Federated learning (FL) is a framework which allows multiple users to jointly train a global machine learning (ML) model by transmitting only model updates under the coordination of a parameter server, while being able to keep their datasets local. One key motivation of such distributed frameworks is to provide privacy guarantees to the users. However, preserving the users' datasets locally is shown to be not sufficient for privacy. Several differential privacy (DP) mechanisms have been proposed to provide provable privacy guarantees by introducing randomness into the framework, and majority of these mechanisms rely on injecting additive noise. FL frameworks also face the challenge of communication efficiency, especially as machine learning models grow in complexity and size. Quantization is a commonly utilized method, reducing the communication cost by transmitting compressed representation of the underlying information. Although there have been several studies on DP and quantization in FL, the potential contribution of the quantization method alone in providing privacy guarantees has not been extensively analyzed yet. We in this paper present a novel stochastic quantization method, utilizing a mixed geometric distribution to introduce the randomness needed to provide DP, without any additive noise. We provide convergence analysis for our framework and empirically study its performance.
Transformer-based models, such as BERT, have been widely applied in a wide range of natural language processing tasks. However, one inevitable side effect is that they require massive memory storage and inference cost when deployed in production. Quantization is one of the popularized ways to alleviate the cost. However, the previous 8-bit quantization strategy based on INT8 data format either suffers from the degradation of accuracy in a Post-Training Quantization (PTQ) fashion or requires an expensive Quantization-Aware Training (QAT) process. Recently, a new numeric format FP8 (i.e. floating-point of 8-bits) has been proposed and supported in commercial AI computing platforms such as H100. In this paper, we empirically validate the effectiveness of FP8 as a way to do Post-Training Quantization without significant loss of accuracy, with a simple calibration and format conversion process. We adopt the FP8 standard proposed by NVIDIA Corp. (2022) in our extensive experiments of BERT variants on GLUE and SQuAD v1.1 datasets, and show that PTQ with FP8 can significantly improve the accuracy upon that with INT8, to the extent of the full-precision model.
Large Language Models (LLMs) stand out for their impressive performance in intricate language modeling tasks. However, their demanding computational and memory needs pose obstacles for broad use on edge devices. Quantization is then introduced to boost LLMs' on-device efficiency. Recent works show that 8-bit or lower weight quantization is feasible with minimal impact on end-to-end task performance, while the activation is still not quantized. On the other hand, mainstream commodity edge devices still struggle to execute these sub-8-bit quantized networks effectively. In this paper, we propose Agile-Quant, an activation-guided quantization framework for popular Large Language Models (LLMs), and implement an end-to-end accelerator on multiple edge devices for faster inference. Considering the hardware profiling and activation analysis, we first introduce a basic activation quantization strategy to balance the trade-off of task performance and real inference speed. Then we leverage the activation-aware token pruning technique to reduce the outliers and the adverse impact on attentivity. Ultimately, we utilize the SIMD-based 4-bit multiplier and our efficient TRIP matrix multiplication to implement the accelerator for LLMs on the edge. We apply our framework on different scales of LLMs including LLaMA, OPT, and BLOOM with 4-bit or 8-bit for the activation and 4-bit for the weight quantization. Experiments show that Agile-Quant achieves simultaneous quantization of model weights and activations while maintaining task performance comparable to existing weight-only quantization methods. Moreover, in the 8- and 4-bit scenario, Agile-Quant achieves an on-device speedup of up to 2.55x compared to its FP16 counterparts across multiple edge devices, marking a pioneering advancement in this domain. Code: https://github.com/shawnricecake/agile-quant
Latent Diffusion Models (LDMs) capture the dynamic evolution of latent variables over time, blending patterns and multimodality in a generative system. Despite the proficiency of LDM in various applications, such as text-to-image generation, facilitated by robust text encoders and a variational autoencoder, the critical need to deploy large generative models on edge devices compels a search for more compact yet effective alternatives. Post Training Quantization (PTQ), a method to compress the operational size of deep learning models, encounters challenges when applied to LDM due to temporal and structural complexities. This study proposes a quantization strategy that efficiently quantize LDMs, leveraging Signal-to-Quantization-Noise Ratio (SQNR) as a pivotal metric for evaluation. By treating the quantization discrepancy as relative noise and identifying sensitive part(s) of a model, we propose an efficient quantization approach encompassing both global and local strategies. The global quantization process mitigates relative quantization noise by initiating higher-precision quantization on sensitive blocks, while local treatments address specific challenges in quantization-sensitive and time-sensitive modules. The outcomes of our experiments reveal that the implementation of both global and local treatments yields a highly efficient and effective Post Training Quantization (PTQ) of LDMs.
