model considering the performance criterion. The source codes are available at https://github.com/CompostieAI/Guava-disease-detection.
A multi-grid multi-block-size vector quantization (MGBVQ) method is proposed for image coding in this work. The fundamental idea of image coding is to remove correlations among pixels before quantization and entropy coding, e.g., the discrete cosine transform (DCT) and intra predictions, adopted by modern image coding standards. We present a new method to remove pixel correlations. First, by decomposing correlations into long- and short-range correlations, we represent long-range correlations in coarser grids due to their smoothness, thus leading to a multi-grid (MG) coding architecture. Second, we show that short-range correlations can be effectively coded by a suite of vector quantizers (VQs). Along this line, we argue the effectiveness of VQs of very large block sizes and present a convenient way to implement them. It is shown by experimental results that MGBVQ offers excellent rate-distortion (RD) performance, which is comparable with existing image coders, at much lower complexity. Besides, it provides a progressive coded bitstream.
Existing advertisements click-through rate (CTR) prediction models are mainly dependent on behavior ID features, which are learned based on the historical user-ad interactions. Nevertheless, behavior ID features relying on historical user behaviors are not feasible to describe new ads without previous interactions with users. To overcome the limitations of behavior ID features in modeling new ads, we exploit the visual content in ads to boost the performance of CTR prediction models. Specifically, we map each ad into a set of visual IDs based on its visual content. These visual IDs are further used for generating the visual embedding for enhancing CTR prediction models. We formulate the learning of visual IDs into a supervised quantization problem. Due to a lack of class labels for commercial images in advertisements, we exploit image textual descriptions as the supervision to optimize the image extractor for generating effective visual IDs. Meanwhile, since the hard quantization is non-differentiable, we soften the quantization operation to make it support the end-to-end network training. After mapping each image into visual IDs, we learn the embedding for each visual ID based on the historical user-ad interactions accumulated in the past. Since the visual ID embedding depends only on the visual content, it generalizes well to new ads. Meanwhile, the visual ID embedding complements the ad behavior ID embedding. Thus, it can considerably boost the performance of the CTR prediction models previously relying on behavior ID features for both new ads and ads that have accumulated rich user behaviors. After incorporating the visual ID embedding in the CTR prediction model of Baidu online advertising, the average CTR of ads improves by 1.46%, and the total charge increases by 1.10%.
Although analog semantic communication systems have received considerable attention in the literature, there is less work on digital semantic communication systems. In this paper, we develop a deep learning (DL)-enabled vector quantized (VQ) semantic communication system for image transmission, named VQ-DeepSC. Specifically, we propose a convolutional neural network (CNN)-based transceiver to extract multi-scale semantic features of images and introduce multi-scale semantic embedding spaces to perform semantic feature quantization, rendering the data compatible with digital communication systems. Furthermore, we employ adversarial training to improve the quality of received images by introducing a PatchGAN discriminator. Experimental results demonstrate that the proposed VQ-DeepSC is more robustness than BPG in digital communication systems and has comparable MS-SSIM performance to the DeepJSCC method.
This paper considers improving wireless communication and computation efficiency in federated learning (FL) via model quantization. In the proposed bitwidth FL scheme, edge devices train and transmit quantized versions of their local FL model parameters to a coordinating server, which aggregates them into a quantized global model and synchronizes the devices. The goal is to jointly determine the bitwidths employed for local FL model quantization and the set of devices participating in FL training at each iteration. We pose this as an optimization problem that aims to minimize the training loss of quantized FL under a per-iteration device sampling budget and delay requirement. However, the formulated problem is difficult to solve without (i) a concrete understanding of how quantization impacts global ML performance and (ii) the ability of the server to construct estimates of this process efficiently. To address the first challenge, we analytically characterize how limited wireless resources and induced quantization errors affect the performance of the proposed FL method. Our results quantify how the improvement of FL training loss between two consecutive iterations depends on the device selection and quantization scheme as well as on several parameters inherent to the model being learned. Then, we show that the FL training process can be described as a Markov decision process and propose a model-based reinforcement learning (RL) method to optimize action selection over iterations. Compared to model-free RL, this model-based RL approach leverages the derived mathematical characterization of the FL training process to discover an effective device selection and quantization scheme without imposing additional device communication overhead. Simulation results show that the proposed FL algorithm can reduce the convergence time.
Volume data is found in many important scientific and engineering applications. Rendering this data for visualization at high quality and interactive rates for demanding applications such as virtual reality is still not easily achievable even using professional-grade hardware. We introduce FoVolNet -- a method to significantly increase the performance of volume data visualization. We develop a cost-effective foveated rendering pipeline that sparsely samples a volume around a focal point and reconstructs the full-frame using a deep neural network. Foveated rendering is a technique that prioritizes rendering computations around the user's focal point. This approach leverages properties of the human visual system, thereby saving computational resources when rendering data in the periphery of the user's field of vision. Our reconstruction network combines direct and kernel prediction methods to produce fast, stable, and perceptually convincing output. With a slim design and the use of quantization, our method outperforms state-of-the-art neural reconstruction techniques in both end-to-end frame times and visual quality. We conduct extensive evaluations of the system's rendering performance, inference speed, and perceptual properties, and we provide comparisons to competing neural image reconstruction techniques. Our test results show that FoVolNet consistently achieves significant time saving over conventional rendering while preserving perceptual quality.
