In this paper, we study the problem of estimating the state of a dynamic state-space system where the output is subject to quantization. We compare some classical approaches and a new development in the literature to obtain the filtering and smoothing distributions of the state conditioned to quantized data. The classical approaches include the Extended Kalman filter/smoother in which we consider an approximation of the quantizer non-linearity based on the arctan function, the quantized Kalman filter/smoother, the Unscented Kalman filter/smoother, and the Sequential Monte Carlo sampling method also called particle filter/smoother. We consider a new approach based on the Gaussian sum filter/smoother where the probability mass function of the quantized data given the state is modeled as an integral equation and approximated using Gauss-Legendre quadrature. The Particle filter is addressed considering some resampling methods used to deal with the degeneracy problem. Also, the sample impoverishment caused by the resampling method is addressed by introducing diversity in the samples set using the Markov Chain Monte Carlo method. In this paper, we discuss the implementation of the aforementioned algorithms and the Particle filter/smoother implementation is studied by using different resampling methods combined with two Markov Chain algorithms. A numerical simulation is presented to analyze the accuracy of the estimation and the computational cost.
Conventional video compression (VC) methods are based on motion compensated transform coding, and the steps of motion estimation, mode and quantization parameter selection, and entropy coding are optimized individually due to the combinatorial nature of the end-to-end optimization problem. Learned VC allows end-to-end rate-distortion (R-D) optimized training of nonlinear transform, motion and entropy model simultaneously. Most works on learned VC consider end-to-end optimization of a sequential video codec based on R-D loss averaged over pairs of successive frames. It is well-known in conventional VC that hierarchical, bi-directional coding outperforms sequential compression because of its ability to use both past and future reference frames. This paper proposes a learned hierarchical bi-directional video codec (LHBDC) that combines the benefits of hierarchical motion-compensated prediction and end-to-end optimization. Experimental results show that we achieve the best R-D results that are reported for learned VC schemes to date in both PSNR and MS-SSIM. Compared to conventional video codecs, the R-D performance of our end-to-end optimized codec outperforms those of both x265 and SVT-HEVC encoders ("veryslow" preset) in PSNR and MS-SSIM as well as HM 16.23 reference software in MS-SSIM. We present ablation studies showing performance gains due to proposed novel tools such as learned masking, flow-field subsampling, and temporal flow vector prediction. The models and instructions to reproduce our results can be found in https://github.com/makinyilmaz/LHBDC/
State-of-the-art neural language models represented by Transformers are becoming increasingly complex and expensive for practical applications. Low-bit deep neural network quantization techniques provides a powerful solution to dramatically reduce their model size. Current low-bit quantization methods are based on uniform precision and fail to account for the varying performance sensitivity at different parts of the system to quantization errors. To this end, novel mixed precision DNN quantization methods are proposed in this paper. The optimal local precision settings are automatically learned using two techniques. The first is based on a quantization sensitivity metric in the form of Hessian trace weighted quantization perturbation. The second is based on mixed precision Transformer architecture search. Alternating direction methods of multipliers (ADMM) are used to efficiently train mixed precision quantized DNN systems. Experiments conducted on Penn Treebank (PTB) and a Switchboard corpus trained LF-MMI TDNN system suggest the proposed mixed precision Transformer quantization techniques achieved model size compression ratios of up to 16 times over the full precision baseline with no recognition performance degradation. When being used to compress a larger full precision Transformer LM with more layers, overall word error rate (WER) reductions up to 1.7% absolute (18% relative) were obtained.
