Prompt Tuning has been largely successful as a parameter-efficient method of conditioning large-scale pre-trained language models to perform downstream tasks. Thus far, soft prompt tuning learns a fixed set of task-specific continuous vectors, i.e., soft tokens that remain static across the task samples. A fixed prompt, however, may not generalize well to the diverse kinds of inputs the task comprises. In order to address this, we propose Vector-quantized Input-contextualized Prompts (VIP) as an extension to the soft prompt tuning framework. VIP particularly focuses on two aspects -- contextual prompts that learns input-specific contextualization of the soft prompt tokens through a small-scale sentence encoder and quantized prompts that maps the contextualized prompts to a set of learnable codebook vectors through a Vector quantization network. On various language understanding tasks like SuperGLUE, QA, Relation classification, NER and NLI, VIP outperforms the soft prompt tuning (PT) baseline by an average margin of 1.19%. Further, our generalization studies show that VIP learns more robust prompt representations, surpassing PT by a margin of 0.6% - 5.3% on Out-of-domain QA and NLI tasks respectively, and by 0.75% on Multi-Task setup over 4 tasks spanning across 12 domains.
Signal classification problems arise in a wide variety of applications, and their demand is only expected to grow. In this paper, we focus on the wireless sensor network signal classification setting, where each sensor forwards quantized signals to a fusion center to be classified. Our primary goal is to train a decision function and quantizers across the sensors to maximize the classification performance in an online manner. Moreover, we are interested in sparse sensor selection using a marginalized weighted kernel approach to improve network resource efficiency by disabling less reliable sensors with minimal effect on classification performance.To achieve our goals, we develop a multi-sensor online kernel scalar quantization (MSOKSQ) learning strategy that operates on the sensor outputs at the fusion center. Our theoretical analysis reveals how the proposed algorithm affects the quantizers across the sensors. Additionally, we provide a convergence analysis of our online learning approach by studying its relationship to batch learning. We conduct numerical studies under different classification and sensor network settings which demonstrate the accuracy gains from optimizing different components of MSOKSQ and robustness to reduction in the number of sensors selected.
Designing efficient sparse recovery algorithms that could handle noisy quantized measurements is important in a variety of applications -- from radar to source localization, spectrum sensing and wireless networking. We take advantage of the approximate message passing (AMP) framework to achieve this goal given its high computational efficiency and state-of-the-art performance. In AMP, the signal of interest is assumed to follow certain prior distribution with unknown parameters. Previous works focused on finding the parameters that maximize the measurement likelihood via expectation maximization -- an increasingly difficult problem to solve in cases involving complicated probability models. In this paper, we treat the parameters as unknown variables and compute their posteriors via AMP. The parameters and signal of interest can then be jointly recovered. Compared to previous methods, the proposed approach leads to a simple and elegant parameter estimation scheme, allowing us to directly work with 1-bit quantization noise model. We then further extend our approach to general multi-bit quantization noise model. Experimental results show that the proposed framework provides significant improvement over state-of-the-art methods across a wide range of sparsity and noise levels.
Large Deep Neural Networks (DNNs) are the backbone of today's artificial intelligence due to their ability to make accurate predictions when being trained on huge datasets. With advancing technologies, such as the Internet of Things, interpreting large quantities of data generated by sensors is becoming an increasingly important task. However, in many applications not only the predictive performance but also the energy consumption of deep learning models is of major interest. This paper investigates the efficient deployment of deep learning models on resource-constrained microcontroller architectures via network compression. We present a methodology for the systematic exploration of different DNN pruning, quantization, and deployment strategies, targeting different ARM Cortex-M based low-power systems. The exploration allows to analyze trade-offs between key metrics such as accuracy, memory consumption, execution time, and power consumption. We discuss experimental results on three different DNN architectures and show that we can compress them to below 10\% of their original parameter count before their predictive quality decreases. This also allows us to deploy and evaluate them on Cortex-M based microcontrollers.
