Increasing the model capacity is a known approach to enhance the adversarial robustness of deep learning networks. On the other hand, various model compression techniques, including pruning and quantization, can reduce the size of the network while preserving its accuracy. Several recent studies have addressed the relationship between model compression and adversarial robustness, while some experiments have reported contradictory results. This work summarizes available evidence and discusses possible explanations for the observed effects.
Spiking neural networks (SNNs) have gained attention in recent years due to their ability to handle sparse and event-based data better than regular artificial neural networks (ANNs). Since the structure of SNNs is less suited for typically used accelerators such as GPUs than conventional ANNs, there is a demand for custom hardware accelerators for processing SNNs. In the past, the main focus was on platforms that resemble the structure of multiprocessor systems. In this work, we propose a lightweight neuron layer architecture that allows network structures to be directly mapped onto digital hardware. Our approach is based on differential time coding of spike sequences and the decoupling of processing time and spike timing that allows the SNN to be processed on different hardware platforms. We present synthesis and performance results showing that this architecture can be implemented for networks of more than 1000 neurons with high clock speeds on a State-of-the-Art FPGA. We furthermore show results on the robustness of our approach to quantization. These results demonstrate that high-accuracy inference can be performed with bit widths as low as 4.
The rapid growth of Internet of Things (IoT) devices and applications has led to an increased demand for advanced analytics and machine learning techniques capable of handling the challenges associated with data privacy, security, and scalability. Federated learning (FL) and blockchain technologies have emerged as promising approaches to address these challenges by enabling decentralized, secure, and privacy-preserving model training on distributed data sources. In this paper, we present a novel IoT solution that combines the incremental learning vector quantization algorithm (XuILVQ) with Ethereum blockchain technology to facilitate secure and efficient data sharing, model training, and prototype storage in a distributed environment. Our proposed architecture addresses the shortcomings of existing blockchain-based FL solutions by reducing computational and communication overheads while maintaining data privacy and security. We assess the performance of our system through a series of experiments, showcasing its potential to enhance the accuracy and efficiency of machine learning tasks in IoT settings.
Recent quantization techniques have enabled heterogeneous precisions at very fine granularity, e.g., each parameter/activation can take on a different precision, resulting in compact neural networks without sacrificing accuracy. However, there is a lack of efficient architectural support for such networks, which require additional hardware to decode the precision settings for individual variables, align the variables, and provide fine-grained mixed-precision compute capabilities. The complexity of these operations introduces high overheads. Thus, the improvements in inference latency/energy of these networks are not commensurate with the compression ratio, and may be inferior to larger quantized networks with uniform precisions. We present an end-to-end co-design approach encompassing computer architecture, training algorithm, and inference optimization to efficiently execute networks with fine-grained heterogeneous precisions. The key to our approach is a novel training algorithm designed to accommodate hardware constraints and inference operation requirements, outputting networks with input-channel-wise heterogeneous precisions and at most three precision levels. Combined with inference optimization techniques, existing architectures with low-cost enhancements can support such networks efficiently, yielding optimized tradeoffs between accuracy, compression ratio and inference latency/energy. We demonstrate the efficacy of our approach across CPU and GPU architectures. For various representative neural networks, our approach achieves >10x improvements in both compression ratio and inference latency, with negligible degradation in accuracy compared to full-precision networks.
3D human pose and shape estimation (HPE) aims to reconstruct the 3D human body, face, and hands from a single image. Although powerful deep learning models have achieved accurate estimation in this task, they require enormous memory and computational resources. Consequently, these methods can hardly be deployed on resource-limited edge devices. In this work, we propose BinaryHPE, a novel binarization method designed to estimate the 3D human body, face, and hands parameters efficiently. Specifically, we propose a novel binary backbone called Binarized Dual Residual Network (BiDRN), designed to retain as much full-precision information as possible. Furthermore, we propose the Binarized BoxNet, an efficient sub-network for predicting face and hands bounding boxes, which further reduces model redundancy. Comprehensive quantitative and qualitative experiments demonstrate the effectiveness of BinaryHPE, which has a significant improvement over state-of-the-art binarization algorithms. Moreover, our BinaryHPE achieves comparable performance with the full-precision method Hand4Whole while using only 22.1% parameters and 14.8% operations. We will release all the code and pretrained models.
