Deep learning (DL) has recently changed the development of intelligent systems and is widely adopted in many real-life applications. Despite their various benefits and potentials, there is a high demand for DL processing in different computationally limited and energy-constrained devices. It is natural to study game-changing technologies such as Binary Neural Networks (BNN) to increase deep learning capabilities. Recently remarkable progress has been made in BNN since they can be implemented and embedded on tiny restricted devices and save a significant amount of storage, computation cost, and energy consumption. However, nearly all BNN acts trade with extra memory, computation cost, and higher performance. This article provides a complete overview of recent developments in BNN. This article focuses exclusively on 1-bit activations and weights 1-bit convolution networks, contrary to previous surveys in which low-bit works are mixed in. It conducted a complete investigation of BNN's development -from their predecessors to the latest BNN algorithms/techniques, presenting a broad design pipeline and discussing each module's variants. Along the way, it examines BNN (a) purpose: their early successes and challenges; (b) BNN optimization: selected representative works that contain essential optimization techniques; (c) deployment: open-source frameworks for BNN modeling and development; (d) terminal: efficient computing architectures and devices for BNN and (e) applications: diverse applications with BNN. Moreover, this paper discusses potential directions and future research opportunities in each section.
Dense Retrieval (DR) has achieved state-of-the-art first-stage ranking effectiveness. However, the efficiency of most existing DR models is limited by the large memory cost of storing dense vectors and the time-consuming nearest neighbor search (NNS) in vector space. Therefore, we present RepCONC, a novel retrieval model that learns discrete Representations via CONstrained Clustering. RepCONC jointly trains dual-encoders and the Product Quantization (PQ) method to learn discrete document representations and enables fast approximate NNS with compact indexes. It models quantization as a constrained clustering process, which requires the document embeddings to be uniformly clustered around the quantization centroids and supports end-to-end optimization of the quantization method and dual-encoders. We theoretically demonstrate the importance of the uniform clustering constraint in RepCONC and derive an efficient approximate solution for constrained clustering by reducing it to an instance of the optimal transport problem. Besides constrained clustering, RepCONC further adopts a vector-based inverted file system (IVF) to support highly efficient vector search on CPUs. Extensive experiments on two popular ad-hoc retrieval benchmarks show that RepCONC achieves better ranking effectiveness than competitive vector quantization baselines under different compression ratio settings. It also substantially outperforms a wide range of existing retrieval models in terms of retrieval effectiveness, memory efficiency, and time efficiency.
Existing deep convolutional neural networks (CNNs) generate massive interlayer feature data during network inference. To maintain real-time processing in embedded systems, large on-chip memory is required to buffer the interlayer feature maps. In this paper, we propose an efficient hardware accelerator with an interlayer feature compression technique to significantly reduce the required on-chip memory size and off-chip memory access bandwidth. The accelerator compresses interlayer feature maps through transforming the stored data into frequency domain using hardware-implemented 8x8 discrete cosine transform (DCT). The high-frequency components are removed after the DCT through quantization. Sparse matrix compression is utilized to further compress the interlayer feature maps. The on-chip memory allocation scheme is designed to support dynamic configuration of the feature map buffer size and scratch pad size according to different network-layer requirements. The hardware accelerator combines compression, decompression, and CNN acceleration into one computing stream, achieving minimal compressing and processing delay. A prototype accelerator is implemented on an FPGA platform and also synthesized in TSMC 28-nm COMS technology. It achieves 403GOPS peak throughput and 1.4x~3.3x interlayer feature map reduction by adding light hardware area overhead, making it a promising hardware accelerator for intelligent IoT devices.
