In this paper, we propose an accurate data-free post-training quantization framework of diffusion models (ADP-DM) for efficient image generation. Conventional data-free quantization methods learn shared quantization functions for tensor discretization regardless of the generation timesteps, while the activation distribution differs significantly across various timesteps. The calibration images are acquired in random timesteps which fail to provide sufficient information for generalizable quantization function learning. Both issues cause sizable quantization errors with obvious image generation performance degradation. On the contrary, we design group-wise quantization functions for activation discretization in different timesteps and sample the optimal timestep for informative calibration image generation, so that our quantized diffusion model can reduce the discretization errors with negligible computational overhead. Specifically, we partition the timesteps according to the importance weights of quantization functions in different groups, which are optimized by differentiable search algorithms. We also select the optimal timestep for calibration image generation by structural risk minimizing principle in order to enhance the generalization ability in the deployment of quantized diffusion model. Extensive experimental results show that our method outperforms the state-of-the-art post-training quantization of diffusion model by a sizable margin with similar computational cost.
We present BRICS, a bi-level feature representation for image collections, which consists of a key code space on top of a feature grid space. Specifically, our representation is learned by an autoencoder to encode images into continuous key codes, which are used to retrieve features from groups of multi-resolution feature grids. Our key codes and feature grids are jointly trained continuously with well-defined gradient flows, leading to high usage rates of the feature grids and improved generative modeling compared to discrete Vector Quantization (VQ). Differently from existing continuous representations such as KL-regularized latent codes, our key codes are strictly bounded in scale and variance. Overall, feature encoding by BRICS is compact, efficient to train, and enables generative modeling over key codes using the diffusion model. Experimental results show that our method achieves comparable reconstruction results to VQ while having a smaller and more efficient decoder network (50% fewer GFlops). By applying the diffusion model over our key code space, we achieve state-of-the-art performance on image synthesis on the FFHQ and LSUN-Church (29% lower than LDM, 32% lower than StyleGAN2, 44% lower than Projected GAN on CLIP-FID) datasets.
In recent months, the monkeypox (mpox) virus -- previously endemic in a limited area of the world -- has started spreading in multiple countries until being declared a ``public health emergency of international concern'' by the World Health Organization. The alert was renewed in February 2023 due to a persisting sustained incidence of the virus in several countries and worries about possible new outbreaks. Low-income countries with inadequate infrastructures for vaccine and testing administration are particularly at risk. A symptom of mpox infection is the appearance of skin rashes and eruptions, which can drive people to seek medical advice. A technology that might help perform a preliminary screening based on the aspect of skin lesions is the use of Machine Learning for image classification. However, to make this technology suitable on a large scale, it should be usable directly on mobile devices of people, with a possible notification to a remote medical expert. In this work, we investigate the adoption of Deep Learning to detect mpox from skin lesion images. The proposal leverages Transfer Learning to cope with the scarce availability of mpox image datasets. As a first step, a homogenous, unpolluted, dataset is produced by manual selection and preprocessing of available image data. It will also be released publicly to researchers in the field. Then, a thorough comparison is conducted amongst several Convolutional Neural Networks, based on a 10-fold stratified cross-validation. The best models are then optimized through quantization for use on mobile devices; measures of classification quality, memory footprint, and processing times validate the feasibility of our proposal. Additionally, the use of eXplainable AI is investigated as a suitable instrument to both technically and clinically validate classification outcomes.
Stochastic learning dynamics based on Langevin or Levy stochastic differential equations (SDEs) in deep neural networks control the variance of noise by varying the size of the mini-batch or directly those of injecting noise. Since the noise variance affects the approximation performance, the design of the additive noise is significant in SDE-based learning and practical implementation. In this paper, we propose an alternative stochastic descent learning equation based on quantized optimization for non-convex objective functions, adopting a stochastic analysis perspective. The proposed method employs a quantized optimization approach that utilizes Langevin SDE dynamics, allowing for controllable noise with an identical distribution without the need for additive noise or adjusting the mini-batch size. Numerical experiments demonstrate the effectiveness of the proposed algorithm on vanilla convolution neural network(CNN) models and the ResNet-50 architecture across various data sets. Furthermore, we provide a simple PyTorch implementation of the proposed algorithm.
In this paper, we present an efficient method for storing fine-tuned models by leveraging the low-rank properties of weight residuals. Our key observation is that weight residuals in large overparameterized models exhibit even stronger low-rank characteristics. Based on this insight, we propose Efficient Residual Encoding (ERE), a novel approach that achieves efficient storage of fine-tuned model weights by approximating the low-rank weight residuals. Furthermore, we analyze the robustness of weight residuals and push the limit of storage efficiency by utilizing additional quantization and layer-wise rank allocation. Our experimental results demonstrate that our method significantly reduces memory footprint while preserving performance in various tasks and modalities. We release our code.
