Event cameras are bio-inspired vision sensors that asynchronously represent pixel-level brightness changes as event streams. Event-based monocular multi-view stereo (EMVS) is a technique that exploits the event streams to estimate semi-dense 3D structure with known trajectory. It is a critical task for event-based monocular SLAM. However, the required intensive computation workloads make it challenging for real-time deployment on embedded platforms. In this paper, Eventor is proposed as a fast and efficient EMVS accelerator by realizing the most critical and time-consuming stages including event back-projection and volumetric ray-counting on FPGA. Highly paralleled and fully pipelined processing elements are specially designed via FPGA and integrated with the embedded ARM as a heterogeneous system to improve the throughput and reduce the memory footprint. Meanwhile, the EMVS algorithm is reformulated to a more hardware-friendly manner by rescheduling, approximate computing and hybrid data quantization. Evaluation results on DAVIS dataset show that Eventor achieves up to $24\times$ improvement in energy efficiency compared with Intel i5 CPU platform.
Transformer based re-ranking models can achieve high search relevance through context-aware soft matching of query tokens with document tokens. To alleviate runtime complexity of such inference, previous work has adopted a late interaction architecture with pre-computed contextual token representations at the cost of a large online storage. This paper proposes contextual quantization of token embeddings by decoupling document-specific and document-independent ranking contributions during codebook-based compression. This allows effective online decompression and embedding composition for better search relevance. This paper presents an evaluation of the above compact token representation model in terms of relevance and space efficiency.
Deep neural networks (DNNs) are ubiquitous in computer vision and natural language processing, but suffer from high inference cost. This problem can be addressed by quantization, which consists in converting floating point operations into a lower bit-width format. With the growing concerns on privacy rights, we focus our efforts on data-free methods. However, such techniques suffer from their lack of adaptability to the target devices, as a hardware typically only support specific bit widths. Thus, to adapt to a variety of devices, a quantization method shall be flexible enough to find good accuracy v.s. speed trade-offs for every bit width and target device. To achieve this, we propose REx, a quantization method that leverages residual error expansion, along with group sparsity and an ensemble approximation for better parallelization. REx is backed off by strong theoretical guarantees and achieves superior performance on every benchmarked application (from vision to NLP tasks), architecture (ConvNets, transformers) and bit-width (from int8 to ternary quantization).
Computationally expensive neural networks are ubiquitous in computer vision and solutions for efficient inference have drawn a growing attention in the machine learning community. Examples of such solutions comprise quantization, i.e. converting the processing values (weights and inputs) from floating point into integers e.g. int8 or int4. Concurrently, the rise of privacy concerns motivated the study of less invasive acceleration methods, such as data-free quantization of pre-trained models weights and activations. Previous approaches either exploit statistical information to deduce scalar ranges and scaling factors for the activations in a static manner, or dynamically adapt this range on-the-fly for each input of each layers (also referred to as activations): the latter generally being more accurate at the expanse of significantly slower inference. In this work, we argue that static input quantization can reach the accuracy levels of dynamic methods by means of a per-channel input quantization scheme that allows one to more finely preserve cross-channel dynamics. We show through a thorough empirical evaluation on multiple computer vision problems (e.g. ImageNet classification, Pascal VOC object detection as well as CityScapes semantic segmentation) that the proposed method, dubbed SPIQ, achieves accuracies rivalling dynamic approaches with static-level inference speed, significantly outperforming state-of-the-art quantization methods on every benchmark.
Lack of large-scale note-level labeled data is the major obstacle to singing transcription from polyphonic music. We address the issue by using pseudo labels from vocal pitch estimation models given unlabeled data. The proposed method first converts the frame-level pseudo labels to note-level through pitch and rhythm quantization steps. Then, it further improves the label quality through self-training in a teacher-student framework. To validate the method, we conduct various experiment settings by investigating two vocal pitch estimation models as pseudo-label generators, two setups of teacher-student frameworks, and the number of iterations in self-training. The results show that the proposed method can effectively leverage large-scale unlabeled audio data and self-training with the noisy student model helps to improve performance. Finally, we show that the model trained with only unlabeled data has comparable performance to previous works and the model trained with additional labeled data achieves higher accuracy than the model trained with only labeled data.
In this paper we introduce "Federated Learning Utilities and Tools for Experimentation" (FLUTE), a high-performance open-source platform for federated learning research and offline simulations. The goal of FLUTE is to enable rapid prototyping and simulation of new federated learning algorithms at scale, including novel optimization, privacy, and communications strategies. We describe the architecture of FLUTE, enabling arbitrary federated modeling schemes to be realized. We compare the platform with other state-of-the-art platforms and describe available features of FLUTE for experimentation in core areas of active research, such as optimization, privacy, and scalability. A comparison with other established platforms shows speed-ups of up to 42x and savings in memory footprint of 3x. A sample of the platform capabilities is also presented for a range of tasks, as well as other functionality, such as linear scaling for the number of participating clients, and a variety of federated optimizers, including FedAdam, DGA, etcetera.