In this correspondence, we propose hierarchical high-altitude platform (HAP)-low-altitude platform (LAP) networks with the aim of maximizing the sum-rate of ground user equipments (UEs). The multiple aerial radio units (RUs) mounted on HAPs and LAPs are managed by the central unit (CU) via constrained fronthaul links. The limitation of fronthaul capacity can be addressed through quantization, employing the cloud radio access network (C-RAN) architecture. For spectral efficiency, we adopt the rate-splitting multiple access (RSMA), leveraging the advantages of both space-division multiple access (SDMA) and non-orthogonal multiple access (NOMA). To achieve this, we jointly optimize rate splitting, transmit power allocation, quantization noise variance, and UAV placement using an alternating optimization (AO) approach coupled with successive convex approximation (SCA) and the weighted minimum mean square error (WMMSE) method. Numerical results validate the superior performance of the proposed method compared to benchmark schemes, including partial optimizations or those without the assistance of LAPs.
Large language models (LLMs) have shown remarkable capabilities in various tasks. However their huge model size and the consequent demand for computational and memory resources also pose challenges to model deployment. Currently, 4-bit post-training quantization (PTQ) has achieved some success in LLMs, reducing the memory footprint by approximately 75% compared to FP16 models, albeit with some accuracy loss. In this paper, we propose SmoothQuant+, an accurate and efficient 4-bit weight-only PTQ that requires no additional training, which enables lossless in accuracy for LLMs for the first time. Based on the fact that the loss of weight quantization is amplified by the activation outliers, SmoothQuant+ smoothes the activation outliers by channel before quantization, while adjusting the corresponding weights for mathematical equivalence, and then performs group-wise 4-bit weight quantization for linear layers. We have integrated SmoothQuant+ into the vLLM framework, an advanced high-throughput inference engine specially developed for LLMs, and equipped it with an efficient W4A16 CUDA kernels, so that vLLM can seamlessly support SmoothQuant+ 4-bit weight quantization. Our results show that, with SmoothQuant+, the Code Llama-34B model can be quantized and deployed on a A100 40GB GPU, achieving lossless accuracy and a throughput increase of 1.9 to 4.0 times compared to the FP16 model deployed on two A100 40GB GPUs. Moreover, the latency per token is only 68% of the FP16 model deployed on two A100 40GB GPUs. This is the state-of-the-art 4-bit weight quantization for LLMs as we know.
Spatio-temporal forecasting is crucial in various fields and requires a careful balance between identifying subtle patterns and filtering out noise. Vector quantization (VQ) appears well-suited for this purpose, as it quantizes input vectors into a set of codebook vectors or patterns. Although VQ has shown promise in various computer vision tasks, it surprisingly falls short in enhancing the accuracy of spatio-temporal forecasting. We attribute this to two main issues: inaccurate optimization due to non-differentiability and limited representation power in hard-VQ. To tackle these challenges, we introduce Differentiable Sparse Soft-Vector Quantization (SVQ), the first VQ method to enhance spatio-temporal forecasting. SVQ balances detail preservation with noise reduction, offering full differentiability and a solid foundation in sparse regression. Our approach employs a two-layer MLP and an extensive codebook to streamline the sparse regression process, significantly cutting computational costs while simplifying training and improving performance. Empirical studies on five spatio-temporal benchmark datasets show SVQ achieves state-of-the-art results, including a 7.9% improvement on the WeatherBench-S temperature dataset and an average mean absolute error reduction of 9.4% in video prediction benchmarks (Human3.6M, KTH, and KittiCaltech), along with a 17.3% enhancement in image quality (LPIPS). Code is publicly available at https://github.com/Pachark/SVQ-Forecasting.