As Deep Neural Networks (DNNs) usually are overparameterized and have millions of weight parameters, it is challenging to deploy these large DNN models on resource-constrained hardware platforms, e.g., smartphones. Numerous network compression methods such as pruning and quantization are proposed to reduce the model size significantly, of which the key is to find suitable compression allocation (e.g., pruning sparsity and quantization codebook) of each layer. Existing solutions obtain the compression allocation in an iterative/manual fashion while finetuning the compressed model, thus suffering from the efficiency issue. Different from the prior art, we propose a novel One-shot Pruning-Quantization (OPQ) in this paper, which analytically solves the compression allocation with pre-trained weight parameters only. During finetuning, the compression module is fixed and only weight parameters are updated. To our knowledge, OPQ is the first work that reveals pre-trained model is sufficient for solving pruning and quantization simultaneously, without any complex iterative/manual optimization at the finetuning stage. Furthermore, we propose a unified channel-wise quantization method that enforces all channels of each layer to share a common codebook, which leads to low bit-rate allocation without introducing extra overhead brought by traditional channel-wise quantization. Comprehensive experiments on ImageNet with AlexNet/MobileNet-V1/ResNet-50 show that our method improves accuracy and training efficiency while obtains significantly higher compression rates compared to the state-of-the-art.
Dense retrieval overcome the lexical gap and has shown great success in ad-hoc information retrieval (IR). Despite their success, dense retrievers are expensive to serve across practical use cases. For use cases requiring to search from millions of documents, the dense index becomes bulky and requires high memory usage for storing the index. More recently, learning-to-hash (LTH) techniques, for e.g., BPR and JPQ, produce binary document vectors, thereby reducing the memory requirement to efficiently store the dense index. LTH techniques are supervised and finetune the retriever using a ranking loss. They outperform their counterparts, i.e., traditional out-of-the-box vector compression techniques such as PCA or PQ. A missing piece from prior work is that existing techniques have been evaluated only in-domain, i.e., on a single dataset such as MS MARCO. In our work, we evaluate LTH and vector compression techniques for improving the downstream zero-shot retrieval accuracy of the TAS-B dense retriever while maintaining efficiency at inference. Our results demonstrate that, unlike prior work, LTH strategies when applied naively can underperform the zero-shot TAS-B dense retriever on average by up to 14% nDCG@10 on the BEIR benchmark. To solve this limitation, in our work, we propose an easy yet effective solution of injecting domain adaptation with existing supervised LTH techniques. We experiment with two well-known unsupervised domain adaptation techniques: GenQ and GPL. Our domain adaptation injection technique can improve the downstream zero-shot retrieval effectiveness for both BPR and JPQ variants of the TAS-B model by on average 11.5% and 8.2% nDCG@10 while both maintaining 32$\times$ memory efficiency and 14$\times$ and 2$\times$ speedup respectively in CPU retrieval latency on BEIR. All our code, models, and data are publicly available at https://github.com/thakur-nandan/income.
Federated learning (FL) over mobile devices has fostered numerous intriguing applications/services, many of which are delay-sensitive. In this paper, we propose a service delay efficient FL (SDEFL) scheme over mobile devices. Unlike traditional communication efficient FL, which regards wireless communications as the bottleneck, we find that under many situations, the local computing delay is comparable to the communication delay during the FL training process, given the development of high-speed wireless transmission techniques. Thus, the service delay in FL should be computing delay + communication delay over training rounds. To minimize the service delay of FL, simply reducing local computing/communication delay independently is not enough. The delay trade-off between local computing and wireless communications must be considered. Besides, we empirically study the impacts of local computing control and compression strategies (i.e., the number of local updates, weight quantization, and gradient quantization) on computing, communication and service delays. Based on those trade-off observation and empirical studies, we develop an optimization scheme to minimize the service delay of FL over heterogeneous devices. We establish testbeds and conduct extensive emulations/experiments to verify our theoretical analysis. The results show that SDEFL reduces notable service delay with a small accuracy drop compared to peer designs.
As the machine learning and systems community strives to achieve higher energy-efficiency through custom DNN accelerators and model compression techniques, there is a need for a design space exploration framework that incorporates quantization-aware processing elements into the accelerator design space while having accurate and fast power, performance, and area models. In this work, we present QAPPA, a highly parameterized quantization-aware power, performance, and area modeling framework for DNN accelerators. Our framework can facilitate the future research on design space exploration of DNN accelerators for various design choices such as bit precision, processing element type, scratchpad sizes of processing elements, global buffer size, device bandwidth, number of total processing elements in the the design, and DNN workloads. Our results show that different bit precisions and processing element types lead to significant differences in terms of performance per area and energy. Specifically, our proposed lightweight processing elements achieve up to 4.9x more performance per area and energy improvement when compared to INT16 based implementation.
