Neural compression has brought tremendous progress in designing lossy compressors with good rate-distortion (RD) performance at low complexity. Thus far, neural compression design involves transforming the source to a latent vector, which is then rounded to integers and entropy coded. While this approach has been shown to be optimal in a one-shot sense on certain sources, we show that it is highly sub-optimal on i.i.d. sequences, and in fact always recovers scalar quantization of the original source sequence. We demonstrate that the sub-optimality is due to the choice of quantization scheme in the latent space, and not the transform design. By employing lattice quantization instead of scalar quantization in the latent space, we demonstrate that Lattice Transform Coding (LTC) is able to recover optimal vector quantization at various dimensions and approach the asymptotically-achievable rate-distortion function at reasonable complexity. On general vector sources, LTC improves upon standard neural compressors in one-shot coding performance. LTC also enables neural compressors that perform block coding on i.i.d. vector sources, which yields coding gain over optimal one-shot coding.
Post-training quantization (PTQ) has emerged as a practical approach to compress large neural networks, making them highly efficient for deployment. However, effectively reducing these models to their low-bit counterparts without compromising the original accuracy remains a key challenge. In this paper, we propose an innovative PTQ algorithm termed COMQ, which sequentially conducts coordinate-wise minimization of the layer-wise reconstruction errors. We consider the widely used integer quantization, where every quantized weight can be decomposed into a shared floating-point scalar and an integer bit-code. Within a fixed layer, COMQ treats all the scaling factor(s) and bit-codes as the variables of the reconstruction error. Every iteration improves this error along a single coordinate while keeping all other variables constant. COMQ is easy to use and requires no hyper-parameter tuning. It instead involves only dot products and rounding operations. We update these variables in a carefully designed greedy order, significantly enhancing the accuracy. COMQ achieves remarkable results in quantizing 4-bit Vision Transformers, with a negligible loss of less than 1% in Top-1 accuracy. In 4-bit INT quantization of convolutional neural networks, COMQ maintains near-lossless accuracy with a minimal drop of merely 0.3% in Top-1 accuracy.
We introduce Chronos, a simple yet effective framework for pretrained probabilistic time series models. Chronos tokenizes time series values using scaling and quantization into a fixed vocabulary and trains existing transformer-based language model architectures on these tokenized time series via the cross-entropy loss. We pretrained Chronos models based on the T5 family (ranging from 20M to 710M parameters) on a large collection of publicly available datasets, complemented by a synthetic dataset that we generated via Gaussian processes to improve generalization. In a comprehensive benchmark consisting of 42 datasets, and comprising both classical local models and deep learning methods, we show that Chronos models: (a) significantly outperform other methods on datasets that were part of the training corpus; and (b) have comparable and occasionally superior zero-shot performance on new datasets, relative to methods that were trained specifically on them. Our results demonstrate that Chronos models can leverage time series data from diverse domains to improve zero-shot accuracy on unseen forecasting tasks, positioning pretrained models as a viable tool to greatly simplify forecasting pipelines.
This paper presents a finite-rate deep-learning (DL)-based channel state information (CSI) feedback method for massive multiple-input multiple-output (MIMO) systems. The presented method provides a finite-bit representation of the latent vector based on a vector-quantized variational autoencoder (VQ-VAE) framework while reducing its computational complexity based on shape-gain vector quantization. In this method, the magnitude of the latent vector is quantized using a non-uniform scalar codebook with a proper transformation function, while the direction of the latent vector is quantized using a trainable Grassmannian codebook. A multi-rate codebook design strategy is also developed by introducing a codeword selection rule for a nested codebook along with the design of a loss function. Simulation results demonstrate that the proposed method reduces the computational complexity associated with VQ-VAE while improving CSI reconstruction performance under a given feedback overhead.