The challenge of kinship verification from facial images represents a cutting-edge and formidable frontier in the realms of pattern recognition and computer vision. This area of study holds a myriad of potential applications, spanning from image annotation and forensic analysis to social media research. Our research stands out by integrating a preprocessing method named Multiscale Retinex (MSR), which elevates image quality and amplifies contrast, ultimately bolstering the end results. Strategically, our methodology capitalizes on the harmonious blend of deep and shallow texture descriptors, merging them proficiently at the score level through the Logistic Regression (LR) method. To elucidate, we employ the Local Phase Quantization (LPQ) descriptor to extract shallow texture characteristics. For deep feature extraction, we turn to the prowess of the VGG16 model, which is pre-trained on a convolutional neural network (CNN). The robustness and efficacy of our method have been put to the test through meticulous experiments on three rigorous kinship datasets, namely: Cornell Kin Face, UB Kin Face, and TS Kin Face.
In recent years, there has been a significant trend in deep neural networks (DNNs), particularly transformer-based models, of developing ever-larger and more capable models. While they demonstrate state-of-the-art performance, their growing scale requires increased computational resources (e.g., GPUs with greater memory capacity). To address this problem, quantization techniques (i.e., low-bit-precision representation and matrix multiplication) have been proposed. Most quantization techniques employ a static strategy in which the model parameters are quantized, either during training or inference, without considering the test-time sample. In contrast, dynamic quantization techniques, which have become increasingly popular, adapt during inference based on the input provided, while maintaining full-precision performance. However, their dynamic behavior and average-case performance assumption makes them vulnerable to a novel threat vector -- adversarial attacks that target the model's efficiency and availability. In this paper, we present QuantAttack, a novel attack that targets the availability of quantized models, slowing down the inference, and increasing memory usage and energy consumption. We show that carefully crafted adversarial examples, which are designed to exhaust the resources of the operating system, can trigger worst-case performance. In our experiments, we demonstrate the effectiveness of our attack on vision transformers on a wide range of tasks, both uni-modal and multi-modal. We also examine the effect of different attack variants (e.g., a universal perturbation) and the transferability between different models.
This paper presents a mixed-computation neural network processing approach for edge applications that incorporates low-precision (low-width) Posit and low-precision fixed point (FixP) number systems. This mixed-computation approach employs 4-bit Posit (Posit4), which has higher precision around zero, for representing weights with high sensitivity, while it uses 4-bit FixP (FixP4) for representing other weights. A heuristic for analyzing the importance and the quantization error of the weights is presented to assign the proper number system to different weights. Additionally, a gradient approximation for Posit representation is introduced to improve the quality of weight updates in the backpropagation process. Due to the high energy consumption of the fully Posit-based computations, neural network operations are carried out in FixP or Posit/FixP. An efficient hardware implementation of a MAC operation with a first Posit operand and FixP for a second operand and accumulator is presented. The efficacy of the proposed low-precision mixed-computation approach is extensively assessed on vision and language models. The results show that, on average, the accuracy of the mixed-computation is about 1.5% higher than that of FixP with a cost of 0.19% energy overhead.
JPEG is still the most widely used image compression algorithm. Most image compression algorithms only consider uncompressed original image, while ignoring a large number of already existing JPEG images. Recently, JPEG recompression approaches have been proposed to further reduce the size of JPEG files. However, those methods only consider JPEG lossless recompression, which is just a special case of the rate-distortion theorem. In this paper, we propose a unified lossly and lossless JPEG recompression framework, which consists of learned quantization table and Markovian hierarchical variational autoencoders. Experiments show that our method can achieve arbitrarily low distortion when the bitrate is close to the upper bound, namely the bitrate of the lossless compression model. To the best of our knowledge, this is the first learned method that bridges the gap between lossy and lossless recompression of JPEG images.
Pruning-quantization joint learning always facilitates the deployment of deep neural networks (DNNs) on resource-constrained edge devices. However, most existing methods do not jointly learn a global criterion for pruning and quantization in an interpretable way. In this paper, we propose a novel physics inspired criterion for pruning-quantization joint learning (PIC-PQ), which is explored from an analogy we first draw between elasticity dynamics (ED) and model compression (MC). Specifically, derived from Hooke's law in ED, we establish a linear relationship between the filters' importance distribution and the filter property (FP) by a learnable deformation scale in the physics inspired criterion (PIC). Furthermore, we extend PIC with a relative shift variable for a global view. To ensure feasibility and flexibility, available maximum bitwidth and penalty factor are introduced in quantization bitwidth assignment. Experiments on benchmarks of image classification demonstrate that PIC-PQ yields a good trade-off between accuracy and bit-operations (BOPs) compression ratio e.g., 54.96X BOPs compression ratio in ResNet56 on CIFAR10 with 0.10% accuracy drop and 53.24X in ResNet18 on ImageNet with 0.61% accuracy drop). The code will be available at https://github.com/fanxxxxyi/PIC-PQ.