Neural image compression (NIC) has outperformed traditional image codecs in rate-distortion (R-D) performance. However, it usually requires a dedicated encoder-decoder pair for each point on R-D curve, which greatly hinders its practical deployment. While some recent works have enabled bitrate control via conditional coding, they impose strong prior during training and provide limited flexibility. In this paper we propose Code Editing, a highly flexible coding method for NIC based on semi-amortized inference and adaptive quantization. Our work is a new paradigm for variable bitrate NIC. Furthermore, experimental results show that our method surpasses existing variable-rate methods, and achieves ROI coding and multi-distortion trade-off with a single decoder.
The latest industrial inference engines, such as FasterTransformer and TurboTransformers, have verified that half-precision floating point (FP16) and 8-bit integer (INT8) quantization can greatly improve model inference speed. However, the existing INT8 quantization methods are too complicated, and improper usage will lead to model performance damage greatly. In this paper, we develop a toolkit for users to easily quantize their models for inference, in which Self-Adaptive Mixed-Precision (SAMP) is proposed to automatically control quantization rate by a mixed-precision architecture to balance model accuracy and efficiency. Experimental results show that our SAMP toolkit has a higher speedup than PyTorch and FasterTransformer while ensuring the required accuracy. In addition, SAMP is based on a modular design, decoupling the tokenizer, embedding, encoder and target layers, which allows users to handle various downstream tasks and can be seamlessly integrated into PyTorch.
Although two-stage Vector Quantized (VQ) generative models allow for synthesizing high-fidelity and high-resolution images, their quantization operator encodes similar patches within an image into the same index, resulting in a repeated artifact for similar adjacent regions using existing decoder architectures. To address this issue, we propose to incorporate the spatially conditional normalization to modulate the quantized vectors so as to insert spatially variant information to the embedded index maps, encouraging the decoder to generate more photorealistic images. Moreover, we use multichannel quantization to increase the recombination capability of the discrete codes without increasing the cost of model and codebook. Additionally, to generate discrete tokens at the second stage, we adopt a Masked Generative Image Transformer (MaskGIT) to learn an underlying prior distribution in the compressed latent space, which is much faster than the conventional autoregressive model. Experiments on two benchmark datasets demonstrate that our proposed modulated VQGAN is able to greatly improve the reconstructed image quality as well as provide high-fidelity image generation.
Processing-in-memory (PIM), an increasingly studied neuromorphic hardware, promises orders of energy and throughput improvements for deep learning inference. Leveraging the massively parallel and efficient analog computing inside memories, PIM circumvents the bottlenecks of data movements in conventional digital hardware. However, an extra quantization step (i.e. PIM quantization), typically with limited resolution due to hardware constraints, is required to convert the analog computing results into digital domain. Meanwhile, non-ideal effects extensively exist in PIM quantization because of the imperfect analog-to-digital interface, which further compromises the inference accuracy. In this paper, we propose a method for training quantized networks to incorporate PIM quantization, which is ubiquitous to all PIM systems. Specifically, we propose a PIM quantization aware training (PIM-QAT) algorithm, and introduce rescaling techniques during backward and forward propagation by analyzing the training dynamics to facilitate training convergence. We also propose two techniques, namely batch normalization (BN) calibration and adjusted precision training, to suppress the adverse effects of non-ideal linearity and stochastic thermal noise involved in real PIM chips. Our method is validated on three mainstream PIM decomposition schemes, and physically on a prototype chip. Comparing with directly deploying conventionally trained quantized model on PIM systems, which does not take into account this extra quantization step and thus fails, our method provides significant improvement. It also achieves comparable inference accuracy on PIM systems as that of conventionally quantized models on digital hardware, across CIFAR10 and CIFAR100 datasets using various network depths for the most popular network topology.
In this paper, we consider decentralized optimization problems where agents have individual cost functions to minimize subject to subspace constraints that require the minimizers across the network to lie in low-dimensional subspaces. This constrained formulation includes consensus or single-task optimization as special cases, and allows for more general task relatedness models such as multitask smoothness and coupled optimization. In order to cope with communication constraints, we propose and study an adaptive decentralized strategy where the agents employ differential randomized quantizers to compress their estimates before communicating with their neighbors. The analysis shows that, under some general conditions on the quantization noise, and for sufficiently small step-sizes $\mu$, the strategy is stable both in terms of mean-square error and average bit rate: by reducing $\mu$, it is possible to keep the estimation errors small (on the order of $\mu$) without increasing indefinitely the bit rate as $\mu\rightarrow 0$. Simulations illustrate the theoretical findings and the effectiveness of the proposed approach, revealing that decentralized learning is achievable at the expense of only a few bits.