State-of-the-art language models (LMs) represented by long-short term memory recurrent neural networks (LSTM-RNNs) and Transformers are becoming increasingly complex and expensive for practical applications. Low-bit neural network quantization provides a powerful solution to dramatically reduce their model size. Current quantization methods are based on uniform precision and fail to account for the varying performance sensitivity at different parts of LMs to quantization errors. To this end, novel mixed precision neural network LM quantization methods are proposed in this paper. The optimal local precision choices for LSTM-RNN and Transformer based neural LMs are automatically learned using three techniques. The first two approaches are based on quantization sensitivity metrics in the form of either the KL-divergence measured between full precision and quantized LMs, or Hessian trace weighted quantization perturbation that can be approximated efficiently using matrix free techniques. The third approach is based on mixed precision neural architecture search. In order to overcome the difficulty in using gradient descent methods to directly estimate discrete quantized weights, alternating direction methods of multipliers (ADMM) are used to efficiently train quantized LMs. Experiments were conducted on state-of-the-art LF-MMI CNN-TDNN systems featuring speed perturbation, i-Vector and learning hidden unit contribution (LHUC) based speaker adaptation on two tasks: Switchboard telephone speech and AMI meeting transcription. The proposed mixed precision quantization techniques achieved "lossless" quantization on both tasks, by producing model size compression ratios of up to approximately 16 times over the full precision LSTM and Transformer baseline LMs, while incurring no statistically significant word error rate increase.
We introduce an algorithm for computing geodesics on sampled manifolds that relies on simulation of quantum dynamics on a graph embedding of the sampled data. Our approach exploits classic results in semiclassical analysis and the quantum-classical correspondence, and forms a basis for techniques to learn the manifold from which a dataset is sampled, and subsequently for nonlinear dimensionality reduction of high-dimensional datasets. We illustrate the new algorithm with data sampled from model manifolds and also by a clustering demonstration based on COVID-19 mobility data. Finally, our method reveals interesting connections between the discretization provided by data sampling and quantization.
Prosody modeling is important, but still challenging in expressive voice conversion. As prosody is difficult to model, and other factors, e.g., speaker, environment and content, which are entangled with prosody in speech, should be removed in prosody modeling. In this paper, we present IQDubbing to solve this problem for expressive voice conversion. To model prosody, we leverage the recent advances in discrete self-supervised speech representation (DSSR). Specifically, prosody vector is first extracted from pre-trained VQ-Wav2Vec model, where rich prosody information is embedded while most speaker and environment information are removed effectively by quantization. To further filter out the redundant information except prosody, such as content and partial speaker information, we propose two kinds of prosody filters to sample prosody from the prosody vector. Experiments show that IQDubbing is superior to baseline and comparison systems in terms of speech quality while maintaining prosody consistency and speaker similarity.
We propose DrBoost, a dense retrieval ensemble inspired by boosting. DrBoost is trained in stages: each component model is learned sequentially and specialized by focusing only on retrieval mistakes made by the current ensemble. The final representation is the concatenation of the output vectors of all the component models, making it a drop-in replacement for standard dense retrievers at test time. DrBoost enjoys several advantages compared to standard dense retrieval models. It produces representations which are 4x more compact, while delivering comparable retrieval results. It also performs surprisingly well under approximate search with coarse quantization, reducing latency and bandwidth needs by another 4x. In practice, this can make the difference between serving indices from disk versus from memory, paving the way for much cheaper deployments.
The celebrated universal approximation theorems for neural networks roughly state that any reasonable function can be arbitrarily well-approximated by a network whose parameters are appropriately chosen real numbers. This paper examines the approximation capabilities of one-bit neural networks -- those whose nonzero parameters are $\pm a$ for some fixed $a\not=0$. One of our main theorems shows that for any $f\in C^s([0,1]^d)$ with $\|f\|_\infty<1$ and error $\varepsilon$, there is a $f_{NN}$ such that $|f(\boldsymbol{x})-f_{NN}(\boldsymbol{x})|\leq \varepsilon$ for all $\boldsymbol{x}$ away from the boundary of $[0,1]^d$, and $f_{NN}$ is either implementable by a $\{\pm 1\}$ quadratic network with $O(\varepsilon^{-2d/s})$ parameters or a $\{\pm \frac 1 2 \}$ ReLU network with $O(\varepsilon^{-2d/s}\log (1/\varepsilon))$ parameters, as $\varepsilon\to0$. We establish new approximation results for iterated multivariate Bernstein operators, error estimates for noise-shaping quantization on the Bernstein basis, and novel implementation of the Bernstein polynomials by one-bit quadratic and ReLU neural networks.
