In hash-based image retrieval systems, degraded or transformed inputs usually generate different codes from the original, deteriorating the retrieval accuracy. To mitigate this issue, data augmentation can be applied during training. However, even if augmented samples of an image are similar in real feature space, the quantization can scatter them far away in Hamming space. This results in representation discrepancies that can impede training and degrade performance. In this work, we propose a novel self-distilled hashing scheme to minimize the discrepancy while exploiting the potential of augmented data. By transferring the hash knowledge of the weakly-transformed samples to the strong ones, we make the hash code insensitive to various transformations. We also introduce hash proxy-based similarity learning and binary cross entropy-based quantization loss to provide fine quality hash codes. Ultimately, we construct a deep hashing framework that not only improves the existing deep hashing approaches, but also achieves the state-of-the-art retrieval results. Extensive experiments are conducted and confirm the effectiveness of our work.
Accelerating the neural network inference by FPGA has emerged as a popular option, since the reconfigurability and high performance computing capability of FPGA intrinsically satisfies the computation demand of the fast-evolving neural algorithms. However, the popular neural accelerators on FPGA (e.g., Xilinx DPU) mainly utilize the DSP resources for constructing their processing units, while the rich LUT resources are not well exploited. Via the software-hardware co-design approach, in this work, we develop an FPGA-based heterogeneous computing system for neural network acceleration. From the hardware perspective, the proposed accelerator consists of DSP- and LUT-based GEneral Matrix-Multiplication (GEMM) computing cores, which forms the entire computing system in a heterogeneous fashion. The DSP- and LUT-based GEMM cores are computed w.r.t a unified Instruction Set Architecture (ISA) and unified buffers. Along the data flow of the neural network inference path, the computation of the convolution/fully-connected layer is split into two portions, handled by the DSP- and LUT-based GEMM cores asynchronously. From the software perspective, we mathematically and systematically model the latency and resource utilization of the proposed heterogeneous accelerator, regarding varying system design configurations. Through leveraging the reinforcement learning technique, we construct a framework to achieve end-to-end selection and optimization of the design specification of target heterogeneous accelerator, including workload split strategy, mixed-precision quantization scheme, and resource allocation of DSP- and LUT-core. In virtue of the proposed design framework and heterogeneous computing system, our design outperforms the state-of-the-art Mix&Match design with latency reduced by 1.12-1.32x with higher inference accuracy. The N3H-core is open-sourced at: https://github.com/elliothe/N3H_Core.
All Lossy compression algorithms employ similar compression schemes -- frequency domain transform followed by quantization and lossless encoding schemes. They target tradeoffs by quantizating high frequency data to increase compression rates which come at the cost of higher image distortion. We propose a new method of optimizing quantization tables using Deep Learning and a minimax loss function that more accurately measures the tradeoffs between rate and distortion parameters (RD) than previous methods. We design a convolutional neural network (CNN) that learns a mapping between image blocks and quantization tables in an unsupervised manner. By processing images across all channels at once, we can achieve stronger performance by also measuring tradeoffs in information loss between different channels. We initially target optimization on JPEG images but feel that this can be expanded to any lossy compressor.
As neural networks have become more powerful, there has been a rising desire to deploy them in the real world; however, the power and accuracy of neural networks is largely due to their depth and complexity, making them difficult to deploy, especially in resource-constrained devices. Neural network quantization has recently arisen to meet this demand of reducing the size and complexity of neural networks by reducing the precision of a network. With smaller and simpler networks, it becomes possible to run neural networks within the constraints of their target hardware. This paper surveys the many neural network quantization techniques that have been developed in the last decade. Based on this survey and comparison of neural network quantization techniques, we propose future directions of research in the area.