In this paper, we study the problem of making brighter images from dark images found in the wild. The images are dark because they are taken in dim environments. They suffer from color shifts caused by quantization and from sensor noise. We don't know the true camera reponse function for such images and they are not RAW. We use a supervised learning method, relying on a straightforward simulation of an imaging pipeline to generate usable dataset for training and testing. On a number of standard datasets, our approach outperforms the state of the art quantitatively. Qualitative comparisons suggest strong improvements in reconstruction accuracy.
Recently, much attention has been drawn to learning the underlying 3D structures of a scene from monocular videos in a fully self-supervised fashion. One of the most challenging aspects of this task is handling the independently moving objects as they break the rigid-scene assumption. For the first time, we show that pixel positional information can be exploited to learn SVDE (Single View Depth Estimation) from videos. Our proposed moving object (MO) masks, which are induced by shifted positional information (SPI) and referred to as `SPIMO' masks, are very robust and consistently remove the independently moving objects in the scenes, allowing for better learning of SVDE from videos. Additionally, we introduce a new adaptive quantization scheme that assigns the best per-pixel quantization curve for our depth discretization. Finally, we employ existing boosting techniques in a new way to further self-supervise the depth of the moving objects. With these features, our pipeline is robust against moving objects and generalizes well to high-resolution images, even when trained with small patches, yielding state-of-the-art (SOTA) results with almost 8.5x fewer parameters than the previous works that learn from videos. We present extensive experiments on KITTI and CityScapes that show the effectiveness of our method.
Recent advancements in machine learning achieved by Deep Neural Networks (DNNs) have been significant. While demonstrating high accuracy, DNNs are associated with a huge number of parameters and computations, which leads to high memory usage and energy consumption. As a result, deploying DNNs on devices with constrained hardware resources poses significant challenges. To overcome this, various compression techniques have been widely employed to optimize DNN accelerators. A promising approach is quantization, in which the full-precision values are stored in low bit-width precision. Quantization not only reduces memory requirements but also replaces high-cost operations with low-cost ones. DNN quantization offers flexibility and efficiency in hardware design, making it a widely adopted technique in various methods. Since quantization has been extensively utilized in previous works, there is a need for an integrated report that provides an understanding, analysis, and comparison of different quantization approaches. Consequently, we present a comprehensive survey of quantization concepts and methods, with a focus on image classification. We describe clustering-based quantization methods and explore the use of a scale factor parameter for approximating full-precision values. Moreover, we thoroughly review the training of a quantized DNN, including the use of a straight-through estimator and quantization regularization. We explain the replacement of floating-point operations with low-cost bitwise operations in a quantized DNN and the sensitivity of different layers in quantization. Furthermore, we highlight the evaluation metrics for quantization methods and important benchmarks in the image classification task. We also present the accuracy of the state-of-the-art methods on CIFAR-10 and ImageNet.
One noted issue of vector-quantized variational autoencoder (VQ-VAE) is that the learned discrete representation uses only a fraction of the full capacity of the codebook, also known as codebook collapse. We hypothesize that the training scheme of VQ-VAE, which involves some carefully designed heuristics, underlies this issue. In this paper, we propose a new training scheme that extends the standard VAE via novel stochastic dequantization and quantization, called stochastically quantized variational autoencoder (SQ-VAE). In SQ-VAE, we observe a trend that the quantization is stochastic at the initial stage of the training but gradually converges toward a deterministic quantization, which we call self-annealing. Our experiments show that SQ-VAE improves codebook utilization without using common heuristics. Furthermore, we empirically show that SQ-VAE is superior to VAE and VQ-VAE in vision- and speech-related tasks.