Emergent properties have been widely adopted as a term to describe behavior not present in smaller models but observed in larger models. Recent work suggests that the trade-off incurred by quantization is also an emergent property, with sharp drops in performance in models over 6B parameters. In this work, we ask "are quantization cliffs in performance solely a factor of scale?" Against a backdrop of increased research focus on why certain emergent properties surface at scale, this work provides a useful counter-example. We posit that it is possible to optimize for a quantization friendly training recipe that suppresses large activation magnitude outliers. Here, we find that outlier dimensions are not an inherent product of scale, but rather sensitive to the optimization conditions present during pre-training. This both opens up directions for more efficient quantization, and poses the question of whether other emergent properties are inherent or can be altered and conditioned by optimization and architecture design choices. We successfully quantize models ranging in size from 410M to 52B with minimal degradation in performance.
Several post-training quantization methods have been applied to large language models (LLMs), and have been shown to perform well down to 8-bits. We find that these methods break down at lower bit precision, and investigate quantization aware training for LLMs (LLM-QAT) to push quantization levels even further. We propose a data-free distillation method that leverages generations produced by the pre-trained model, which better preserves the original output distribution and allows quantizing any generative model independent of its training data, similar to post-training quantization methods. In addition to quantizing weights and activations, we also quantize the KV cache, which is critical for increasing throughput and support long sequence dependencies at current model sizes. We experiment with LLaMA models of sizes 7B, 13B, and 30B, at quantization levels down to 4-bits. We observe large improvements over training-free methods, especially in the low-bit settings.
Static deep neural network (DNN) watermarking techniques typically employ irreversible methods to embed watermarks into the DNN model weights. However, this approach causes permanent damage to the watermarked model and fails to meet the requirements of integrity authentication. Reversible data hiding (RDH) methods offer a potential solution, but existing approaches suffer from weaknesses in terms of usability, capacity, and fidelity, hindering their practical adoption. In this paper, we propose a novel RDH-based static DNN watermarking scheme using quantization index modulation (QIM). Our scheme incorporates a novel approach based on a one-dimensional quantizer for watermark embedding. Furthermore, we design two schemes to address the challenges of integrity protection and legitimate authentication for DNNs. Through simulation results on training loss and classification accuracy, we demonstrate the feasibility and effectiveness of our proposed schemes, highlighting their superior adaptability compared to existing methods.
Large speech models are rapidly gaining traction in research community. As a result, model compression has become an important topic, so that these models can fit in memory and be served with reduced cost. Practical approaches for compressing automatic speech recognition (ASR) model use int8 or int4 weight quantization. In this study, we propose to develop 2-bit ASR models. We explore the impact of symmetric and asymmetric quantization combined with sub-channel quantization and clipping on both LibriSpeech dataset and large-scale training data. We obtain a lossless 2-bit Conformer model with 32% model size reduction when compared to state of the art 4-bit Conformer model for LibriSpeech. With the large-scale training data, we obtain a 2-bit Conformer model with over 40% model size reduction against the 4-bit version at the cost of 17% relative word error rate degradation
Large language models(LLMs) have sparked a new wave of exciting AI applications. Hosting these models at scale requires significant memory resources. One crucial memory bottleneck for the deployment stems from the context window. It is commonly recognized that model weights are memory hungry; however, the size of key-value embedding stored during the generation process (KV cache) can easily surpass the model size. The enormous size of the KV cache puts constraints on the inference batch size, which is crucial for high throughput inference workload. Inspired by an interesting observation of the attention scores, we hypothesize the persistence of importance: only pivotal tokens, which had a substantial influence at one step, will significantly influence future generations. Based on our empirical verification and theoretical analysis around this hypothesis, we propose Scissorhands, a system that maintains the memory usage of the KV cache at a fixed budget without finetuning the model. In essence, Scissorhands manages the KV cache by storing the pivotal tokens with a higher probability. We validate that Scissorhands reduces the inference memory usage of the KV cache by up to 5X without compromising model quality. We further demonstrate that Scissorhands can be combined with 4-bit quantization, traditionally used to compress model weights, to achieve up to 20X compression.
Accurate facial landmark detection is critical for facial analysis tasks, yet prevailing heatmap and coordinate regression methods grapple with prohibitive computational costs and quantization errors. Through comprehensive theoretical analysis and experimentation, we identify and elucidate the limitations of existing techniques. To overcome these challenges, we pioneer the application of True-Range Multilateration, originally devised for GPS localization, to facial landmark detection. We propose KeyPoint Positioning System (KeyPosS) - the first framework to deduce exact landmark coordinates by triangulating distances between points of interest and anchor points predicted by a fully convolutional network. A key advantage of KeyPosS is its plug-and-play nature, enabling flexible integration into diverse decoding pipelines. Extensive experiments on four datasets demonstrate state-of-the-art performance, with KeyPosS outperforming existing methods in low-resolution settings despite minimal computational overhead. By spearheading the integration of Multilateration with facial analysis, KeyPosS marks a paradigm shift in facial landmark detection. The code is available at https://github.com/zhiqic/KeyPosS.