In previous research, we showed that 'texts that tell a story' exhibit a statistical structure that is not Maxwell-Boltzmann but Bose-Einstein. Our explanation is that this is due to the presence of 'indistinguishability' in human language as a result of the same words in different parts of the story being indistinguishable from one another. In the current article, we set out to provide an explanation for this Bose-Einstein statistics. We show that it is the presence of 'meaning' in 'stories' that gives rise to the lack of independence characteristic of Bose-Einstein, and provides conclusive evidence that 'words can be considered the quanta of human language', structurally similar to how 'photons are the quanta of light'. Using several studies on entanglement from our Brussels research group, we also show that it is also the presence of 'meaning' in texts that makes the von Neumann entropy of a total text smaller relative to the entropy of the words composing it. We explain how the new insights in this article fit in with the research domain called 'quantum cognition', where quantum probability models and quantum vector spaces are used in human cognition, and are also relevant to the use of quantum structures in information retrieval and natural language processing, and how they introduce 'quantization' and 'Bose-Einstein statistics' as relevant quantum effects there. Inspired by the conceptuality interpretation of quantum mechanics, and relying on the new insights, we put forward hypotheses about the nature of physical reality. In doing so, we note how this new type of decrease in entropy, and its explanation, may be important for the development of quantum thermodynamics. We likewise note how it can also give rise to an original explanatory picture of the nature of physical reality on the surface of planet Earth, in which human culture emerges as a reinforcing continuation of life.
Graph Convolutional Networks (GCNs) are widely used in a variety of applications, and can be seen as an unstructured version of standard Convolutional Neural Networks (CNNs). As in CNNs, the computational cost of GCNs for large input graphs (such as large point clouds or meshes) can be high and inhibit the use of these networks, especially in environments with low computational resources. To ease these costs, quantization can be applied to GCNs. However, aggressive quantization of the feature maps can lead to a significant degradation in performance. On a different note, Haar wavelet transforms are known to be one of the most effective and efficient approaches to compress signals. Therefore, instead of applying aggressive quantization to feature maps, we propose to utilize Haar wavelet compression and light quantization to reduce the computations and the bandwidth involved with the network. We demonstrate that this approach surpasses aggressive feature quantization by a significant margin, for a variety of problems ranging from node classification to point cloud classification and part and semantic segmentation.
Matrix multiplication is the bedrock in Deep Learning inference application. When it comes to hardware acceleration on edge computing devices, matrix multiplication often takes up a great majority of the time. To achieve better performance in edge computing, we introduce a low-power Multi-layer Perceptron (MLP) accelerator based on a pipelined matrix multiplication scheme and a nonuniform quantization methodology. The implementation is running on Field-programmable Gate Array (FPGA) devices and tested its performance on handwritten digit classification and Q-learning tasks. Results show that our method can achieve better performance with fewer power consumption.
When deploying deep learning models to a device, it is traditionally assumed that available computational resources (compute, memory, and power) remain static. However, real-world computing systems do not always provide stable resource guarantees. Computational resources need to be conserved when load from other processes is high or battery power is low. Inspired by recent works on neural network subspaces, we propose a method for training a "compressible subspace" of neural networks that contains a fine-grained spectrum of models that range from highly efficient to highly accurate. Our models require no retraining, thus our subspace of models can be deployed entirely on-device to allow adaptive network compression at inference time. We present results for achieving arbitrarily fine-grained accuracy-efficiency trade-offs at inference time for structured and unstructured sparsity. We achieve accuracies on-par with standard models when testing our uncompressed models, and maintain high accuracy for sparsity rates above 90% when testing our compressed models. We also demonstrate that our algorithm extends to quantization at variable bit widths, achieving accuracy on par with individually trained networks.
We propose an approach for cognitive coding of speech by unsupervised extraction of contextual representations in two hierarchical levels of abstraction. Speech attributes such as phoneme identity that last one hundred milliseconds or less are captured in the lower level of abstraction, while speech attributes such as speaker identity and emotion that persist up to one second are captured in the higher level of abstraction. This decomposition is achieved by a two-stage neural network, with a lower and an upper stage operating at different time scales. Both stages are trained to predict the content of the signal in their respective latent spaces. A top-down pathway between stages further improves the predictive capability of the network. With an application in speech compression in mind, we investigate the effect of dimensionality reduction and low bitrate quantization on the extracted representations. The performance measured on the LibriSpeech and EmoV-DB datasets reaches, and for some speech attributes even exceeds, that of state-of-the-art approaches.