In this paper, we investigate how field programmable gate arrays can serve as hardware accelerators for real-time semantic segmentation tasks relevant for autonomous driving. Considering compressed versions of the ENet convolutional neural network architecture, we demonstrate a fully-on-chip deployment with a latency of 4.9 ms per image, using less than 30% of the available resources on a Xilinx ZCU102 evaluation board. The latency is reduced to 3 ms per image when increasing the batch size to ten, corresponding to the use case where the autonomous vehicle receives inputs from multiple cameras simultaneously. We show, through aggressive filter reduction and heterogeneous quantization-aware training, and an optimized implementation of convolutional layers, that the power consumption and resource utilization can be significantly reduced while maintaining accuracy on the Cityscapes dataset.
Spiking neural networks (SNNs) operating with asynchronous discrete events show higher energy efficiency with sparse computation. A popular approach for implementing deep SNNs is ANN-SNN conversion combining both efficient training of ANNs and efficient inference of SNNs. However, the accuracy loss is usually non-negligible, especially under a few time steps, which restricts the applications of SNN on latency-sensitive edge devices greatly. In this paper, we first identify that such performance degradation stems from the misrepresentation of the negative or overflow residual membrane potential in SNNs. Inspired by this, we decompose the conversion error into three parts: quantization error, clipping error, and residual membrane potential representation error. With such insights, we propose a two-stage conversion algorithm to minimize those errors respectively. Besides, We show each stage achieves significant performance gains in a complementary manner. By evaluating on challenging datasets including CIFAR-10, CIFAR- 100 and ImageNet, the proposed method demonstrates the state-of-the-art performance in terms of accuracy, latency and energy preservation. Furthermore, our method is evaluated using a more challenging object detection task, revealing notable gains in regression performance under ultra-low latency when compared to existing spike-based detection algorithms. Codes are available at https://github.com/Windere/snn-cvt-dual-phase.
In federated learning (FL), it is commonly assumed that all data are placed at clients in the beginning of machine learning (ML) optimization (i.e., offline learning). However, in many real-world applications, it is expected to proceed in an online fashion. To this end, online FL (OFL) has been introduced, which aims at learning a sequence of global models from decentralized streaming data such that the so-called cumulative regret is minimized. Combining online gradient descent and model averaging, in this framework, FedOGD is constructed as the counterpart of FedSGD in FL. While it can enjoy an optimal sublinear regret, FedOGD suffers from heavy communication costs. In this paper, we present a communication-efficient method (named OFedIQ) by means of intermittent transmission (enabled by client subsampling and periodic transmission) and quantization. For the first time, we derive the regret bound that captures the impact of data-heterogeneity and the communication-efficient techniques. Through this, we efficiently optimize the parameters of OFedIQ such as sampling rate, transmission period, and quantization levels. Also, it is proved that the optimized OFedIQ can asymptotically achieve the performance of FedOGD while reducing the communication costs by 99%. Via experiments with real datasets, we demonstrate the effectiveness of the optimized OFedIQ.
Analog mixed-signal (AMS) devices promise faster, more energy-efficient deep neural network (DNN) inference than their digital counterparts. However, recent studies show that DNNs on AMS devices with fixed-point numbers can incur an accuracy penalty because of precision loss. To mitigate this penalty, we present a novel AMS-compatible adaptive block floating-point (ABFP) number representation. We also introduce amplification (or gain) as a method for increasing the accuracy of the number representation without increasing the bit precision of the output. We evaluate the effectiveness of ABFP on the DNNs in the MLPerf datacenter inference benchmark -- realizing less than $1\%$ loss in accuracy compared to FLOAT32. We also propose a novel method of finetuning for AMS devices, Differential Noise Finetuning (DNF), which samples device noise to speed up finetuning compared to conventional Quantization-Aware Training.
Although generative facial prior and geometric prior have recently demonstrated high-quality results for blind face restoration, producing fine-grained facial details faithful to inputs remains a challenging problem. Motivated by the classical dictionary-based methods and the recent vector quantization (VQ) technique, we propose a VQ-based face restoration method - VQFR. VQFR takes advantage of high-quality low-level feature banks extracted from high-quality faces and can thus help recover realistic facial details. However, the simple application of the VQ codebook cannot achieve good results with faithful details and identity preservation. Therefore, we further introduce two special network designs. 1). We first investigate the compression patch size in the VQ codebook and find that the VQ codebook designed with a proper compression patch size is crucial to balance the quality and fidelity. 2). To further fuse low-level features from inputs while not "contaminating" the realistic details generated from the VQ codebook, we proposed a parallel decoder consisting of a texture decoder and a main decoder. Those two decoders then interact with a texture warping module with deformable convolution. Equipped with the VQ codebook as a facial detail dictionary and the parallel decoder design, the proposed VQFR can largely enhance the restored quality of facial details while keeping the fidelity to previous methods.