Neural Radiance Fields (NeRFs) have demonstrated remarkable potential in capturing complex 3D scenes with high fidelity. However, one persistent challenge that hinders the widespread adoption of NeRFs is the computational bottleneck due to the volumetric rendering. On the other hand, 3D Gaussian splatting (3DGS) has recently emerged as an alternative representation that leverages a 3D Gaussisan-based representation and adopts the rasterization pipeline to render the images rather than volumetric rendering, achieving very fast rendering speed and promising image quality. However, a significant drawback arises as 3DGS entails a substantial number of 3D Gaussians to maintain the high fidelity of the rendered images, which requires a large amount of memory and storage. To address this critical issue, we place a specific emphasis on two key objectives: reducing the number of Gaussian points without sacrificing performance and compressing the Gaussian attributes, such as view-dependent color and covariance. To this end, we propose a learnable mask strategy that significantly reduces the number of Gaussians while preserving high performance. In addition, we propose a compact but effective representation of view-dependent color by employing a grid-based neural field rather than relying on spherical harmonics. Finally, we learn codebooks to compactly represent the geometric attributes of Gaussian by vector quantization. With model compression techniques such as quantization and entropy coding, we consistently show over 25$\times$ reduced storage and enhanced rendering speed, while maintaining the quality of the scene representation, compared to 3DGS. Our work provides a comprehensive framework for 3D scene representation, achieving high performance, fast training, compactness, and real-time rendering. Our project page is available at https://maincold2.github.io/c3dgs/.
Traditionally, IoT edge devices have been perceived primarily as low-power components with limited capabilities for autonomous operations. Yet, with emerging advancements in embedded AI hardware design, a foundational shift paves the way for future possibilities. Thus, the aim of the KDT NEUROKIT2E project is to establish a new open-source framework to further facilitate AI applications on edge devices by developing new methods in quantization, pruning-aware training, and sparsification. These innovations hold the potential to expand the functional range of such devices considerably, enabling them to manage complex Machine Learning (ML) tasks utilizing local resources and laying the groundwork for innovative learning approaches. In the context of 6G's transformative potential, distributed learning among independent agents emerges as a pivotal application, attributed to 6G networks' support for ultra-reliable low-latency communication, enhanced data rates, and advanced edge computing capabilities. Our research focuses on the mechanisms and methodologies that allow edge network-enabled agents to engage in collaborative learning in distributed environments. Particularly, one of the key issues within distributed collaborative learning is determining the degree of confidence in the learning results, considering the spatio-temporal locality of data sets perceived by independent agents.
In high-dynamic range (HDR) analog-to-digital converters (ADCs), having many quantization bits minimizes quantization errors but results in high bit rates, limiting their application scope. A strategy combining modulo-folding with a low-DR ADC can create an efficient HDR-ADC with fewer bits. However, this typically demands oversampling, increasing the overall bit rate. An alternative method using phase modulation (PM) achieves HDR-ADC functionality by modulating the phase of a carrier signal with the analog input. This allows a low-DR ADC with fewer bits. We've derived identifiability results enabling reconstruction of the original signal from PM samples acquired at the Nyquist rate, adaptable to various signals and non-uniform sampling. Using discrete phase demodulation algorithms for practical implementation, our PM-based approach doesn't require oversampling in noise-free conditions, contrasting with modulo-based ADCs. With noise, our PM-based HDR method demonstrates efficiency with lower reconstruction errors and reduced sampling rates. Our hardware prototype illustrates reconstructing signals ten times greater than the ADC's DR from Nyquist rate samples, potentially replacing high-bit rate HDR-ADCs while meeting existing bit rate needs.
Ensuring high-quality video content for wireless users has become increasingly vital. Nevertheless, maintaining a consistent level of video quality faces challenges due to the fluctuating encoded bitrate, primarily caused by dynamic video content, especially in live streaming scenarios. Video compression is typically employed to eliminate unnecessary redundancies within and between video frames, thereby reducing the required bandwidth for video transmission. The encoded bitrate and the quality of the compressed video depend on encoder parameters, specifically, the quantization parameter (QP). Poor choices of encoder parameters can result in reduced bandwidth efficiency and high likelihood of non-conformance. Non-conformance refers to the violation of the peak signal-to-noise ratio (PSNR) constraint for an encoded video segment. To address these issues, a real-time deep learning-based H.264 controller is proposed. This controller dynamically estimates the optimal encoder parameters based on the content of a video chunk with minimal delay. The objective is to maintain video quality in terms of PSNR above a specified threshold while minimizing the average bitrate of the compressed video. Experimental results, conducted on both QCIF dataset and a diverse range of random videos from public datasets, validate the effectiveness of this approach. Notably, it achieves improvements of up to 2.5 times in average bandwidth usage compared to the state-of-the-art adaptive bitrate video streaming, with a negligible non-conformance probability below $10^{-2}$.