Information-theoretic measures have been widely adopted in the design of features for learning and decision problems. Inspired by this, we look at the relationship between i) a weak form of information loss in the Shannon sense and ii) the operation loss in the minimum probability of error (MPE) sense when considering a family of lossy continuous representations (features) of a continuous observation. We present several results that shed light on this interplay. Our first result offers a lower bound on a weak form of information loss as a function of its respective operation loss when adopting a discrete lossy representation (quantization) instead of the original raw observation. From this, our main result shows that a specific form of vanishing information loss (a weak notion of asymptotic informational sufficiency) implies a vanishing MPE loss (or asymptotic operational sufficiency) when considering a general family of lossy continuous representations. Our theoretical findings support the observation that the selection of feature representations that attempt to capture informational sufficiency is appropriate for learning, but this selection is a rather conservative design principle if the intended goal is achieving MPE in classification. Supporting this last point, and under some structural conditions, we show that it is possible to adopt an alternative notion of informational sufficiency (strictly weaker than pure sufficiency in the mutual information sense) to achieve operational sufficiency in learning.
Large DNNs with mixed-precision quantization can achieve ultra-high compression while retaining high classification performance. However, because of the challenges in finding an accurate metric that can guide the optimization process, these methods either sacrifice significant performance compared to the 32-bit floating-point (FP-32) baseline or rely on a compute-expensive, iterative training policy that requires the availability of a pre-trained baseline. To address this issue, this paper presents BMPQ, a training method that uses bit gradients to analyze layer sensitivities and yield mixed-precision quantized models. BMPQ requires a single training iteration but does not need a pre-trained baseline. It uses an integer linear program (ILP) to dynamically adjust the precision of layers during training, subject to a fixed hardware budget. To evaluate the efficacy of BMPQ, we conduct extensive experiments with VGG16 and ResNet18 on CIFAR-10, CIFAR-100, and Tiny-ImageNet datasets. Compared to the baseline FP-32 models, BMPQ can yield models that have 15.4x fewer parameter bits with a negligible drop in accuracy. Compared to the SOTA "during training", mixed-precision training scheme, our models are 2.1x, 2.2x, and 2.9x smaller, on CIFAR-10, CIFAR-100, and Tiny-ImageNet, respectively, with an improved accuracy of up to 14.54%.
Conventional methods for the image-text generation tasks mainly tackle the naturally bidirectional generation tasks separately, focusing on designing task-specific frameworks to improve the quality and fidelity of the generated samples. Recently, Vision-Language Pre-training models have greatly improved the performance of the image-to-text generation tasks, but large-scale pre-training models for text-to-image synthesis task are still under-developed. In this paper, we propose ERNIE-ViLG, a unified generative pre-training framework for bidirectional image-text generation with transformer model. Based on the image quantization models, we formulate both image generation and text generation as autoregressive generative tasks conditioned on the text/image input. The bidirectional image-text generative modeling eases the semantic alignments across vision and language. For the text-to-image generation process, we further propose an end-to-end training method to jointly learn the visual sequence generator and the image reconstructor. To explore the landscape of large-scale pre-training for bidirectional text-image generation, we train a 10-billion parameter ERNIE-ViLG model on a large-scale dataset of 145 million (Chinese) image-text pairs which achieves state-of-the-art performance for both text-to-image and image-to-text tasks, obtaining an FID of 7.9 on MS-COCO for text-to-image synthesis and best results on COCO-CN and AIC-ICC for image captioning.
Neural video codecs have demonstrated great potential in video transmission and storage applications. Existing neural hybrid video coding approaches rely on optical flow or Gaussian-scale flow for prediction, which cannot support fine-grained adaptation to diverse motion content. Towards more content-adaptive prediction, we propose a novel cross-scale prediction module that achieves more effective motion compensation. Specifically, on the one hand, we produce a reference feature pyramid as prediction sources and then transmit cross-scale flows that leverage the feature scale to control the precision of prediction. On the other hand, for the first time, a weighted prediction mechanism is introduced even if only a single reference frame is available, which can help synthesize a fine prediction result by transmitting cross-scale weight maps. In addition to the cross-scale prediction module, we further propose a multi-stage quantization strategy, which improves the rate-distortion performance with no extra computational penalty during inference. We show the encouraging performance of our efficient neural video codec (ENVC) on several benchmark datasets. In particular, the proposed ENVC can compete with the latest coding standard H.266/VVC in terms of sRGB PSNR on UVG dataset for the low-latency mode. We also analyze in detail the effectiveness of the cross-scale prediction module in handling various video content, and provide a comprehensive ablation study to analyze those important components. Test code is available at https://github.com/USTC-IMCL/ENVC .