Hierarchical VAEs have emerged in recent years as a reliable option for maximum likelihood estimation. However, instability issues and demanding computational requirements have hindered research progress in the area. We present simple modifications to the Very Deep VAE to make it converge up to $2.6\times$ faster, save up to $20\times$ in memory load and improve stability during training. Despite these changes, our models achieve comparable or better negative log-likelihood performance than current state-of-the-art models on all $7$ commonly used image datasets we evaluated on. We also make an argument against using 5-bit benchmarks as a way to measure hierarchical VAE's performance due to undesirable biases caused by the 5-bit quantization. Additionally, we empirically demonstrate that roughly $3\%$ of the hierarchical VAE's latent space dimensions is sufficient to encode most of the image information, without loss of performance, opening up the doors to efficiently leverage the hierarchical VAEs' latent space in downstream tasks. We release our source code and models at https://github.com/Rayhane-mamah/Efficient-VDVAE .
Black-box functions are broadly used to model complex problems that provide no explicit information but the input and output. Despite existing studies of black-box function optimization, the solution set satisfying an inequality with a black-box function plays a more significant role than only one optimum in many practical situations. Covering as much as possible of the solution set through limited evaluations to the black-box objective function is defined as the Black-Box Coverage (BBC) problem in this paper. We formalized this problem in a sample-based search paradigm and constructed a coverage criterion with Confusion Matrix Analysis. Further, we propose LAMBDA (Latent-Action Monte-Carlo Beam Search with Density Adaption) to solve BBC problems. LAMBDA can focus around the solution set quickly by recursively partitioning the search space into accepted and rejected sub-spaces. Compared with La-MCTS, LAMBDA introduces density information to overcome the sampling bias of optimization and obtain more exploration. Benchmarking shows, LAMBDA achieved state-of-the-art performance among all baselines and was at most 33x faster to get 95% coverage than Random Search. Experiments also demonstrate that LAMBDA has a promising future in the verification of autonomous systems in virtual tests.
Recently, pre-trained Transformer based language models, such as BERT, have shown great superiority over the traditional methods in many Natural Language Processing (NLP) tasks. However, the computational cost for deploying these models is prohibitive on resource-restricted devices. One method to alleviate this computation overhead is to quantize the original model into fewer bits representation, and previous work has proved that we can at most quantize both weights and activations of BERT into 8-bits, without degrading its performance. In this work, we propose MKQ-BERT, which further improves the compression level and uses 4-bits for quantization. In MKQ-BERT, we propose a novel way for computing the gradient of the quantization scale, combined with an advanced distillation strategy. On the one hand, we prove that MKQ-BERT outperforms the existing BERT quantization methods for achieving a higher accuracy under the same compression level. On the other hand, we are the first work that successfully deploys the 4-bits BERT and achieves an end-to-end speedup for inference. Our results suggest that we could achieve 5.3x of bits reduction without degrading the model accuracy, and the inference speed of one int4 layer is 15x faster than a float32 layer in Transformer based model.
Increasingly larger and better Transformer models keep advancing state-of-the-art accuracy and capability for Natural Language Processing applications. These models demand more computational power, storage, and energy. Mokey reduces the footprint of state-of-the-art 32-bit or 16-bit floating-point transformer models by quantizing all values to 4-bit indexes into dictionaries of representative 16-bit fixed-point centroids. Mokey does not need fine-tuning, an essential feature as often the training resources or datasets are not available to many. Exploiting the range of values that naturally occur in transformer models, Mokey selects centroid values to also fit an exponential curve. This unique feature enables Mokey to replace the bulk of the original multiply-accumulate operations with narrow 3b fixed-point additions resulting in an area- and energy-efficient hardware accelerator design. Over a set of state-of-the-art transformer models, the Mokey accelerator delivers an order of magnitude improvements in energy efficiency over a Tensor Cores-based accelerator while improving performance by at least $4\times$ and as much as $15\times$ depending on the model and on-chip buffering capacity. Optionally, Mokey can be used as a memory compression assist for any other accelerator, transparently stashing wide floating-point or fixed-point activations or weights into narrow 4-bit indexes. Mokey proves superior to prior state-of-the-art quantization methods for Transformers.