Transformers are the backbone of powerful foundation models for many Vision and Natural Language Processing tasks. But their compute and memory/storage footprint is large, and so, serving such models is expensive often requiring high-end hardware. To mitigate this difficulty, Post-Training Quantization seeks to modify a pre-trained model and quantize it to eight bits or lower, significantly boosting compute/memory/latency efficiency. Such models have been successfully quantized to four bits with some performance loss. In this work, we outline a simple scheme to quantize Transformer-based models to just two bits (plus some overhead) with only a small drop in accuracy. Key to our formulation is a concept borrowed from Harmonic analysis called Fusion Frames. Our main finding is that the quantization must take place not in the original weight space, but instead in the Fusion Frame representations. If quantization is interpreted as the addition of noise, our casting of the problem allows invoking an extensive body of known consistent recovery and noise robustness guarantees. Further, if desired, de-noising filters are known in closed form. We show empirically, via a variety of experiments, that (almost) two-bit quantization for Transformer models promises sizable efficiency gains. The code is available at https://github.com/vsingh-group/FrameQuant
Solar energy is rapidly becoming a robust renewable energy source to conventional finite resources such as fossil fuels. It is harvested using interconnected photovoltaic panels, typically built with crystalline silicon cells, i.e. semiconducting materials that convert effectively the solar radiation into electricity. However, crystalline silicon is fragile and vulnerable to cracking over time or in predictive maintenance tasks, which can lead to electric isolation of parts of the solar cell and even failure, thus affecting the panel performance and reducing electricity generation. This work aims to developing a system for detecting cell cracks in solar panels to anticipate and alaert of a potential failure of the photovoltaic system by using computer vision techniques. Three scenarios are defined where these techniques will bring value. In scenario A, images are taken manually and the system detecting failures in the solar cells is not subject to any computationa constraints. In scenario B, an Edge device is placed near the solar farm, able to make inferences. Finally, in scenario C, a small microcontroller is placed in a drone flying over the solar farm and making inferences about the solar cells' states. Three different architectures are found the most suitable solutions, one for each scenario, namely the InceptionV3 model, an EfficientNetB0 model shrunk into full integer quantization, and a customized CNN architechture built with VGG16 blocks.
Transformer-based models have gained widespread popularity in both the computer vision (CV) and natural language processing (NLP) fields. However, significant challenges arise during post-training linear quantization, leading to noticeable reductions in inference accuracy. Our study focuses on uncovering the underlying causes of these accuracy drops and proposing a quantization-friendly fine-tuning method, \textbf{QuantTune}. Firstly, our analysis revealed that, on average, 65\% of quantization errors result from the precision loss incurred by the dynamic range amplification effect of outliers across the target Transformer-based models. Secondly, \textbf{QuantTune} adjusts weights based on the deviation of outlier activations and effectively constrains the dynamic ranges of the problematic activations. As a result, it successfully mitigates the negative impact of outliers on the inference accuracy of quantized models. Lastly, \textbf{QuantTune} can be seamlessly integrated into the back-propagation pass in the fine-tuning process without requiring extra complexity in inference software and hardware design. Our approach showcases significant improvements in post-training quantization across a range of Transformer-based models, including ViT, Bert-base, and OPT. QuantTune reduces accuracy drops by 12.09\% at 8-bit quantization and 33.8\% at 7-bit compared to top calibration methods, outperforming state-of-the-art solutions by over 18.84\% across ViT models.
Generating emotional talking faces is a practical yet challenging endeavor. To create a lifelike avatar, we draw upon two critical insights from a human perspective: 1) The connection between audio and the non-deterministic facial dynamics, encompassing expressions, blinks, poses, should exhibit synchronous and one-to-many mapping. 2) Vibrant expressions are often accompanied by emotion-aware high-definition (HD) textures and finely detailed teeth. However, both aspects are frequently overlooked by existing methods. To this end, this paper proposes using normalizing Flow and Vector-Quantization modeling to produce emotional talking faces that satisfy both insights concurrently (FlowVQTalker). Specifically, we develop a flow-based coefficient generator that encodes the dynamics of facial emotion into a multi-emotion-class latent space represented as a mixture distribution. The generation process commences with random sampling from the modeled distribution, guided by the accompanying audio, enabling both lip-synchronization and the uncertain nonverbal facial cues generation. Furthermore, our designed vector-quantization image generator treats the creation of expressive facial images as a code query task, utilizing a learned codebook to provide rich, high-quality textures that enhance the emotional perception of the results. Extensive experiments are conducted to showcase the effectiveness of our approach.