Just Noticeable Distortion (JND) finds the minimum distortion level perceivable by humans. This can be a natural solution for setting the compression for each video region in perceptual video coding. However, existing JND-based solutions estimate JND levels for each video frame and ignore the fact that different video regions have different perceptual importance. To address this issue, we propose a Block-Level Just Noticeable Distortion-based Perceptual (BL-JUNIPER) framework for video coding. The proposed four-stage framework combines different perceptual information to further improve the prediction accuracy. The JND mapping in the first stage derives block-level JNDs from frame-level information without the need to collect a new bock-level JND dataset. In the second stage, an efficient CNN-based model is proposed to predict JND levels for each block according to spatial and temporal characteristics. Unlike existing methods, BL-JUNIPER works on raw video frames and avoids re-encoding each frame several times, making it computationally practical. Third, the visual importance of each block is measured using a visual attention model. Finally, a proposed quantization control algorithm uses both JND levels and visual importance to adjust the Quantization Parameter (QP) for each block. The specific algorithm for each stage of the proposed framework can be changed, as long as the input and output formats of each block are followed, without the need to change other stages, based on any current or future methods, providing a flexible and robust solution. Extensive experimental results demonstrate that BL-JUNIPER achieves a mean bitrate reduction of 27.75% with a Delta Mean Opinion Score (DMOS) close to zero and BD-Rate gains of 25.44% based on MOS, compared to the baseline encoding, and also gains a better performance compared to competing methods.
With the rapid increase in the size of neural networks, model compression has become an important area of research. Quantization is an effective technique at decreasing the model size, memory access, and compute load of large models. Despite recent advances in quantization aware training (QAT) technique, most papers present evaluations that are focused on computer vision tasks, which have different training dynamics compared to sequence tasks. In this paper, we first benchmark the impact of popular techniques such as straight through estimator, pseudo-quantization noise, learnable scale parameter, clipping, etc. on 4-bit seq2seq models across a suite of speech recognition datasets ranging from 1,000 hours to 1 million hours, as well as one machine translation dataset to illustrate its applicability outside of speech. Through the experiments, we report that noise based QAT suffers when there is insufficient regularization signal flowing back to the quantization scale. We propose low complexity changes to the QAT process to improve model accuracy (outperforming popular learnable scale and clipping methods). With the improved accuracy, it opens up the possibility to exploit some of the other benefits of noise based QAT: 1) training a single model that performs well in mixed precision mode and 2) improved generalization on long form speech recognition.
In theory, vector quantization (VQ) is always better than scalar quantization (SQ) in terms of rate-distortion (R-D) performance. Recent state-of-the-art methods for neural image compression are mainly based on nonlinear transform coding (NTC) with uniform scalar quantization, overlooking the benefits of VQ due to its exponentially increased complexity. In this paper, we first investigate on some toy sources, demonstrating that even if modern neural networks considerably enhance the compression performance of SQ with nonlinear transform, there is still an insurmountable chasm between SQ and VQ. Therefore, revolving around VQ, we propose a novel framework for neural image compression named Nonlinear Vector Transform Coding (NVTC). NVTC solves the critical complexity issue of VQ through (1) a multi-stage quantization strategy and (2) nonlinear vector transforms. In addition, we apply entropy-constrained VQ in latent space to adaptively determine the quantization boundaries for joint rate-distortion optimization, which improves the performance both theoretically and experimentally. Compared to previous NTC approaches, NVTC demonstrates superior rate-distortion performance, faster decoding speed, and smaller model size. Our code is available at https://github.com/USTC-IMCL/NVTC
Vision Transformers (ViTs) have emerged as the fundamental architecture for most computer vision fields, but the considerable memory and computation costs hinders their application on resource-limited devices. As one of the most powerful compression methods, binarization reduces the computation of the neural network by quantizing the weights and activation values as $\pm$1. Although existing binarization methods have demonstrated excellent performance on Convolutional Neural Networks (CNNs), the full binarization of ViTs is still under-studied and suffering a significant performance drop. In this paper, we first argue empirically that the severe performance degradation is mainly caused by the weight oscillation in the binarization training and the information distortion in the activation of ViTs. Based on these analyses, we propose $\textbf{BinaryViT}$, an accurate full binarization scheme for ViTs, which pushes the quantization of ViTs to the limit. Specifically, we propose a novel gradient regularization scheme (GRS) for driving a bimodal distribution of the weights to reduce oscillation in binarization training. Moreover, we design an activation shift module (ASM) to adaptively tune the activation distribution to reduce the information distortion caused by binarization. Extensive experiments on ImageNet dataset show that our BinaryViT consistently surpasses the strong baseline by 2.05% and improve the accuracy of fully binarized ViTs to a usable level. Furthermore, our method achieves impressive savings of 16.2$\times$ and 17.7$\times$ in model size and OPs compared to the full-precision DeiT-S.