This paper proposes a multi-section vector quantization approach for on-line signature recognition. We have used the MCYT database, which consists of 330 users and 25 skilled forgeries per person performed by 5 different impostors. This database is larger than those typically used in the literature. Nevertheless, we also provide results from the SVC database. Our proposed system outperforms the winner of SVC with a reduced computational requirement, which is around 47 times lower than DTW. In addition, our system improves the database storage requirements due to vector compression, and is more privacy-friendly as it is not possible to recover the original signature using the codebooks. Experimental results with MCYT provide a 99.76% identification rate and 2.46% EER (skilled forgeries and individual threshold). Experimental results with SVC are 100% of identification rate and 0% (individual threshold) and 0.31% (general threshold) when using a two-section VQ approach.
Deep neural networks (DNNs) have demonstrated their effectiveness in a wide range of computer vision tasks, with the state-of-the-art results obtained through complex and deep structures that require intensive computation and memory. Now-a-days, efficient model inference is crucial for consumer applications on resource-constrained platforms. As a result, there is much interest in the research and development of dedicated deep learning (DL) hardware to improve the throughput and energy efficiency of DNNs. Low-precision representation of DNN data-structures through quantization would bring great benefits to specialized DL hardware. However, the rigorous quantization leads to a severe accuracy drop. As such, quantization opens a large hyper-parameter space at bit-precision levels, the exploration of which is a major challenge. In this paper, we propose a novel framework referred to as the Fixed-Point Quantizer of deep neural Networks (FxP-QNet) that flexibly designs a mixed low-precision DNN for integer-arithmetic-only deployment. Specifically, the FxP-QNet gradually adapts the quantization level for each data-structure of each layer based on the trade-off between the network accuracy and the low-precision requirements. Additionally, it employs post-training self-distillation and network prediction error statistics to optimize the quantization of floating-point values into fixed-point numbers. Examining FxP-QNet on state-of-the-art architectures and the benchmark ImageNet dataset, we empirically demonstrate the effectiveness of FxP-QNet in achieving the accuracy-compression trade-off without the need for training. The results show that FxP-QNet-quantized AlexNet, VGG-16, and ResNet-18 reduce the overall memory requirements of their full-precision counterparts by 7.16x, 10.36x, and 6.44x with less than 0.95%, 0.95%, and 1.99% accuracy drop, respectively.
Large-scale pre-trained sequence-to-sequence models like BART and T5 achieve state-of-the-art performance on many generative NLP tasks. However, such models pose a great challenge in resource-constrained scenarios owing to their large memory requirements and high latency. To alleviate this issue, we propose to jointly distill and quantize the model, where knowledge is transferred from the full-precision teacher model to the quantized and distilled low-precision student model. Empirical analyses show that, despite the challenging nature of generative tasks, we were able to achieve a 16.5x model footprint compression ratio with little performance drop relative to the full-precision counterparts on multiple summarization and QA datasets. We further pushed the limit of compression ratio to 27.7x and presented the performance-efficiency trade-off for generative tasks using pre-trained models. To the best of our knowledge, this is the first work aiming to effectively distill and quantize sequence-to-sequence pre-trained models for language generation tasks.
In this work, we consider a full-duplex (FD) massive multiple-input multiple-output (MIMO) cellular network with low resolution analog-to-digital converters (ADCs) and digital-to-analog converters (DACs). Our first contribution is to propose a unified framework for forward link analysis where matched filter precoders are applied at the FD base stations (BSs) under channel hardening. Second, we derive expressions for the signal-to-quantization-plus-interference-plus-noise ratio (SQINR) for general and special cases. Finally, we quantify effects of quantization error, pilot contamination, and full duplexing for a hexagonal cell lattice on spectral efficiency and cumulative distribution function (CDF) to show that FD outperforms half duplex (HD) in a wide variety of scenarios.
Interstellar communication transmitters, intended to be discovered and decoded to information bits, are expected to transmit signals that contain message symbols quantized in at least one of the degrees of freedom of the transmitted signal. A hypothesis is proposed that signal quantization, in the form of multiplicative values of one or more signal measurements, may be observable during the reception of hypothetical discoverable interstellar communication signals. In previous work, using single and multiple synchronized radio telescopes, candidate hypothetical interstellar communication signals comprising delta-t delta-f opposite circular polarized pulse pairs have been reported and analyzed (ref. arXiv:2105.03727, arXiv:2106.10168, arXiv:2202.12791). In the latter report, an apparent quantization of delta-f at multiples of 58.575 Hz was observed. In the current work, a machine process has been implemented to further examine anomalous delta-f and delta-t quantization, with results reported in this paper. As in some past work, a 26 foot diameter radio telescope with fixed azimuth and elevation pointing is used to enable a Right Ascension filter to measure signals associated with a celestial direction of interest, relative to other directions, over a 6.3 hour range of Right Ascension. The 5.25 plus or minus 0.15 hour Right Ascension, -7.6 degrees plus or minus 1 degree Declination celestial direction presents repetition and quantization anomalies, during an experiment lasting 157 days, with the first 143 days overlapping the previous experiment.