Quantization has emerged as a promising technique for improving the memory and computational efficiency of large language models (LLMs). Though the trade-off between performance and efficiency is well-known, there is still much to be learned about the relationship between quantization and LLM performance. To shed light on this relationship, we propose a new perspective on quantization, viewing it as perturbations added to the weights and activations of LLMs. We call this approach "the lens of perturbation". Using this lens, we conduct experiments with various artificial perturbations to explore their impact on LLM performance. Our findings reveal several connections between the properties of perturbations and LLM performance, providing insights into the failure cases of uniform quantization and suggesting potential solutions to improve the robustness of LLM quantization. To demonstrate the significance of our findings, we implement a simple non-uniform quantization approach based on our insights. Our experiments show that this approach achieves minimal performance degradation on both 4-bit weight quantization and 8-bit quantization for weights and activations. These results validate the correctness of our approach and highlight its potential to improve the efficiency of LLMs without sacrificing performance.
Transformer-based models have made remarkable advancements in various NLP areas. Nevertheless, these models often exhibit vulnerabilities when confronted with adversarial attacks. In this paper, we explore the effect of quantization on the robustness of Transformer-based models. Quantization usually involves mapping a high-precision real number to a lower-precision value, aiming at reducing the size of the model at hand. To the best of our knowledge, this work is the first application of quantization on the robustness of NLP models. In our experiments, we evaluate the impact of quantization on BERT and DistilBERT models in text classification using SST-2, Emotion, and MR datasets. We also evaluate the performance of these models against TextFooler, PWWS, and PSO adversarial attacks. Our findings show that quantization significantly improves (by an average of 18.68%) the adversarial accuracy of the models. Furthermore, we compare the effect of quantization versus that of the adversarial training approach on robustness. Our experiments indicate that quantization increases the robustness of the model by 18.80% on average compared to adversarial training without imposing any extra computational overhead during training. Therefore, our results highlight the effectiveness of quantization in improving the robustness of NLP models.
Recent advancements in Vision-Language (VL) models have sparked interest in their deployment on edge devices, yet challenges in handling diverse visual modalities, manual annotation, and computational constraints remain. We introduce EdgeVL, a novel framework that bridges this gap by seamlessly integrating dual-modality knowledge distillation and quantization-aware contrastive learning. This approach enables the adaptation of large VL models, like CLIP, for efficient use with both RGB and non-RGB images on resource-limited devices without the need for manual annotations. EdgeVL not only transfers visual language alignment capabilities to compact models but also maintains feature quality post-quantization, significantly enhancing open-vocabulary classification performance across various visual modalities. Our work represents the first systematic effort to adapt large VL models for edge deployment, showcasing up to 15.4% accuracy improvements on multiple datasets and up to 93-fold reduction in model size.
Key-value (KV) caching has become the de-facto to accelerate generation speed for large language models (LLMs) inference. However, the growing cache demand with increasing sequence length has transformed LLM inference to be a memory bound problem, significantly constraining the system throughput. Existing methods rely on dropping unimportant tokens or quantizing all entries uniformly. Such methods, however, often incur high approximation errors to represent the compressed matrices. The autoregressive decoding process further compounds the error of each step, resulting in critical deviation in model generation and deterioration of performance. To tackle this challenge, we propose GEAR, an efficient KV cache compression framework that achieves near-lossless high-ratio compression. GEAR first applies quantization to majority of entries of similar magnitudes to ultra-low precision. It then employs a low rank matrix to approximate the quantization error, and a sparse matrix to remedy individual errors from outlier entries. By adeptly integrating three techniques, GEAR is able to fully exploit their synergistic potentials. Our experiments demonstrate that compared to alternatives, GEAR achieves near-lossless 4-bit KV cache compression with up to 2.38x throughput improvement, while reducing peak-memory size up to 2.29x. Our code is publicly available at https://github.com/HaoKang-Timmy/GEAR.