Conventional frequentist FL schemes are known to yield overconfident decisions. Bayesian FL addresses this issue by allowing agents to process and exchange uncertainty information encoded in distributions over the model parameters. However, this comes at the cost of a larger per-iteration communication overhead. This letter investigates whether Bayesian FL can still provide advantages in terms of calibration when constraining communication bandwidth. We present compressed particle-based Bayesian FL protocols for FL and federated "unlearning" that apply quantization and sparsification across multiple particles. The experimental results confirm that the benefits of Bayesian FL are robust to bandwidth constraints.
The exponentially growing model size drives the continued success of deep learning, but it brings prohibitive computation and memory cost. From the algorithm perspective, model sparsification and quantization have been studied to alleviate the problem. From the architecture perspective, hardware vendors provide Tensor cores for acceleration. However, it is very challenging to gain practical speedups from sparse, low-precision matrix operations on Tensor cores, because of the strict requirements for data layout and lack of support for efficiently manipulating the low-precision integers. We propose Magicube, a high-performance sparse-matrix library for low-precision integers on Tensor cores. Magicube supports SpMM and SDDMM, two major sparse operations in deep learning with mixed precision. Experimental results on an NVIDIA A100 GPU show that Magicube achieves on average 1.44x (up to 2.37x) speedup over the vendor-optimized library for sparse kernels, and 1.43x speedup over the state-of-the-art with a comparable accuracy for end-to-end sparse Transformer inference.
Quantization is wildly taken as a model compression technique, which obtains efficient models by converting floating-point weights and activations in the neural network into lower-bit integers. Quantization has been proven to work well on convolutional neural networks and transformer-based models. Despite the decency of these models, recent works have shown that MLP-based models are able to achieve comparable results on various tasks ranging from computer vision, NLP to 3D point cloud, while achieving higher throughput due to the parallelism and network simplicity. However, as we show in the paper, directly applying quantization to MLP-based models will lead to significant accuracy degradation. Based on our analysis, two major issues account for the accuracy gap: 1) the range of activations in MLP-based models can be too large to quantize, and 2) specific components in the MLP-based models are sensitive to quantization. Consequently, we propose to 1) apply LayerNorm to control the quantization range of activations, 2) utilize bounded activation functions, 3) apply percentile quantization on activations, 4) use our improved module named multiple token-mixing MLPs, and 5) apply linear asymmetric quantizer for sensitive operations. Equipped with the abovementioned techniques, our Q-MLP models can achieve 79.68% accuracy on ImageNet with 8-bit uniform quantization (model size 30 MB) and 78.47% with 4-bit quantization (15 MB).
Data-free quantization can potentially address data privacy and security concerns in model compression, and thus has been widely investigated. Recently, PSAQ-ViT designs a relative value metric, patch similarity, to generate data from pre-trained vision transformers (ViTs), achieving the first attempt at data-free quantization for ViTs. In this paper, we propose PSAQ-ViT V2, a more accurate and general data-free quantization framework for ViTs, built on top of PSAQ-ViT. More specifically, following the patch similarity metric in PSAQ-ViT, we introduce an adaptive teacher-student strategy, which facilitates the constant cyclic evolution of the generated samples and the quantized model (student) in a competitive and interactive fashion under the supervision of the full-precision model (teacher), thus significantly improving the accuracy of the quantized model. Moreover, without the auxiliary category guidance, we employ the task- and model-independent prior information, making the general-purpose scheme compatible with a broad range of vision tasks and models. Extensive experiments are conducted on various models on image classification, object detection, and semantic segmentation tasks, and PSAQ-ViT V2, with the naive quantization strategy and without access to real-world data, consistently achieves competitive results, showing potential as a powerful baseline on data-free quantization for ViTs. For instance, with Swin-S as the (backbone) model, 8-bit quantization reaches 82.13 top-1 accuracy on ImageNet, 50.9 box AP and 44.1 mask AP on COCO, and 47.2 mIoU on ADE20K. We hope that accurate and general PSAQ-ViT V2 can serve as a potential and practice solution in real-world applications involving sensitive data. Code is released and merged at: https://github.com/zkkli/PSAQ-ViT.
We present SeRP, a framework for Self-Supervised Learning of 3D point clouds. SeRP consists of encoder-decoder architecture that takes perturbed or corrupted point clouds as inputs and aims to reconstruct the original point cloud without corruption. The encoder learns the high-level latent representations of the points clouds in a low-dimensional subspace and recovers the original structure. In this work, we have used Transformers and PointNet-based Autoencoders. The proposed framework also addresses some of the limitations of Transformers-based Masked Autoencoders which are prone to leakage of location information and uneven information density. We trained our models on the complete ShapeNet dataset and evaluated them on ModelNet40 as a downstream classification task. We have shown that the pretrained models achieved 0.5-1% higher classification accuracies than the networks trained from scratch. Furthermore, we also proposed VASP: Vector-Quantized Autoencoder for Self-supervised Representation Learning for Point Clouds that employs Vector-Quantization for discrete representation learning for Transformer-based autoencoders.