Emerging edge applications require both a fast response latency and complex processing. This is infeasible without expensive hardware that can process complex operations -- such as object detection -- within a short time. Many approach this problem by addressing the complexity of the models -- via model compression, pruning and quantization -- or compressing the input. In this paper, we propose a different perspective when addressing the performance challenges. Croesus is a multi-stage approach to edge-cloud systems that provides the ability to find the balance between accuracy and performance. Croesus consists of two stages (that can be generalized to multiple stages): an initial and a final stage. The initial stage performs the computation in real-time using approximate/best-effort computation at the edge. The final stage performs the full computation at the cloud, and uses the results to correct any errors made at the initial stage. In this paper, we demonstrate the implications of such an approach on a video analytics use-case and show how multi-stage processing yields a better balance between accuracy and performance. Moreover, we study the safety of multi-stage transactions via two proposals: multi-stage serializability (MS-SR) and multi-stage invariant confluence with Apologies (MS-IA).
This work is focused on the pruning of some convolutional neural networks (CNNs) and improving theirs efficiency on graphic processing units (GPU) by using a direct sparse algorithm. The Nvidia deep neural network (cuDnn) library is the most effective implementations of deep learning (DL) algorithms for GPUs. GPUs are the most commonly used accelerators for deep learning computations. One of the most common techniques for improving the efficiency of CNN models is weight pruning and quantization. There are two main types of pruning: structural and non-structural. The first enables much easier acceleration on many type of accelerators, but with this type it is difficult to achieve a sparsity level and accuracy as high as that obtained with the second type. Non-structural pruning with retraining can generate a weight tensors up to 90% or more of sparsity in some deep CNN models. In this article the pruning algorithm is presented which makes it possible to achieve high sparsity levels without accuracy drop. In the next stage the linear and non-linear quantization is adapted for further time and footprint reduction. This paper is an extended of previously published paper concerning effective pruning techniques and present real models pruned with high sparsities and reduced precision which can achieve better performance than the CuDnn library.
Pre-trained language models such as BERT have shown remarkable effectiveness in various natural language processing tasks. However, these models usually contain millions of parameters, which prevents them from practical deployment on resource-constrained devices. Knowledge distillation, Weight pruning, and Quantization are known to be the main directions in model compression. However, compact models obtained through knowledge distillation may suffer from significant accuracy drop even for a relatively small compression ratio. On the other hand, there are only a few quantization attempts that are specifically designed for natural language processing tasks. They suffer from a small compression ratio or a large error rate since manual setting on hyper-parameters is required and fine-grained subgroup-wise quantization is not supported. In this paper, we proposed an automatic mixed-precision quantization framework designed for BERT that can simultaneously conduct quantization and pruning in a subgroup-wise level. Specifically, our proposed method leverages Differentiable Neural Architecture Search to assign scale and precision for parameters in each sub-group automatically, and at the same time pruning out redundant groups of parameters. Extensive evaluations on BERT downstream tasks reveal that our proposed method outperforms baselines by providing the same performance with much smaller model size. We also show the feasibility of obtaining the extremely light-weight model by combining our solution with orthogonal methods such as DistilBERT.
This work considers a challenging Deep Neural Network(DNN) quantization task that seeks to train quantized DNNs without involving any full-precision operations. Most previous quantization approaches are not applicable to this task since they rely on full-precision gradients to update network weights. To fill this gap, in this work we advocate using Evolutionary Algorithms (EAs) to search for the optimal low-bits weights of DNNs. To efficiently solve the induced large-scale discrete problem, we propose a novel EA based on cooperative coevolution that repeatedly groups the network weights based on the confidence in their values and focuses on optimizing the ones with the least confidence. To the best of our knowledge, this is the first work that applies EAs to train quantized DNNs. Experiments show that our approach surpasses previous quantization approaches and can train a 4-bit ResNet-20 on the Cifar-10 dataset with the same test accuracy as its full-precision counterpart.