Large language models (LLMs) enable unparalleled few- and zero-shot reasoning capabilities but at a high computational footprint. A growing assortment of methods for compression promises to reduce the computational burden of LLMs in deployment, but so far, only quantization approaches have been demonstrated to be effective for LLM compression while maintaining zero-shot performance. A critical step in the compression process, the pretrain-then-finetune paradigm, has largely been overlooked when adapting existing pruning strategies to LLMs or proposing new ones. In this work, we show that embarrassingly simple layer pruning coupled with an extended language model pretraining as the finetuning phase produces state-of-the-art results against structured and even semi-structured compression of models at a 7B scale while being more inference efficient. We call this method LayerChop, where we deterministically remove layers from a model followed by task-agnostic finetuning of the remaining weights by continued self-supervised pretraining. At this scale, we also show how distillation, which has been super effective in task-agnostic compression of smaller BERT-style models, becomes inefficient against our simple pruning technique.
Semantic communication, an intelligent communication paradigm that aims to transmit useful information in the semantic domain, is facilitated by deep learning techniques. Robust semantic features can be learned and transmitted in an analog fashion, but it poses new challenges to hardware, protocol, and encryption. In this paper, we propose a digital semantic communication system, which consists of an encoding network deployed on a resource-limited device and a decoding network deployed at the edge. To acquire better semantic representation for digital transmission, a novel non-linear quantization module is proposed to efficiently quantize semantic features with trainable quantization levels. Additionally, structured pruning is incorporated to reduce the dimension of the transmitted features. We also introduce a semantic learning loss (SLL) function to reduce semantic error. To adapt to various channel conditions and inputs under constraints of communication and computing resources, a policy network is designed to adaptively choose the split point and the dimension of the transmitted semantic features. Experiments using the CIFAR-10 and ImageNet dataset for image classification are employed to evaluate the proposed digital semantic communication network, and ablation studies are conducted to assess the proposed quantization module, structured pruning and SLL.
Intelligent reflecting surfaces (IRSs) are being widely investigated as a potential low-cost and energy-efficient alternative to active relays for improving coverage in next-generation cellular networks. However, technical constraints in the configuration of IRSs should be taken into account in the design of scheduling solutions and the assessment of their performance. To this end, we examine an IRS-assisted time division multiple access (TDMA) cellular network where the reconfiguration of the IRS incurs a communication cost; thus, we aim at limiting the number of reconfigurations over time. Along these lines, we propose a clustering-based heuristic scheduling scheme that maximizes the cell sum capacity, subject to a fixed number of reconfigurations within a TDMA frame. First, the best configuration of each user equipment (UE), in terms of joint beamforming and optimal IRS configuration, is determined using an iterative algorithm. Then, we propose different clustering techniques to divide the UEs into subsets sharing the same sub-optimal IRS configuration, derived through distance- and capacity-based algorithms. Finally, UEs within the same cluster are scheduled accordingly. We provide extensive numerical results for different propagation scenarios, IRS sizes, and phase shifters quantization constraints, showing the effectiveness of our approach in supporting multi-user IRS systems with practical constraints.
Building upon Randomized Discretization, we develop two novel adversarial defenses against white-box PGD attacks, utilizing vector quantization in higher dimensional spaces. These methods, termed pRD and swRD, not only offer a theoretical guarantee in terms of certified accuracy, they are also shown, via abundant experiments, to perform comparably or even superior to the current art of adversarial defenses. These methods can be extended to a version that allows further training of the target classifier and demonstrates further improved performance.
Existing autoregressive models follow the two-stage generation paradigm that first learns a codebook in the latent space for image reconstruction and then completes the image generation autoregressively based on the learned codebook. However, existing codebook learning simply models all local region information of images without distinguishing their different perceptual importance, which brings redundancy in the learned codebook that not only limits the next stage's autoregressive model's ability to model important structure but also results in high training cost and slow generation speed. In this study, we borrow the idea of importance perception from classical image coding theory and propose a novel two-stage framework, which consists of Masked Quantization VAE (MQ-VAE) and Stackformer, to relieve the model from modeling redundancy. Specifically, MQ-VAE incorporates an adaptive mask module for masking redundant region features before quantization and an adaptive de-mask module for recovering the original grid image feature map to faithfully reconstruct the original images after quantization. Then, Stackformer learns to predict the combination of the next code and its position in the feature map. Comprehensive experiments on various image generation validate our effectiveness and efficiency. Code will be released at https://github.com/CrossmodalGroup/MaskedVectorQuantization.