Accurate and efficient pedestrian detection is crucial for the intelligent transportation system regarding pedestrian safety and mobility, e.g., Advanced Driver Assistance Systems, and smart pedestrian crosswalk systems. Among all pedestrian detection methods, vision-based detection method is demonstrated to be the most effective in previous studies. However, the existing vision-based pedestrian detection algorithms still have two limitations that restrict their implementations, those being real-time performance as well as the resistance to the impacts of environmental factors, e.g., low illumination conditions. To address these issues, this study proposes a lightweight Illumination and Temperature-aware Multispectral Network (IT-MN) for accurate and efficient pedestrian detection. The proposed IT-MN is an efficient one-stage detector. For accommodating the impacts of environmental factors and enhancing the sensing accuracy, thermal image data is fused by the proposed IT-MN with visual images to enrich useful information when visual image quality is limited. In addition, an innovative and effective late fusion strategy is also developed to optimize the image fusion performance. To make the proposed model implementable for edge computing, the model quantization is applied to reduce the model size by 75% while shortening the inference time significantly. The proposed algorithm is evaluated by comparing with the selected state-of-the-art algorithms using a public dataset collected by in-vehicle cameras. The results show that the proposed algorithm achieves a low miss rate and inference time at 14.19% and 0.03 seconds per image pair on GPU. Besides, the quantized IT-MN achieves an inference time of 0.21 seconds per image pair on the edge device, which also demonstrates the potentiality of deploying the proposed model on edge devices as a highly efficient pedestrian detection algorithm.
With the rapid increase of big data, distributed Machine Learning (ML) has been widely applied in training large-scale models. Stochastic Gradient Descent (SGD) is arguably the workhorse algorithm of ML. Distributed ML models trained by SGD involve large amounts of gradient communication, which limits the scalability of distributed ML. Thus, it is important to compress the gradients for reducing communication. In this paper, we propose FastSGD, a Fast compressed SGD framework for distributed ML. To achieve a high compression ratio at a low cost, FastSGD represents the gradients as key-value pairs, and compresses both the gradient keys and values in linear time complexity. For the gradient value compression, FastSGD first uses a reciprocal mapper to transform original values into reciprocal values, and then, it utilizes a logarithm quantization to further reduce reciprocal values to small integers. Finally, FastSGD filters reduced gradient integers by a given threshold. For the gradient key compression, FastSGD provides an adaptive fine-grained delta encoding method to store gradient keys with fewer bits. Extensive experiments on practical ML models and datasets demonstrate that FastSGD achieves the compression ratio up to 4 orders of magnitude, and accelerates the convergence time up to 8x, compared with state-of-the-art methods.
Recently Implicit Neural Representations (INRs) gained attention as a novel and effective representation for various data types. Thus far, prior work mostly focused on optimizing their reconstruction performance. This work investigates INRs from a novel perspective, i.e., as a tool for image compression. To this end, we propose the first comprehensive compression pipeline based on INRs including quantization, quantization-aware retraining and entropy coding. Encoding with INRs, i.e. overfitting to a data sample, is typically orders of magnitude slower. To mitigate this drawback, we leverage meta-learned initializations based on MAML to reach the encoding in fewer gradient updates which also generally improves rate-distortion performance of INRs. We find that our approach to source compression with INRs vastly outperforms similar prior work, is competitive with common compression algorithms designed specifically for images and closes the gap to state-of-the-art learned approaches based on Rate-Distortion Autoencoders. Moreover, we provide an extensive ablation study on the importance of individual components of our method which we hope facilitates future research on this novel approach to image compression.
More and more orgainizations and institutions make efforts on using external data to improve the performance of AI services. To address the data privacy and security concerns, federated learning has attracted increasing attention from both academia and industry to securely construct AI models across multiple isolated data providers. In this paper, we studied the efficiency problem of adapting widely used XGBoost model in real-world applications to vertical federated learning setting. State-of-the-art vertical federated XGBoost frameworks requires large number of encryption operations and ciphertext transmissions, which makes the model training much less efficient than training XGBoost models locally. To bridge this gap, we proposed a novel batch homomorphic encryption method to cut the cost of encryption-related computation and transmission in nearly half. This is achieved by encoding the first-order derivative and the second-order derivative into a single number for encryption, ciphertext transmission, and homomorphic addition operations. The sum of multiple first-order derivatives and second-order derivatives can be simultaneously decoded from the sum of encoded values. We are motivated by the batch idea in the work of BatchCrypt for horizontal federated learning, and design a novel batch method to address the limitations of allowing quite few number of negative numbers. The encode procedure of the proposed batch method consists of four steps, including shifting, truncating, quantizing and batching, while the decoding procedure consists of de-quantization and shifting back. The advantages of our method are demonstrated through theoretical analysis and extensive numerical experiments.
