Denoising diffusion (score-based) generative models have recently achieved significant accomplishments in generating realistic and diverse data. These approaches define a forward diffusion process for transforming data into noise and a backward denoising process for sampling data from noise. Unfortunately, the generation process of current denoising diffusion models is notoriously slow due to the lengthy iterative noise estimations, which rely on cumbersome neural networks. It prevents the diffusion models from being widely deployed, especially on edge devices. Previous works accelerate the generation process of diffusion model (DM) via finding shorter yet effective sampling trajectories. However, they overlook the cost of noise estimation with a heavy network in every iteration. In this work, we accelerate generation from the perspective of compressing the noise estimation network. Due to the difficulty of retraining DMs, we exclude mainstream training-aware compression paradigms and introduce post-training quantization (PTQ) into DM acceleration. However, the output distributions of noise estimation networks change with time-step, making previous PTQ methods fail in DMs since they are designed for single-time step scenarios. To devise a DM-specific PTQ method, we explore PTQ on DM in three aspects: quantized operations, calibration dataset, and calibration metric. We summarize and use several observations derived from all-inclusive investigations to formulate our method, which especially targets the unique multi-time-step structure of DMs. Experimentally, our method can directly quantize full-precision DMs into 8-bit models while maintaining or even improving their performance in a training-free manner. Importantly, our method can serve as a plug-and-play module on other fast-sampling methods, e.g., DDIM. The code is available at https://github.com/42Shawn/PTQ4DM .
Approximating radiance fields with volumetric grids is one of promising directions for improving NeRF, represented by methods like Plenoxels and DVGO, which achieve super-fast training convergence and real-time rendering. However, these methods typically require a tremendous storage overhead, costing up to hundreds of megabytes of disk space and runtime memory for a single scene. We address this issue in this paper by introducing a simple yet effective framework, called vector quantized radiance fields (VQRF), for compressing these volume-grid-based radiance fields. We first present a robust and adaptive metric for estimating redundancy in grid models and performing voxel pruning by better exploring intermediate outputs of volumetric rendering. A trainable vector quantization is further proposed to improve the compactness of grid models. In combination with an efficient joint tuning strategy and post-processing, our method can achieve a compression ratio of 100$\times$ by reducing the overall model size to 1 MB with negligible loss on visual quality. Extensive experiments demonstrate that the proposed framework is capable of achieving unrivaled performance and well generalization across multiple methods with distinct volumetric structures, facilitating the wide use of volumetric radiance fields methods in real-world applications. Code Available at \url{https://github.com/AlgoHunt/VQRF}
The complicated architecture and high training cost of vision transformers urge the exploration of post-training quantization. However, the heavy-tailed distribution of vision transformer activations hinders the effectiveness of previous post-training quantization methods, even with advanced quantizer designs. Instead of tuning the quantizer to better fit the complicated activation distribution, this paper proposes NoisyQuant, a quantizer-agnostic enhancement for the post-training activation quantization performance of vision transformers. We make a surprising theoretical discovery that for a given quantizer, adding a fixed Uniform noisy bias to the values being quantized can significantly reduce the quantization error under provable conditions. Building on the theoretical insight, NoisyQuant achieves the first success on actively altering the heavy-tailed activation distribution with additive noisy bias to fit a given quantizer. Extensive experiments show NoisyQuant largely improves the post-training quantization performance of vision transformer with minimal computation overhead. For instance, on linear uniform 6-bit activation quantization, NoisyQuant improves SOTA top-1 accuracy on ImageNet by up to 1.7%, 1.1% and 0.5% for ViT, DeiT, and Swin Transformer respectively, achieving on-par or even higher performance than previous nonlinear, mixed-precision quantization.