Compressing large language models (LLMs), often consisting of billions of parameters, provides faster inference, smaller memory footprints, and enables local deployment. Two standard compression techniques are pruning and quantization, with the former eliminating redundant connections in model layers and the latter representing model parameters with fewer bits. The key tradeoff is between the degree of compression and the impact on the quality of the compressed model. Existing research on LLM compression primarily focuses on performance in terms of general metrics like perplexity or downstream task accuracy. More fine-grained metrics, such as those measuring parametric knowledge, remain significantly underexplored. To help bridge this gap, we present a comprehensive analysis across multiple model families (ENCODER, ENCODER-DECODER, and DECODER) using the LAMA and LM-HARNESS benchmarks in order to systematically quantify the effect of commonly employed compression techniques on model performance. A particular focus is on tradeoffs involving parametric knowledge, with the goal of providing practitioners with practical insights to help make informed decisions on compression. We release our codebase1 to enable further research.
Efficient inference of Deep Neural Networks (DNNs) on resource-constrained edge devices is essential. Quantization and sparsity are key techniques that translate to repetition and sparsity within tensors at the hardware-software interface. This paper introduces the concept of repetition-sparsity trade-off that helps explain computational efficiency during inference. We propose PLUM, a unified co-design framework that integrates DNN inference systems and quantization (forward and backward pass) to leverage the repetition-sparsity trade-off to improve inference efficiency. Our results demonstrate that PLUM's quantization method is more accurate than binary quantization with the same number of non-zero weights. Detailed analysis indicates that signed binarization generates a smaller distribution of effectual (non-zero) parameters nested within a larger distribution of total parameters of latent full-precision weights for a DNN block. Finally, the proposed PLUM framework achieves a 26% speedup on real hardware, doubles energy efficiency, and reduces density by 2.8x compared to binary methods while retaining top-1 accuracy when compared to prior-art methods for ResNets on ImageNet (by achieving 66.2% top-1 accuracy), presenting an alternative solution for deploying efficient models in resource-limited environments.
Semantic communication, recognized as a promising technology for future intelligent applications, has received widespread research attention. Despite the potential of semantic communication to enhance transmission reliability, especially in low signal-to-noise (SNR) environments, the critical issue of resource allocation and compatibility in the dynamic wireless environment remains largely unexplored. In this paper, we propose an adaptive semantic resource allocation paradigm with semantic-bit quantization (SBQ) compatibly for existing wireless communications, where the inaccurate environment perception introduced by the additional mapping relationship between semantic metrics and transmission metrics is solved. In order to investigate the performance of semantic communication networks, the quality of service for semantic communication (SC-QoS), including the semantic quantization efficiency (SQE) and transmission latency, is proposed for the first time. A problem of maximizing the overall effective SC-QoS is formulated by jointly optimizing the transmit beamforming of the base station, the bits for semantic representation, the subchannel assignment, and the bandwidth resource allocation. To address the non-convex formulated problem, an intelligent resource allocation scheme is proposed based on a hybrid deep reinforcement learning (DRL) algorithm, where the intelligent agent can perceive both semantic tasks and dynamic wireless environments. Simulation results demonstrate that our design can effectively combat semantic noise and achieve superior performance in wireless communications compared to several benchmark schemes. Furthermore, compared to mapping-guided paradigm based resource allocation schemes, our proposed adaptive scheme can achieve up to 13% performance improvement in terms of SC-QoS.
Most real-world applications that employ deep neural networks (DNNs) quantize them to low precision to reduce the compute needs. We present a method to improve the robustness of quantized DNNs to white-box adversarial attacks. We first tackle the limitation of deterministic quantization to fixed ``bins'' by introducing a differentiable Stochastic Quantizer (SQ). We explore the hypothesis that different quantizations may collectively be more robust than each quantized DNN. We formulate a training objective to encourage different quantized DNNs to learn different representations of the input image. The training objective captures diversity and accuracy via mutual information between ensemble members. Through experimentation, we demonstrate substantial improvement in robustness against $L_\infty$ attacks even if the attacker is allowed to backpropagate through SQ (e.g., > 50\% accuracy to PGD(5/255) on CIFAR10 without adversarial training), compared to vanilla DNNs as well as existing ensembles of quantized DNNs. We extend the method to detect attacks and generate robustness profiles in the adversarial information plane (AIP), towards a unified analysis of different threat models by correlating the MI and accuracy.