Binary networks are extremely efficient as they use only two symbols to define the network: $\{+1,-1\}$. One can make the prior distribution of these symbols a design choice. The recent IR-Net of Qin et al. argues that imposing a Bernoulli distribution with equal priors (equal bit ratios) over the binary weights leads to maximum entropy and thus minimizes information loss. However, prior work cannot precisely control the binary weight distribution during training, and therefore cannot guarantee maximum entropy. Here, we show that quantizing using optimal transport can guarantee any bit ratio, including equal ratios. We investigate experimentally that equal bit ratios are indeed preferable and show that our method leads to optimization benefits. We show that our quantization method is effective when compared to state-of-the-art binarization methods, even when using binary weight pruning.
Deep hashing has shown promising performance in large-scale image retrieval. However, latent codes extracted by Deep Neural Networks (DNNs) will inevitably lose semantic information during the binarization process, which damages the retrieval accuracy and makes it challenging. Although many existing approaches perform regularization to alleviate quantization errors, we figure out an incompatible conflict between metric learning and quantization learning. The metric loss penalizes the inter-class distances to push different classes unconstrained far away. Worse still, it tends to map the latent code deviate from ideal binarization point and generate severe ambiguity in the binarization process. Based on the minimum distance of the binary linear code, we creatively propose Hashing-guided Hinge Function (HHF) to avoid such conflict. In detail, the carefully-designed inflection point, which relies on the hash bit length and category numbers, is explicitly adopted to balance the metric term and quantization term. Such a modification prevents the network from falling into local metric optimal minima in deep hashing. Extensive experiments in CIFAR-10, CIFAR-100, ImageNet, and MS-COCO show that HHF consistently outperforms existing techniques, and is robust and flexible to transplant into other methods. Code is available at https://github.com/JerryXu0129/HHF.
Performance of the sensor-based camera identification (SCI) method heavily relies on the denoising filter in estimating Photo-Response Non-Uniformity (PRNU). Given various attempts on enhancing the quality of the extracted PRNU, it still suffers from unsatisfactory performance in low-resolution images and high computational demand. Leveraging the similarity of PRNU estimation and image denoising, we take advantage of the latest achievements of Convolutional Neural Network (CNN)-based denoisers for PRNU extraction. In this paper, a comparative evaluation of such CNN denoisers on SCI performance is carried out on the public "Dresden Image Database". Our findings are two-fold. From one aspect, both the PRNU extraction and image denoising separate noise from the image content. Hence, SCI can benefit from the recent CNN denoisers if carefully trained. From another aspect, the goals and the scenarios of PRNU extraction and image denoising are different since one optimizes the quality of noise and the other optimizes the image quality. A carefully tailored training is needed when CNN denoisers are used for PRNU estimation. Alternative strategies of training data preparation and loss function design are analyzed theoretically and evaluated experimentally. We point out that feeding the CNNs with image-PRNU pairs and training them with correlation-based loss function result in the best PRNU estimation performance. To facilitate further studies of SCI, we also propose a minimum-loss camera fingerprint quantization scheme using which we save the fingerprints as image files in PNG format. Furthermore, we make the quantized fingerprints of the cameras from the "Dresden Image Database" publicly available.
Quantizing the floating-point weights and activations of deep convolutional neural networks to fixed-point representation yields reduced memory footprints and inference time. Recently, efforts have been afoot towards zero-shot quantization that does not require original unlabelled training samples of a given task. These best-published works heavily rely on the learned batch normalization (BN) parameters to infer the range of the activations for quantization. In particular, these methods are built upon either empirical estimation framework or the data distillation approach, for computing the range of the activations. However, the performance of such schemes severely degrades when presented with a network that does not accommodate BN layers. In this line of thought, we propose a generalized zero-shot quantization (GZSQ) framework that neither requires original data nor relies on BN layer statistics. We have utilized the data distillation approach and leveraged only the pre-trained weights of the model to estimate enriched data for range calibration of the activations. To the best of our knowledge, this is the first work that utilizes the distribution of the pretrained weights to assist the process of zero-shot quantization. The proposed scheme has significantly outperformed the existing zero-shot works, e.g., an improvement of ~ 33% in classification accuracy for MobileNetV2 and several other models that are w & w/o BN layers, for a variety of tasks. We have also demonstrated the efficacy of the proposed work across multiple open-source quantization frameworks. Importantly, our work is the first attempt towards the post-training zero-shot quantization of futuristic unnormalized deep neural networks.