Generating controllable and editable human motion sequences is a key challenge in 3D Avatar generation. It has been labor-intensive to generate and animate human motion for a long time until learning-based approaches have been developed and applied recently. However, these approaches are still task-specific or modality-specific\cite {ahuja2019language2pose}\cite{ghosh2021synthesis}\cite{ferreira2021learning}\cite{li2021ai}. In this paper, we propose ``UDE", the first unified driving engine that enables generating human motion sequences from natural language or audio sequences (see Fig.~\ref{fig:teaser}). Specifically, UDE consists of the following key components: 1) a motion quantization module based on VQVAE that represents continuous motion sequence as discrete latent code\cite{van2017neural}, 2) a modality-agnostic transformer encoder\cite{vaswani2017attention} that learns to map modality-aware driving signals to a joint space, and 3) a unified token transformer (GPT-like\cite{radford2019language}) network to predict the quantized latent code index in an auto-regressive manner. 4) a diffusion motion decoder that takes as input the motion tokens and decodes them into motion sequences with high diversity. We evaluate our method on HumanML3D\cite{Guo_2022_CVPR} and AIST++\cite{li2021learn} benchmarks, and the experiment results demonstrate our method achieves state-of-the-art performance. Project website: \url{https://github.com/zixiangzhou916/UDE/
This paper describes an entropy regularization term for vector quantization (VQ) based on the analysis of persistent homology of the VQ embeddings. Higher embedding entropy positively correlates with higher codebook utilization, mitigating overfit towards the identity and codebook collapse in VQ-based autoencoders [1]. We show that homology-constrained regularization is an effective way to increase entropy of the VQ process (approximated to input entropy) while preserving the approximated topology in the quantized latent space, averaged over mini batches. This work further explores some patterns of persistent homology diagrams of latents formed by vector quantization. We implement and test the proposed algorithm as a module integrated into a sample VQ-VAE. Linked code repository provides a functioning implementation of the proposed architecture, referred to as homology-constrained vector quantization (HC-VQ) further in this work.
In deep neural networks (DNNs), there are a huge number of weights and multiply-and-accumulate (MAC) operations. Accordingly, it is challenging to apply DNNs on resource-constrained platforms, e.g., mobile phones. Quantization is a method to reduce the size and the computational complexity of DNNs. Existing quantization methods either require hardware overhead to achieve a non-uniform quantization or focus on model-wise and layer-wise uniform quantization, which are not as fine-grained as filter-wise quantization. In this paper, we propose a class-based quantization method to determine the minimum number of quantization bits for each filter or neuron in DNNs individually. In the proposed method, the importance score of each filter or neuron with respect to the number of classes in the dataset is first evaluated. The larger the score is, the more important the filter or neuron is and thus the larger the number of quantization bits should be. Afterwards, a search algorithm is adopted to exploit the different importance of filters and neurons to determine the number of quantization bits of each filter or neuron. Experimental results demonstrate that the proposed method can maintain the inference accuracy with low bit-width quantization. Given the same number of quantization bits, the proposed method can also achieve a better inference accuracy than the existing methods.
Existing Temporal Action Detection (TAD) methods typically take a pre-processing step in converting an input varying-length video into a fixed-length snippet representation sequence, before temporal boundary estimation and action classification. This pre-processing step would temporally downsample the video, reducing the inference resolution and hampering the detection performance in the original temporal resolution. In essence, this is due to a temporal quantization error introduced during the resolution downsampling and recovery. This could negatively impact the TAD performance, but is largely ignored by existing methods. To address this problem, in this work we introduce a novel model-agnostic post-processing method without model redesign and retraining. Specifically, we model the start and end points of action instances with a Gaussian distribution for enabling temporal boundary inference at a sub-snippet level. We further introduce an efficient Taylor-expansion based approximation, dubbed as Gaussian Approximated Post-processing (GAP). Extensive experiments demonstrate that our GAP can consistently improve a wide variety of pre-trained off-the-shelf TAD models on the challenging ActivityNet (+0.2% -0.7% in average mAP) and THUMOS (+0.2% -0.5% in average mAP) benchmarks. Such performance gains are already significant and highly comparable to those achieved by novel model designs. Also, GAP can be integrated with model training for further performance gain. Importantly, GAP enables lower temporal resolutions for more efficient inference, facilitating low-resource applications. The code will be available in https://github.com/sauradip/GAP
A novel control design framework is proposed for a class of non-Lipschitz nonlinear systems with quantized states, meanwhile prescribed transient performance and lower control design complexity could be guaranteed. Firstly, different from all existing control methods for systems with state quantization, global stability of strict-feedback nonlinear systems is achieved without requiring the condition that the nonlinearities of the system model satisfy global Lipschitz continuity. Secondly, a novel barrier function-free prescribed performance control (BFPPC) method is proposed, which can guarantee prescribed transient performance under quantized states. Thirdly, a new \textit{W}-function-based control scheme is designed such that virtual control signals are not required to be differentiated repeatedly and the controller could be designed in a simple way, which guarantees global stability and lower design complexity compared with traditional dynamic surface control (DSC). Simulation results demonstrate the effectiveness of our method.