Federated learning allows collaborative workers to solve a machine learning problem while preserving data privacy. Recent studies have tackled various challenges in federated learning, but the joint optimization of communication overhead, learning reliability, and deployment efficiency is still an open problem. To this end, we propose a new scheme named federated learning via plurality vote (FedVote). In each communication round of FedVote, workers transmit binary or ternary weights to the server with low communication overhead. The model parameters are aggregated via weighted voting to enhance the resilience against Byzantine attacks. When deployed for inference, the model with binary or ternary weights is resource-friendly to edge devices. We show that our proposed method can reduce quantization error and converges faster compared with the methods directly quantizing the model updates.
Federated learning (FL) is an emerging learning paradigm without violating users' privacy. However, large model size and frequent model aggregation cause serious communication bottleneck for FL. To reduce the communication volume, techniques such as model compression and quantization have been proposed. Besides the fixed-bit quantization, existing adaptive quantization schemes use ascending-trend quantization, where the quantization level increases with the training stages. In this paper, we first investigate the impact of quantization on model convergence, and show that the optimal quantization level is directly related to the range of the model updates. Given the model is supposed to converge with the progress of the training, the range of the model updates will gradually shrink, indicating that the quantization level should decrease with the training stages. Based on the theoretical analysis, a descending quantization scheme named FedDQ is proposed. Experimental results show that the proposed descending quantization scheme can save up to 65.2% of the communicated bit volume and up to 68% of the communication rounds, when compared with existing schemes.
Stateful optimizers maintain gradient statistics over time, e.g., the exponentially smoothed sum (SGD with momentum) or squared sum (Adam) of past gradient values. This state can be used to accelerate optimization compared to plain stochastic gradient descent but uses memory that might otherwise be allocated to model parameters, thereby limiting the maximum size of models trained in practice. In this paper, we develop the first optimizers that use 8-bit statistics while maintaining the performance levels of using 32-bit optimizer states. To overcome the resulting computational, quantization, and stability challenges, we develop block-wise dynamic quantization. Block-wise quantization divides input tensors into smaller blocks that are independently quantized. Each block is processed in parallel across cores, yielding faster optimization and high precision quantization. To maintain stability and performance, we combine block-wise quantization with two additional changes: (1) dynamic quantization, a form of non-linear optimization that is precise for both large and small magnitude values, and (2) a stable embedding layer to reduce gradient variance that comes from the highly non-uniform distribution of input tokens in language models. As a result, our 8-bit optimizers maintain 32-bit performance with a small fraction of the memory footprint on a range of tasks, including 1.5B parameter language modeling, GLUE finetuning, ImageNet classification, WMT'14 machine translation, MoCo v2 contrastive ImageNet pretraining+finetuning, and RoBERTa pretraining, without changes to the original optimizer hyperparameters. We open-source our 8-bit optimizers as a drop-in replacement that only requires a two-line code change.
Capsule networks (CapsNets) are an emerging trend in image processing. In contrast to a convolutional neural network, CapsNets are not vulnerable to object deformation, as the relative spatial information of the objects is preserved across the network. However, their complexity is mainly related to the capsule structure and the dynamic routing mechanism, which makes it almost unreasonable to deploy a CapsNet, in its original form, in a resource-constrained device powered by a small microcontroller (MCU). In an era where intelligence is rapidly shifting from the cloud to the edge, this high complexity imposes serious challenges to the adoption of CapsNets at the very edge. To tackle this issue, we present an API for the execution of quantized CapsNets in Arm Cortex-M and RISC-V MCUs. Our software kernels extend the Arm CMSIS-NN and RISC-V PULP-NN to support capsule operations with 8-bit integers as operands. Along with it, we propose a framework to perform post-training quantization of a CapsNet. Results show a reduction in memory footprint of almost 75%, with accuracy loss ranging from 0.07% to 0.18%. In terms of throughput, our Arm Cortex-M API enables the execution of primary capsule and capsule layers with medium-sized kernels in just 119.94 and 90.60 milliseconds (ms), respectively (STM32H755ZIT6U, Cortex-M7 @ 480 MHz). For the GAP-8 SoC (RISC-V RV32IMCXpulp @ 170 MHz), the latency drops to 7.02 and 38.03 ms, respectively.