Due to the promising advantages in space compression and inference acceleration, quantized representation learning for recommender systems has become an emerging research direction recently. As the target is to embed latent features in the discrete embedding space, developing quantization for user-item representations with a few low-precision integers confronts the challenge of high information loss, thus leading to unsatisfactory performance in Top-K recommendation. In this work, we study the problem of representation learning for recommendation with 1-bit quantization. We propose a model named Low-loss Quantized Graph Convolutional Network (L^2Q-GCN). Different from previous work that plugs quantization as the final encoder of user-item embeddings, L^2Q-GCN learns the quantized representations whilst capturing the structural information of user-item interaction graphs at different semantic levels. This achieves the substantial retention of intermediate interactive information, alleviating the feature smoothing issue for ranking caused by numerical quantization. To further improve the model performance, we also present an advanced solution named L^2Q-GCN-anl with quantization approximation and annealing training strategy. We conduct extensive experiments on four benchmarks over Top-K recommendation task. The experimental results show that, with nearly 9x representation storage compression, L^2Q-GCN-anl attains about 90~99% performance recovery compared to the state-of-the-art model.
Quantization is emerging as an efficient approach to promote hardware-friendly deep learning and run deep neural networks on resource-limited hardware. However, it still causes a significant decrease to the network in accuracy. We summarize challenges of quantization into two categories: Quantization for Diverse Architectures and Quantization on Complex Scenes. Our studies focus mainly on applying quantization on various architectures and scenes and pushing the limit of quantization to extremely compress and accelerate networks. The comprehensive research on quantization will achieve more powerful, more efficient, and more flexible hardware-friendly deep learning, and make it better suited to more real-world applications.
WiFi-based imaging enables pervasive sensing in a privacy-preserving and cost-effective way. However, most of existing methods either require specialized hardware modification or suffer from poor imaging performance due to the fundamental limit of off-the-shelf commodity WiFi devices in spatial resolution. We observe that the recently developed reconfigurable intelligent surface (RIS) could be a promising solution to overcome these challenges. Thus, in this paper, we propose a RIS-aided WiFi imaging framework to achieve high-resolution imaging with the off-the-shelf WiFi devices. Specifically, we first design a beamforming method to achieve the first-stage imaging by separating the signals from different spatial locations with the aid of the RIS. Then, we propose an optimization-based super-resolution imaging algorithm by leveraging the low rank nature of the reconstructed object. During the optimization, we also explicitly take into account the effect of finite phase quantization in RIS to avoid the resolution degradation due to quantization errors. Simulation results demonstrate that our framework achieves median root mean square error (RMSE) of 0.03 and median structural similarity (SSIM) of 0.52. The visual results show that high-resolution imaging results are achieved with simulation signals at 5 GHz that are matched with commercial WiFi 802.11n/ac protocols.
Geometry-based point cloud compression (G-PCC) can achieve remarkable compression efficiency for point clouds. However, it still leads to serious attribute compression artifacts, especially under low bitrate scenarios. In this paper, we propose a Multi-Scale Graph Attention Network (MS-GAT) to remove the artifacts of point cloud attributes compressed by G-PCC. We first construct a graph based on point cloud geometry coordinates and then use the Chebyshev graph convolutions to extract features of point cloud attributes. Considering that one point may be correlated with points both near and far away from it, we propose a multi-scale scheme to capture the short- and long-range correlations between the current point and its neighboring and distant points. To address the problem that various points may have different degrees of artifacts caused by adaptive quantization, we introduce the quantization step per point as an extra input to the proposed network. We also incorporate a weighted graph attentional layer into the network to pay special attention to the points with more attribute artifacts. To the best of our knowledge, this is the first attribute artifacts removal method for G-PCC. We validate the effectiveness of our method over various point clouds. Objective comparison results show that our proposed method achieves an average of 9.74% BD-rate reduction compared with Predlift and 10.13% BD-rate reduction compared with RAHT. Subjective comparison results present that visual artifacts such as color shifting, blurring, and quantization noise are reduced.