Orthogonal Frequency Division Multiplexing (OFDM)-based waveforms are used for communication links in many current and emerging Internet of Things (IoT) applications, including the latest WiFi standards. For such OFDM-based transceivers, many core physical layer functions related to channel estimation, demapping, and decoding are implemented for specific choices of channel types and modulation schemes, among others. To decouple hard-wired choices from the receiver chain and thereby enhance the flexibility of IoT deployment in many novel scenarios without changing the underlying hardware, we explore a novel, modular Machine Learning (ML)-based receiver chain design. Here, ML blocks replace the individual processing blocks of an OFDM receiver, and we specifically describe this swapping for the legacy channel estimation, symbol demapping, and decoding blocks with Neural Networks (NNs). A unique aspect of this modular design is providing flexible allocation of processing functions to the legacy or ML blocks, allowing them to interchangeably coexist. Furthermore, we study the implementation cost-benefits of the proposed NNs in resource-constrained IoT devices through pruning and quantization, as well as emulation of these compressed NNs within Field Programmable Gate Arrays (FPGAs). Our evaluations demonstrate that the proposed modular NN-based receiver improves bit error rate of the traditional non-ML receiver by averagely 61% and 10% for the simulated and over-the-air datasets, respectively. We further show complexity-performance tradeoffs by presenting computational complexity comparisons between the traditional algorithms and the proposed compressed NNs.
Building on our recent work on {\em neuromimetic control theory}, new results on resilience and neuro-inspired quantization are reported. The term neuromimetic refers to the models having features that are characteristic of the neurobiology of biological motor control. As in previous work, the focus is on what we call {\em overcomplete} linear systems that are characterized by larger numbers of input and output channels than the dimensions of the state. The specific contributions of the present paper include a proposed {\em resilient} observer whose operation tolerates output channel intermittency and even complete dropouts. Tying these ideas together with our previous work on resilient stability, a resilient separation principle is established. We also propose a {\em principled quantization} in which control signals are encoded as simple discrete inputs which act collectively through the many channels of input that are the hallmarks of the overcomplete models. Aligned with the neuromimetic paradigm, an {\em emulation} problem is proposed and this in turn defines an optimal quantization problem. Several possible solutions are discussed including direct combinatorial optimization, a Hebbian-like iterative learning algorithm, and a deep Q-learning (DQN) approach. For the problems being considered, machine learning approaches to optimization provide valuable insights regarding comparisons between optimal and nearby suboptimal solutions. These are useful in understanding the kinds of resilience to intermittency and channel dropouts that were earlier demonstrated for continuous systems.
With the advent of ubiquitous deployment of smart devices and the Internet of Things, data sources for machine learning inference have increasingly moved to the edge of the network. Existing machine learning inference platforms typically assume a homogeneous infrastructure and do not take into account the more complex and tiered computing infrastructure that includes edge devices, local hubs, edge datacenters, and cloud datacenters. On the other hand, recent AutoML efforts have provided viable solutions for model compression, pruning and quantization for heterogeneous environments; for a machine learning model, now we may easily find or even generate a series of models with different tradeoffs between accuracy and efficiency. We design and implement JellyBean, a system for serving and optimizing machine learning inference workflows on heterogeneous infrastructures. Given service-level objectives (e.g., throughput, accuracy), JellyBean picks the most cost-efficient models that meet the accuracy target and decides how to deploy them across different tiers of infrastructures. Evaluations show that JellyBean reduces the total serving cost of visual question answering by up to 58%, and vehicle tracking from the NVIDIA AI City Challenge by up to 36% compared with state-of-the-art model selection and worker assignment solutions. JellyBean also outperforms prior ML serving systems (e.g., Spark on the cloud) up to 5x in serving costs.