In this article, we investigate the spectral behavior of random features kernel matrices of the type ${\bf K} = \mathbb{E}_{{\bf w}} \left[\sigma\left({\bf w}^{\sf T}{\bf x}_i\right)\sigma\left({\bf w}^{\sf T}{\bf x}_j\right)\right]_{i,j=1}^n$, with nonlinear function $\sigma(\cdot)$, data ${\bf x}_1, \ldots, {\bf x}_n \in \mathbb{R}^p$, and random projection vector ${\bf w} \in \mathbb{R}^p$ having i.i.d. entries. In a high-dimensional setting where the number of data $n$ and their dimension $p$ are both large and comparable, we show, under a Gaussian mixture model for the data, that the eigenspectrum of ${\bf K}$ is independent of the distribution of the i.i.d.(zero-mean and unit-variance) entries of ${\bf w}$, and only depends on $\sigma(\cdot)$ via its (generalized) Gaussian moments $\mathbb{E}_{z\sim \mathcal N(0,1)}[\sigma'(z)]$ and $\mathbb{E}_{z\sim \mathcal N(0,1)}[\sigma''(z)]$. As a result, for any kernel matrix ${\bf K}$ of the form above, we propose a novel random features technique, called Ternary Random Feature (TRF), that (i) asymptotically yields the same limiting kernel as the original ${\bf K}$ in a spectral sense and (ii) can be computed and stored much more efficiently, by wisely tuning (in a data-dependent manner) the function $\sigma$ and the random vector ${\bf w}$, both taking values in $\{-1,0,1\}$. The computation of the proposed random features requires no multiplication, and a factor of $b$ times less bits for storage compared to classical random features such as random Fourier features, with $b$ the number of bits to store full precision values. Besides, it appears in our experiments on real data that the substantial gains in computation and storage are accompanied with somewhat improved performances compared to state-of-the-art random features compression/quantization methods.
This paper presents a methodology to separate the quantization process from the hardware-specific model compilation stage via a pre-quantized deep learning model description in standard ONNX format. Separating the quantization process from the model compilation stage enables independent development. The methodology is expressive to convey hardware-specific operations and to embed key quantization parameters into a ONNX model which enables hardware/software co-design. Detailed examples are given for both MLP and CNN based networks, which can be extended to other networks in a straightforward fashion.
BERT based ranking models have achieved superior performance on various information retrieval tasks. However, the large number of parameters and complex self-attention operation come at a significant latency overhead. To remedy this, recent works propose late-interaction architectures, which allow pre-computation of intermediate document representations, thus reducing the runtime latency. Nonetheless, having solved the immediate latency issue, these methods now introduce storage costs and network fetching latency, which limits their adoption in real-life production systems. In this work, we propose the Succinct Document Representation (SDR) scheme that computes \emph{highly compressed} intermediate document representations, mitigating the storage/network issue. Our approach first reduces the dimension of token representations by encoding them using a novel autoencoder architecture that uses the document's textual content in both the encoding and decoding phases. After this token encoding step, we further reduce the size of entire document representations using a modern quantization technique. Extensive evaluations on passage re-reranking on the MSMARCO dataset show that compared to existing approaches using compressed document representations, our method is highly efficient, achieving 4x-11.6x better compression rates for the same ranking quality.
Time-series classification is one of the most frequently performed tasks in industrial data science, and one of the most widely used data representation in the industrial setting is tabular representation. In this work, we propose a novel scalable architecture for learning representations from tabular time-series data and subsequently performing downstream tasks such as time-series classification. The representation learning framework is end-to-end, akin to bidirectional encoder representations from transformers (BERT) in language modeling, however, we introduce novel masking technique suitable for pretraining of time-series data. Additionally, we also use one-dimensional convolutions augmented with transformers and explore their effectiveness, since the time-series datasets lend themselves naturally for one-dimensional convolutions. We also propose a novel timestamp embedding technique, which helps in handling both periodic cycles at different time granularity levels, and aperiodic trends present in the time-series data. Our proposed model is end-to-end and can handle both categorical and continuous valued inputs, and does not require any quantization or encoding of continuous features.