This is an exploratory study that discovers the current image quantization (vector quantization) do not satisfy translation equivariance in the quantized space due to aliasing. Instead of focusing on anti-aliasing, we propose a simple yet effective way to achieve translation-equivariant image quantization by enforcing orthogonality among the codebook embeddings. To explore the advantages of translation-equivariant image quantization, we conduct three proof-of-concept experiments with a carefully controlled dataset: (1) text-to-image generation, where the quantized image indices are the target to predict, (2) image-to-text generation, where the quantized image indices are given as a condition, (3) using a smaller training set to analyze sample efficiency. From the strictly controlled experiments, we empirically verify that the translation-equivariant image quantizer improves not only sample efficiency but also the accuracy over VQGAN up to +11.9% in text-to-image generation and +3.9% in image-to-text generation.
The high memory consumption and computational costs of Recurrent neural network language models (RNNLMs) limit their wider application on resource constrained devices. In recent years, neural network quantization techniques that are capable of producing extremely low-bit compression, for example, binarized RNNLMs, are gaining increasing research interests. Directly training of quantized neural networks is difficult. By formulating quantized RNNLMs training as an optimization problem, this paper presents a novel method to train quantized RNNLMs from scratch using alternating direction methods of multipliers (ADMM). This method can also flexibly adjust the trade-off between the compression rate and model performance using tied low-bit quantization tables. Experiments on two tasks: Penn Treebank (PTB), and Switchboard (SWBD) suggest the proposed ADMM quantization achieved a model size compression factor of up to 31 times over the full precision baseline RNNLMs. Faster convergence of 5 times in model training over the baseline binarized RNNLM quantization was also obtained. Index Terms: Language models, Recurrent neural networks, Quantization, Alternating direction methods of multipliers.
Compression has emerged as one of the essential deep learning research topics, especially for the edge devices that have limited computation power and storage capacity. Among the main compression techniques, low-rank compression via matrix factorization has been known to have two problems. First, an extensive tuning is required. Second, the resulting compression performance is typically not impressive. In this work, we propose a low-rank compression method that utilizes a modified beam-search for an automatic rank selection and a modified stable rank for a compression-friendly training. The resulting BSR (Beam-search and Stable Rank) algorithm requires only a single hyperparameter to be tuned for the desired compression ratio. The performance of BSR in terms of accuracy and compression ratio trade-off curve turns out to be superior to the previously known low-rank compression methods. Furthermore, BSR can perform on par with or better than the state-of-the-art structured pruning methods. As with pruning, BSR can be easily combined with quantization for an additional compression.
In this paper, we present a communication-efficient federated learning framework inspired by quantized compressed sensing. The presented framework consists of gradient compression for wireless devices and gradient reconstruction for a parameter server (PS). Our strategy for gradient compression is to sequentially perform block sparsification, dimensional reduction, and quantization. Thanks to gradient sparsification and quantization, our strategy can achieve a higher compression ratio than one-bit gradient compression. For accurate aggregation of the local gradients from the compressed signals at the PS, we put forth an approximate minimum mean square error (MMSE) approach for gradient reconstruction using the expectation-maximization generalized-approximate-message-passing (EM-GAMP) algorithm. Assuming Bernoulli Gaussian-mixture prior, this algorithm iteratively updates the posterior mean and variance of local gradients from the compressed signals. We also present a low-complexity approach for the gradient reconstruction. In this approach, we use the Bussgang theorem to aggregate local gradients from the compressed signals, then compute an approximate MMSE estimate of the aggregated gradient using the EM-GAMP algorithm. We also provide a convergence rate analysis of the presented framework. Using the MNIST dataset, we demonstrate that the presented framework achieves almost identical performance with the case that performs no compression, while significantly reducing communication overhead for federated learning.