When training neural networks with simulated quantization, we observe that quantized weights can, rather unexpectedly, oscillate between two grid-points. The importance of this effect and its impact on quantization-aware training (QAT) are not well-understood or investigated in literature. In this paper, we delve deeper into the phenomenon of weight oscillations and show that it can lead to a significant accuracy degradation due to wrongly estimated batch-normalization statistics during inference and increased noise during training. These effects are particularly pronounced in low-bit ($\leq$ 4-bits) quantization of efficient networks with depth-wise separable layers, such as MobileNets and EfficientNets. In our analysis we investigate several previously proposed QAT algorithms and show that most of these are unable to overcome oscillations. Finally, we propose two novel QAT algorithms to overcome oscillations during training: oscillation dampening and iterative weight freezing. We demonstrate that our algorithms achieve state-of-the-art accuracy for low-bit (3 & 4 bits) weight and activation quantization of efficient architectures, such as MobileNetV2, MobileNetV3, and EfficentNet-lite on ImageNet. Our source code is available at {https://github.com/qualcomm-ai-research/oscillations-qat}.
The increasing size of generative Pre-trained Language Models (PLMs) has greatly increased the demand for model compression. Despite various methods to compress BERT or its variants, there are few attempts to compress generative PLMs, and the underlying difficulty remains unclear. In this paper, we compress generative PLMs by quantization. We find that previous quantization methods fail on generative tasks due to the \textit{homogeneous word embeddings} caused by reduced capacity, and \textit{varied distribution of weights}. Correspondingly, we propose a token-level contrastive distillation to learn distinguishable word embeddings, and a module-wise dynamic scaling to make quantizers adaptive to different modules. Empirical results on various tasks show that our proposed method outperforms the state-of-the-art compression methods on generative PLMs by a clear margin. With comparable performance with the full-precision models, we achieve 14.4x and 13.4x compression rates on GPT-2 and BART, respectively.
Modeling latent variables with priors and hyperpriors is an essential problem in variational image compression. Formally, trade-off between rate and distortion is handled well if priors and hyperpriors precisely describe latent variables. Current practices only adopt univariate priors and process each variable individually. However, we find inter-correlations and intra-correlations exist when observing latent variables in a vectorized perspective. These findings reveal visual redundancies to improve rate-distortion performance and parallel processing ability to speed up compression. This encourages us to propose a novel vectorized prior. Specifically, a multivariate Gaussian mixture is proposed with means and covariances to be estimated. Then, a novel probabilistic vector quantization is utilized to effectively approximate means, and remaining covariances are further induced to a unified mixture and solved by cascaded estimation without context models involved. Furthermore, codebooks involved in quantization are extended to multi-codebooks for complexity reduction, which formulates an efficient compression procedure. Extensive experiments on benchmark datasets against state-of-the-art indicate our model has better rate-distortion performance and an impressive $3.18\times$ compression speed up, giving us the ability to perform real-time, high-quality variational image compression in practice. Our source code is publicly available at \url{https://github.com/xiaosu-zhu/McQuic}.
Over the past years, the ever-growing trend on data storage demand, more specifically for "cold" data (rarely accessed data), has motivated research for alternative systems of data storage. Because of its biochemical characteristics, synthetic DNA molecules are now considered as serious candidates for this new kind of storage. This paper presents some results on lossy image compression methods based on convolutional autoencoders adapted to DNA data storage. The model architectures presented here have been designed to efficiently compress images, encode them into a quaternary code, and finally store them into synthetic DNA molecules. This work also aims at making the compression models better fit the problematics that we encounter when storing data into DNA, namely the fact that the DNA writing, storing and reading methods are error prone processes. The main take away of this kind of compressive autoencoder is our quantization and the robustness to substitution errors thanks to the noise model that we use during training.
With the popularity of virtual assistants (e.g., Siri, Alexa), the use of speech recognition is now becoming more and more widespread.However, speech signals contain a lot of sensitive information, such as the speaker's identity, which raises privacy concerns.The presented experiments show that the representations extracted by the deep layers of speech recognition networks contain speaker information.This paper aims to produce an anonymous representation while preserving speech recognition performance.To this end, we propose to use vector quantization to constrain the representation space and induce the network to suppress the speaker identity.The choice of the quantization dictionary size allows to configure the trade-off between utility (speech recognition) and privacy (speaker identity concealment).