The rapid scaling of language models is motivating research using low-bitwidth quantization. In this work, we propose a novel binarization technique for Transformers applied to machine translation (BMT), the first of its kind. We identify and address the problem of inflated dot-product variance when using one-bit weights and activations. Specifically, BMT leverages additional LayerNorms and residual connections to improve binarization quality. Experiments on the WMT dataset show that a one-bit weight-only Transformer can achieve the same quality as a float one, while being 16x smaller in size. One-bit activations incur varying degrees of quality drop, but mitigated by the proposed architectural changes. We further conduct a scaling law study using production-scale translation datasets, which shows that one-bit weight Transformers scale and generalize well in both in-domain and out-of-domain settings. Implementation in JAX/Flax will be open sourced.
We present a novel, power- & hardware-efficient, multiuser, multibeam RIS (Reflective Intelligent Surface) architecture for multiuser MIMO, especially for very high frequency bands (e.g., high mmWave and sub-THz), where channels are typically sparse in the beamspace and LOS is the dominant component. The key module is formed by an active multiantenna feeder (AMAF) with a small number of active antennas, placed in the near field of a RIS with a much larger number of passive controllable reflecting elements. We propose a pragmatic approach to obtain a steerable beam with high gain and very low sidelobes. Then K independently controlled beams can be achieved by closely stacking K such AMAF-RIS modules. Our analysis includes the mutual interference between the modules and the fact that, due to the delay difference of propagation through the AMAF-RIS structure, the resulting channel matrix is frequency selective even for pure LOS propagation. We consider a 3D geometry and show that "beam focusing" is in fact possible (and much more effective in terms of coverage) also in the far-field, by creating spotbeams with limited footprint both in angle and in range. Our results show that: 1) simple RF beamforming (BF) without computationally expensive baseband multiuser precoding is sufficient to practically eliminate multiuser interference when the users are chosen with sufficient angular/range separation, thanks to the extremely low sidelobe beams; 2) the impact of beam pointing errors with standard deviation as large as 2.5 deg and RIS quantized phase-shifters with quantization bits > 2 is essentially negligible; 3) The proposed architecture is more power efficient & much simpler from a hardware implementation viewpoint than standard active arrays with the same BF performance. We show also that the array gain of the proposed AMAF-RIS structure is linear with the RIS aperture.
Generative Language Models (GLMs) have shown impressive performance in tasks such as text generation, understanding, and reasoning. However, the large model size poses challenges for practical deployment. To solve this problem, Quantization-Aware Training (QAT) has become increasingly popular. However, current QAT methods for generative models have resulted in a noticeable loss of accuracy. To counteract this issue, we propose a novel knowledge distillation method specifically designed for GLMs. Our method, called token-scaled logit distillation, prevents overfitting and provides superior learning from the teacher model and ground truth. This research marks the first evaluation of ternary weight quantization-aware training of large-scale GLMs with less than 1.0 degradation in perplexity and achieves enhanced accuracy in tasks like common-sense QA and arithmetic reasoning as well as natural language understanding. Our code is available at https://github.com/aiha-lab/TSLD.
This work studies post-training parameter quantization in large language models (LLMs). We introduce quantization with incoherence processing (QuIP), a new method based on the insight that quantization benefits from $\textit{incoherent}$ weight and Hessian matrices, i.e., from the weights being even in magnitude and the directions in which it is important to round them accurately being unaligned with the coordinate axes. QuIP consists of two steps: (1) an adaptive rounding procedure minimizing a quadratic proxy objective; (2) efficient pre- and post-processing that ensures weight and Hessian incoherence via multiplication by random orthogonal matrices. We complement QuIP with the first theoretical analysis for an LLM-scale quantization algorithm, and show that our theory also applies to an existing method, OPTQ. Empirically, we find that our incoherence preprocessing improves several existing quantization algorithms and yields the first LLM quantization methods that produce viable results using only two bits per weight. Our code can be found at https://github.com/Cornell-RelaxML/QuIP.