Quantized neural networks typically require smaller memory footprints and lower computation complexity, which is crucial for efficient deployment. However, quantization inevitably leads to a distribution divergence from the original network, which generally degrades the performance. To tackle this issue, massive efforts have been made, but most existing approaches lack statistical considerations and depend on several manual configurations. In this paper, we present an adaptive-mapping quantization method to learn an optimal latent sub-distribution that is inherent within models and smoothly approximated with a concrete Gaussian Mixture (GM). In particular, the network weights are projected in compliance with the GM-approximated sub-distribution. This sub-distribution evolves along with the weight update in a co-tuning schema guided by the direct task-objective optimization. Sufficient experiments on image classification and object detection over various modern architectures demonstrate the effectiveness, generalization property, and transferability of the proposed method. Besides, an efficient deployment flow for the mobile CPU is developed, achieving up to 7.46$\times$ inference acceleration on an octa-core ARM CPU. Our codes have been publicly released at \url{https://github.com/RunpeiDong/DGMS}.
Quantization for Convolutional Neural Network (CNN) has shown significant progress with the intention of reducing the cost of computation and storage with low-bitwidth data inputs. There are, however, no systematic studies on how an existing full-bitwidth processing unit, such as CPUs and DSPs, can be better utilized to carry out significantly higher computation throughput for convolution under various quantized bitwidths. In this study, we propose HiKonv, a unified solution that maximizes the compute throughput of a given underlying processing unit to process low-bitwidth quantized data inputs through novel bit-wise parallel computation. We establish theoretical performance bounds using a full-bitwidth multiplier for highly parallelized low-bitwidth convolution, and demonstrate new breakthroughs for high-performance computing in this critical domain. For example, a single 32-bit processing unit can deliver 128 binarized convolution operations (multiplications and additions) under one CPU instruction, and a single 27x18 DSP core can deliver eight convolution operations with 4-bit inputs in one cycle. We demonstrate the effectiveness of HiKonv on CPU and FPGA for both convolutional layers or a complete DNN model. For a convolutional layer quantized to 4-bit, HiKonv achieves a 3.17x latency improvement over the baseline implementation using C++ on CPU. Compared to the DAC-SDC 2020 champion model for FPGA, HiKonv achieves a 2.37x throughput improvement and 2.61x DSP efficiency improvement, respectively.
In this paper, we propose a quantized learning equation with a monotone increasing resolution of quantization and stochastic analysis for the proposed algorithm. According to the white noise hypothesis for the quantization error with dense and uniform distribution, we can regard the quantization error as i.i.d.\ white noise. Based on this, we show that the learning equation with monotonically increasing quantization resolution converges weakly as the distribution viewpoint. The analysis of this paper shows that global optimization is possible for a domain that satisfies the Lipschitz condition instead of local convergence properties such as the Hessian constraint of the objective function.
Running large-scale pre-trained language models in computationally constrained environments remains a challenging problem yet to be addressed, while transfer learning from these models has become prevalent in Natural Language Processing tasks. Several solutions, including knowledge distillation, network quantization, or network pruning have been previously proposed; however, these approaches focus mostly on the English language, thus widening the gap when considering low-resource languages. In this work, we introduce three light and fast versions of distilled BERT models for the Romanian language: Distil-BERT-base-ro, Distil-RoBERT-base, and DistilMulti-BERT-base-ro. The first two models resulted from the individual distillation of knowledge from two base versions of Romanian BERTs available in literature, while the last one was obtained by distilling their ensemble. To our knowledge, this is the first attempt to create publicly available Romanian distilled BERT models, which were thoroughly evaluated on five tasks: part-of-speech tagging, named entity recognition, sentiment analysis, semantic textual similarity, and dialect identification. Our experimental results argue that the three distilled models offer performance comparable to their teachers, while being twice as fast on a GPU and ~35% smaller. In addition, we further test the similarity between the predictions of our students versus their teachers by measuring their label and probability loyalty, together with regression loyalty - a new metric introduced in this work.