An effective unsupervised hashing algorithm leads to compact binary codes preserving the neighborhood structure of data as much as possible. One of the most established schemes for unsupervised hashing is to reduce the dimensionality of data and then find a rigid (neighbourhood-preserving) transformation that reduces the quantization error. Although employing rigid transformations is effective, we may not reduce quantization loss to the ultimate limits. As well, reducing dimensionality and quantization loss in two separate steps seems to be sub-optimal. Motivated by these shortcomings, we propose to employ both rigid and non-rigid transformations to reduce quantization error and dimensionality simultaneously. We relax the orthogonality constraint on the projection in a PCA-formulation and regularize this by a quantization term. We show that both the non-rigid projection matrix and rotation matrix contribute towards minimizing quantization loss but in different ways. A scalable nested coordinate descent approach is proposed to optimize this mixed-integer optimization problem. We evaluate the proposed method on five public benchmark datasets providing almost half a million images. Comparative results indicate that the proposed method mostly outperforms state-of-art linear methods and competes with end-to-end deep solutions.
Network quantization has gained increasing attention with the rapid growth of large pre-trained language models~(PLMs). However, most existing quantization methods for PLMs follow quantization-aware training~(QAT) that requires end-to-end training with full access to the entire dataset. Therefore, they suffer from slow training, large memory overhead, and data security issues. In this paper, we study post-training quantization~(PTQ) of PLMs, and propose module-wise quantization error minimization~(MREM), an efficient solution to mitigate these issues. By partitioning the PLM into multiple modules, we minimize the reconstruction error incurred by quantization for each module. In addition, we design a new model parallel training strategy such that each module can be trained locally on separate computing devices without waiting for preceding modules, which brings nearly the theoretical training speed-up (e.g., $4\times$ on $4$ GPUs). Experiments on GLUE and SQuAD benchmarks show that our proposed PTQ solution not only performs close to QAT, but also enjoys significant reductions in training time, memory overhead, and data consumption.
A deep hashing model typically has two main learning objectives: to make the learned binary hash codes discriminative and to minimize a quantization error. With further constraints such as bit balance and code orthogonality, it is not uncommon for existing models to employ a large number (>4) of losses. This leads to difficulties in model training and subsequently impedes their effectiveness. In this work, we propose a novel deep hashing model with only a single learning objective. Specifically, we show that maximizing the cosine similarity between the continuous codes and their corresponding binary orthogonal codes can ensure both hash code discriminativeness and quantization error minimization. Further, with this learning objective, code balancing can be achieved by simply using a Batch Normalization (BN) layer and multi-label classification is also straightforward with label smoothing. The result is an one-loss deep hashing model that removes all the hassles of tuning the weights of various losses. Importantly, extensive experiments show that our model is highly effective, outperforming the state-of-the-art multi-loss hashing models on three large-scale instance retrieval benchmarks, often by significant margins. Code is available at https://github.com/kamwoh/orthohash
Click-through rate (CTR) prediction becomes indispensable in ubiquitous web recommendation applications. Nevertheless, the current methods are struggling under the cold-start scenarios where the user interactions are extremely sparse. We consider this problem as an automatic identification about whether the user behaviors are rich enough to capture the interests for prediction, and propose an Auto-Quantized Contrastive Learning (AQCL) loss to regularize the model. Different from previous methods, AQCL explores both the instance-instance and the instance-cluster similarity to robustify the latent representation, and automatically reduces the information loss to the active users due to the quantization. The proposed framework is agnostic to different model architectures and can be trained in an end-to-end fashion. Extensive results show that it consistently improves the current state-of-the-art CTR models.