We examine the problem of uplink cell-free access point (AP) placement in the context of optimal throughput. In this regard, we formulate two main placement problems, namely the sum rate and minimum rate maximization problems, and discuss the challenges associated with solving the underlying optimization problem with the help of some simple scenarios. As a practical solution to the AP placement problem, we suggest a vector quantization (VQ) approach. The suitability of the VQ approach to cell-free AP placement is investigated by examining three VQ-based solutions. First, the standard VQ approach, that is the Lloyd algorithm (using the squared error distortion function) is described. Second, the tree-structured VQ (TSVQ), which performs successive partitioning of the distribution space is applied. Third, a probability density function optimized VQ (PDFVQ) procedure is outlined, which enables efficient, low complexity, and scalable placement, and is aimed at a massive distributed multiple-input-multiple-output (MIMO) scenario. While the VQ-based solutions do not solve the cell-free AP placement problems explicitly, numerical experiments show that their sum and minimum rate performances are good enough, and offer a good starting point for gradient-based optimization methods. Among the VQ solutions, PDFVQ, with advantages over the other VQ methods, offers a good trade-off between sum and minimum rates.
We consider the general problem of recovering a high-dimensional signal from noisy quantized measurements. Quantization, especially coarse quantization such as 1-bit sign measurements, leads to severe information loss and thus a good prior knowledge of the unknown signal is helpful for accurate recovery. Motivated by the power of score-based generative models (SGM, also known as diffusion models) in capturing the rich structure of natural signals beyond simple sparsity, we propose an unsupervised data-driven approach called quantized compressed sensing with SGM (QCS-SGM), where the prior distribution is modeled by a pre-trained SGM. To perform posterior sampling, an annealed pseudo-likelihood score called noise perturbed pseudo-likelihood score is introduced and combined with the prior score of SGM. The proposed QCS-SGM applies to an arbitrary number of quantization bits. Experiments on a variety of baseline datasets demonstrate that the proposed QCS-SGM significantly outperforms existing state-of-the-art algorithms by a large margin for both in-distribution and out-of-distribution samples. Moreover, as a posterior sampling method, QCS-SGM can be easily used to obtain confidence intervals or uncertainty estimates of the reconstructed results. The code is available at https://github.com/mengxiangming/QCS-SGM.
Closed-loop sleep modulation is an emerging research paradigm to treat sleep disorders and enhance sleep benefits. However, two major barriers hinder the widespread application of this research paradigm. First, subjects often need to be wire-connected to rack-mount instrumentation for data acquisition, which negatively affects sleep quality. Second, conventional real-time sleep stage classification algorithms give limited performance. In this work, we conquer these two limitations by developing a sleep modulation system that supports closed-loop operations on the device. Sleep stage classification is performed using a lightweight deep learning (DL) model accelerated by a low-power field-programmable gate array (FPGA) device. The DL model uses a single channel electroencephalogram (EEG) as input. Two convolutional neural networks (CNNs) are used to capture general and detailed features, and a bidirectional long-short-term memory (LSTM) network is used to capture time-variant sequence features. An 8-bit quantization is used to reduce the computational cost without compromising performance. The DL model has been validated using a public sleep database containing 81 subjects, achieving a state-of-the-art classification accuracy of 85.8% and a F1-score of 79%. The developed model has also shown the potential to be generalized to different channels and input data lengths. Closed-loop in-phase auditory stimulation has been demonstrated on the test bench.
The wireless in-band full-duplex (IBFD) technology can in theory double the system capacity over the conventional frequency division duplex (FDD) or time-division duplex (TDD) alternatives. But the strong self-interference of the IBFD can cause excessive quantization noise in the analog-to-digital converters (ADC), which represents the hurdle for its real implementation. In this paper, we consider employing a reconfigurable intelligent surface (RIS) for IBFD communications. While the BS transmits and receives the signals to and from the users simultaneously on the same frequency band, it can adjust the reflection coefficients of the RIS to configure the wireless channel so that the self-interference of the BS is sufficiently mitigated in the propagation domain. Taking the impact of the quantization noise into account, we propose to jointly design the downlink (DL) precoding matrix and the RIS coefficients to maximize the sum of uplink (UL) and DL rates. The effectiveness of the proposed RIS-assisted in-band full-duplex (RAIBFD) system is verified by simulation studies, even taking into considerations that the phases of the RIS have only finite resolution.