Traditional Deep Neural Network (DNN) quantization methods using integer, fixed-point, or floating-point data types struggle to capture diverse DNN parameter distributions at low precision, and often require large silicon overhead and intensive quantization-aware training. In this study, we introduce Logarithmic Posits (LP), an adaptive, hardware-friendly data type inspired by posits that dynamically adapts to DNN weight/activation distributions by parameterizing LP bit fields. We also develop a novel genetic-algorithm based framework, LP Quantization (LPQ), to find optimal layer-wise LP parameters while reducing representational divergence between quantized and full-precision models through a novel global-local contrastive objective. Additionally, we design a unified mixed-precision LP accelerator (LPA) architecture comprising of processing elements (PEs) incorporating LP in the computational datapath. Our algorithm-hardware co-design demonstrates on average <1% drop in top-1 accuracy across various CNN and ViT models. It also achieves ~ 2x improvements in performance per unit area and 2.2x gains in energy efficiency compared to state-of-the-art quantization accelerators using different data types.
To enhance the interpretability of multimodal unified representations, many studies have focused on discrete unified representations. These efforts typically start with contrastive learning and gradually extend to the disentanglement of modal information, achieving solid multimodal discrete unified representations. However, existing research often overlooks two critical issues: 1) The use of Euclidean distance for quantization in discrete representations often overlooks the important distinctions among different dimensions of features, resulting in redundant representations after quantization; 2) Different modalities have unique characteristics, and a uniform alignment approach does not fully exploit these traits. To address these issues, we propose Training-free Optimization of Codebook (TOC) and Fine and Coarse cross-modal Information Disentangling (FCID). These methods refine the unified discrete representations from pretraining and perform fine- and coarse-grained information disentanglement tailored to the specific characteristics of each modality, achieving significant performance improvements over previous state-of-the-art models.
The emergence of LLMs has ignited a fresh surge of breakthroughs in NLP applications, particularly in domains such as question-answering systems and text generation. As the need for longer context grows, a significant bottleneck in model deployment emerges due to the linear expansion of the Key-Value (KV) cache with the context length. Existing methods primarily rely on various hypotheses, such as sorting the KV cache based on attention scores for replacement or eviction, to compress the KV cache and improve model throughput. However, heuristics used by these strategies may wrongly evict essential KV cache, which can significantly degrade model performance. In this paper, we propose QAQ, a Quality Adaptive Quantization scheme for the KV cache. We theoretically demonstrate that key cache and value cache exhibit distinct sensitivities to quantization, leading to the formulation of separate quantization strategies for their non-uniform quantization. Through the integration of dedicated outlier handling, as well as an improved attention-aware approach, QAQ achieves up to 10x the compression ratio of the KV cache size with a neglectable impact on model performance. QAQ significantly reduces the practical hurdles of deploying LLMs, opening up new possibilities for longer-context applications. The code is available at github.com/ClubieDong/KVCacheQuantization.
In Distributed optimization and Learning, and even more in the modern framework of federated learning, communication, which is slow and costly, is critical. We introduce LoCoDL, a communication-efficient algorithm that leverages the two popular and effective techniques of Local training, which reduces the communication frequency, and Compression, in which short bitstreams are sent instead of full-dimensional vectors of floats. LoCoDL works with a large class of unbiased compressors that includes widely-used sparsification and quantization methods. LoCoDL provably benefits from local training and compression and enjoys a doubly-accelerated communication complexity, with respect to the condition number of the functions and the model dimension, in the general heterogenous regime with strongly convex functions. This is confirmed in practice, with LoCoDL outperforming existing algorithms.