Binary neural networks are the extreme case of network quantization, which has long been thought of as a potential edge machine learning solution. However, the significant accuracy gap to the full-precision counterparts restricts their creative potential for mobile applications. In this work, we revisit the potential of binary neural networks and focus on a compelling but unanswered problem: how can a binary neural network achieve the crucial accuracy level (e.g., 80%) on ILSVRC-2012 ImageNet? We achieve this goal by enhancing the optimization process from three complementary perspectives: (1) We design a novel binary architecture BNext based on a comprehensive study of binary architectures and their optimization process. (2) We propose a novel knowledge-distillation technique to alleviate the counter-intuitive overfitting problem observed when attempting to train extremely accurate binary models. (3) We analyze the data augmentation pipeline for binary networks and modernize it with up-to-date techniques from full-precision models. The evaluation results on ImageNet show that BNext, for the first time, pushes the binary model accuracy boundary to 80.57% and significantly outperforms all the existing binary networks. Code and trained models are available at: https://github.com/hpi-xnor/BNext.git.
JPEG is arguably the most popular image coding format, achieving high compression ratios via lossy quantization that may create visual artifacts degradation. Numerous attempts to remove these artifacts were conceived over the years, and common to most of these is the use of deterministic post-processing algorithms that optimize some distortion measure (e.g., PSNR, SSIM). In this paper we propose a different paradigm for JPEG artifact correction: Our method is stochastic, and the objective we target is high perceptual quality -- striving to obtain sharp, detailed and visually pleasing reconstructed images, while being consistent with the compressed input. These goals are achieved by training a stochastic conditional generator (conditioned on the compressed input), accompanied by a theoretically well-founded loss term, resulting in a sampler from the posterior distribution. Our solution offers a diverse set of plausible and fast reconstructions for a given input with perfect consistency. We demonstrate our scheme's unique properties and its superiority to a variety of alternative methods on the FFHQ and ImageNet datasets.
Fiber nonlinearity effects cap achievable rates and ranges in long-haul optical fiber communication links. Conventional nonlinearity compensation methods, such as perturbation theory-based nonlinearity compensation (PB-NLC), attempt to compensate for the nonlinearity by approximating analytical solutions to the signal propagation over optical fibers. However, their practical usability is limited by model mismatch and the immense computational complexity associated with the analytical computation of perturbation triplets and the nonlinearity distortion field. Recently, machine learning techniques have been used to optimise parameters of PB-based approaches, which traditionally have been determined analytically from physical models. It has been claimed in the literature that the learned PB-NLC approaches have improved performance and/or reduced computational complexity over their non-learned counterparts. In this paper, we first revisit the acclaimed benefits of the learned PB-NLC approaches by carefully carrying out a comprehensive performance-complexity analysis utilizing state-of-the-art complexity reduction methods. Interestingly, our results show that least squares-based PB-NLC with clustering quantization has the best performance-complexity trade-off among the learned PB-NLC approaches. Second, we advance the state-of-the-art of learned PB-NLC by proposing and designing a fully learned structure. We apply a bi-directional recurrent neural network for learning perturbation triplets that are alike those obtained from the analytical computation and are used as input features for the neural network to estimate the nonlinearity distortion field. Finally, we demonstrate through numerical simulations that our proposed fully learned approach achieves an improved performance-complexity trade-off compared to the existing learned and non-learned PB-NLC techniques.
Vector Quantization (VQ) is an appealing model compression method to obtain a tiny model with less accuracy loss. While methods to obtain better codebooks and codes under fixed clustering dimensionality have been extensively studied, optimizations of the vectors in favour of clustering performance are not carefully considered, especially via the reduction of vector dimensionality. This paper reports our recent progress on the combination of dimensionality compression and vector quantization, proposing a Low-Rank Representation Vector Quantization ($\text{LR}^2\text{VQ}$) method that outperforms previous VQ algorithms in various tasks and architectures. $\text{LR}^2\text{VQ}$ joins low-rank representation with subvector clustering to construct a new kind of building block that is directly optimized through end-to-end training over the task loss. Our proposed design pattern introduces three hyper-parameters, the number of clusters $k$, the size of subvectors $m$ and the clustering dimensionality $\tilde{d}$. In our method, the compression ratio could be directly controlled by $m$, and the final accuracy is solely determined by $\tilde{d}$. We recognize $\tilde{d}$ as a trade-off between low-rank approximation error and clustering error and carry out both theoretical analysis and experimental observations that empower the estimation of the proper $\tilde{d}$ before fine-tunning. With a proper $\tilde{d}$, we evaluate $\text{LR}^2\text{VQ}$ with ResNet-18/ResNet-50 on ImageNet classification datasets, achieving 2.8\%/1.0\% top-1 accuracy improvements over the current state-of-the-art VQ-based compression algorithms with 43$\times$/31$\times$ compression factor.