Generative Artificial Intelligence (GAI) shows remarkable productivity and creativity in Mobile Edge Networks, such as the metaverse and the Industrial Internet of Things. Federated learning is a promising technique for effectively training GAI models in mobile edge networks due to its data distribution. However, there is a notable issue with communication consumption when training large GAI models like generative diffusion models in mobile edge networks. Additionally, the substantial energy consumption associated with training diffusion-based models, along with the limited resources of edge devices and complexities of network environments, pose challenges for improving the training efficiency of GAI models. To address this challenge, we propose an on-demand quantized energy-efficient federated diffusion approach for mobile edge networks. Specifically, we first design a dynamic quantized federated diffusion training scheme considering various demands from the edge devices. Then, we study an energy efficiency problem based on specific quantization requirements. Numerical results show that our proposed method significantly reduces system energy consumption and transmitted model size compared to both baseline federated diffusion and fixed quantized federated diffusion methods while effectively maintaining reasonable quality and diversity of generated data.
An integrate-and-fire time-encoding machine (IF-TEM) is an effective asynchronous sampler that translates amplitude information into non-uniform time sequences. In this work, we propose a novel Adaptive IF-TEM (AIF-TEM) approach. This design dynamically adjusts the TEM's sensitivity to changes in the input signal's amplitude and frequency in real-time. We provide a comprehensive analysis of AIF-TEM's oversampling and distortion properties. By the adaptive adjustments, AIF-TEM as we show can achieve significant performance improvements in terms of sampling rate-distortion in a practical finite regime. We demonstrate empirically that in the scenarios tested AIF-TEM outperforms classical IF-TEM and traditional Nyquist (i.e., periodic) sampling methods for band-limited signals. In terms of Mean Square Error (MSE), the reduction reaches at least 12dB (fixing the oversampling rate). Additionally, we investigate the quantization process for AIF-TEM and analyze the quantization MSE bound. Empirical results show that classic quantization for AIF-TEM improves performance by at least 14 dB compared to IF-TEM. We introduce a dynamic quantization technique for AIF-TEM, which further improves performance compared to classic quantization. Empirically, this reduction reaches at least 10 dB compared to classic quantization for AIF-TEM.
In this paper, we examine the role of stochastic quantizers for privacy preservation. We first employ a static stochastic quantizer and investigate its corresponding privacy-preserving properties. Specifically, we demonstrate that a sufficiently large quantization step guarantees $(0, \delta)$ differential privacy. Additionally, the degradation of control performance caused by quantization is evaluated as the tracking error of output regulation. These two analyses characterize the trade-off between privacy and control performance, determined by the quantization step. This insight enables us to use quantization intentionally as a means to achieve the seemingly conflicting two goals of maintaining control performance and preserving privacy at the same time; towards this end, we further investigate a dynamic stochastic quantizer. Under a stability assumption, the dynamic stochastic quantizer can enhance privacy, more than the static one, while achieving the same control performance. We further handle the unstable case by additionally applying input Gaussian noise.
As Large Language Models (LLMs) continue to advance in performance, their size has escalated significantly, with current LLMs containing billions or even trillions of parameters. However, in this study, we discovered that many layers of LLMs exhibit high similarity, and some layers play a negligible role in network functionality. Based on this observation, we define a metric called Block Influence (BI) to gauge the significance of each layer in LLMs. We then propose a straightforward pruning approach: layer removal, in which we directly delete the redundant layers in LLMs based on their BI scores. Experiments demonstrate that our method, which we call ShortGPT, significantly outperforms previous state-of-the-art (SOTA) methods in model pruning. Moreover, ShortGPT is orthogonal to quantization-like methods, enabling further reduction in parameters and computation. The ability to achieve better results through simple layer removal, as opposed to more complex pruning techniques, suggests a high degree of redundancy in the model architecture.