Knowledge distillation (KD) has been a ubiquitous method for model compression to strengthen the capability of a lightweight model with the transferred knowledge from the teacher. In particular, KD has been employed in quantization-aware training (QAT) of Transformer encoders like BERT to improve the accuracy of the student model with the reduced-precision weight parameters. However, little is understood about which of the various KD approaches best fits the QAT of Transformers. In this work, we provide an in-depth analysis of the mechanism of KD on attention recovery of quantized large Transformers. In particular, we reveal that the previously adopted MSE loss on the attention score is insufficient for recovering the self-attention information. Therefore, we propose two KD methods; attention-map and attention-output losses. Furthermore, we explore the unification of both losses to address task-dependent preference between attention-map and output losses. The experimental results on various Transformer encoder models demonstrate that the proposed KD methods achieve state-of-the-art accuracy for QAT with sub-2-bit weight quantization.
Low-resolution analog-to-digital converters (ADCs) are promising for reducing energy consumption and costs of multiuser multiple-input multiple-output (MIMO) systems with many antennas. We propose low-resolution multiuser MIMO receivers where the signals are simultaneously processed by 1-bit ADCs and a comparator network, which can be interpreted as additional virtual channels with binary outputs. We distinguish the proposed comparator networks in fully and partially connected. For such receivers, we develop the low-resolution aware linear minimum mean-squared error (LRA-LMMSE) channel estimator and detector according to the Bussgang theorem. We also develop a robust detector which takes into account the channel state information (CSI) mismatch statistics. By exploiting knowledge of the channel coefficients we devise a mean-square error (MSE) greedy search and a sequential signal-to-interference-plus-noise ratio (SINR) search for optimization of partially connected networks. Numerical results show that a system with extra virtual channels can outperform a system with additional receive antennas, in terms of bit error rate (BER). Furthermore, by employing the proposed channel estimation with its error statistics, we construct a lower bound on the ergodic sum rate for a linear receiver. Simulation results confirm that the proposed approach outperforms the conventional 1-bit MIMO system in terms of BER, MSE and sum rate.
This paper presents a dynamic predictive sampling (DPS) based analog-to-digital converter (ADC) that provides a non-uniform sampling of input analog continuous-time signals. The processing unit generates a dynamic prediction of the input signal using two prior-quantized samplings to compute digital values of an upper threshold and a lower threshold. The digital threshold values are converted to analog thresholds to form a tracking window. A comparator compares the input analog signal with the tracking window to determine if the prediction is successful. A counter records timestamps between the unsuccessful predictions, which are the selected sampling points for quantization. No quantization is performed for successfully predicted sampling points so that the data throughput and power can be saved. The proposed circuits were designed as a 10-bit ADC using 0.18 micro CMOS process sampling at 1 kHz. The results show that the proposed system can achieve a data compression factor of 6.17 and a power saving factor of 31% compared to a Nyquist rate SAR ADC for ECG monitoring.
This paper presents a fully integrated second-order level-crossing sampling data converter for real-time data compression and feature extraction. Compared with level-sampling ADCs which sample at fixed voltage levels, the proposed circuits updates tracking thresholds using linear extrapolation, which forms a second-order level-crossing sampling ADC that has sloped sampling levels. The computing is done digitally and is implemented by modifying the digital control logic of a conventional SAR ADC. The system selects only the turning points in the input waveform for quantization. The output of the proposed data converter consists of both the digital value of the selected sampling points and the timestamp between the selected sampling points. The main advantages are data savings and power savings for the data converter and the following digital signal processing or communication circuits, which are ideal for low-power sensors. The test chip was fabricated using a 180nm CMOS process. The proposed ADC saves 30% compared to a conventional SAR ADC and achieves a compression factor of 6.17 for tracking ECG signals.