Parameter-efficient fine-tuning (PEFT) techniques make it possible to efficiently adapt a language model to create "expert" models that specialize to new tasks or domains. Recent techniques in model merging and compositional generalization leverage these expert models by dynamically composing modules to improve zero/few-shot generalization. Despite the efficiency of PEFT methods, the size of expert models can make it onerous to retrieve expert models per query over high-latency networks like the Internet or serve multiple experts on a single GPU. To address these issues, we present ComPEFT, a novel method for compressing fine-tuning residuals (task vectors) of PEFT based models. ComPEFT employs sparsification and ternary quantization to reduce the size of the PEFT module without performing any additional retraining while preserving or enhancing model performance. In extensive evaluation across T5, T0, and LLaMA-based models with 200M - 65B parameters, ComPEFT achieves compression ratios of 8x - 50x. In particular, we show that ComPEFT improves with scale - stronger models exhibit higher compressibility and better performance. For example, we show that ComPEFT applied to LLaMA outperforms QLoRA by 4.16% on MMLU with a storage size reduction of up to 26x. In addition, we show that the compressed experts produced by ComPEFT maintain few-shot compositional generalization capabilities, facilitate efficient communication and computation, and exhibit enhanced performance when merged. Lastly, we provide an analysis of different method components, compare it with other PEFT methods, and test ComPEFT's efficacy for compressing the residual of full-finetuning. Our code is available at https://github.com/prateeky2806/compeft.
Sound Source Localization (SSL) involves estimating the Direction of Arrival (DOA) of sound sources. Since the DOA estimation output space is continuous, regression might be more suitable for DOA, offering higher precision. However, in practice, classification often outperforms regression, exhibiting greater robustness to interference. Conversely, classification's drawback is inherent quantization error. Within the classification paradigm, the DOA output space is discretized into intervals, each treated as a class. These classes show strong inter-class correlations, being inherently ordered, with higher similarity as intervals grow closer. Nevertheless, this has not been fully exploited. To address this, we propose an Unbiased Label Distribution (ULD) to eliminate quantization error in training targets. Furthermore, we tailor two loss functions for the soft label family: Negative Log Absolute Error (NLAE) and Mean Squared Error without activation (MSE(wo)). Finally, we introduce Weighted Adjacent Decoding (WAD) to overcome quantization error during model prediction decoding. Experimental results demonstrate our approach surpasses classification quantization limits, achieving state-of-the-art performance. Our code and supplementary materials are available at https://github.com/linfeng-feng/ULD.
Accelerating compute intensive non-real-time beam-forming algorithms in ultrasound imaging using deep learning architectures has been gaining momentum in the recent past. Nonetheless, the complexity of the state-of-the-art deep learning techniques poses challenges for deployment on resource-constrained edge devices. In this work, we propose a novel vision transformer based tiny beamformer (Tiny-VBF), which works on the raw radio-frequency channel data acquired through single-angle plane wave insonification. The output of our Tiny-VBF provides fast envelope detection requiring very low frame rate, i.e. 0.34 GOPs/Frame for a frame size of 368 x 128 in comparison to the state-of-the-art deep learning models. It also exhibited an 8% increase in contrast and gains of 5% and 33% in axial and lateral resolution respectively when compared to Tiny-CNN on in-vitro dataset. Additionally, our model showed a 4.2% increase in contrast and gains of 4% and 20% in axial and lateral resolution respectively when compared against conventional Delay-and-Sum (DAS) beamformer. We further propose an accelerator architecture and implement our Tiny-VBF model on a Zynq UltraScale+ MPSoC ZCU104 FPGA using a hybrid quantization scheme with 50% less resource consumption compared to the floating-point implementation, while preserving the image quality.
Post-training quantization (PTQ) is a powerful technique for model compression, reducing the numerical precision in neural networks without additional training overhead. Recent works have investigated adopting 8-bit floating-point formats(FP8) in the context of PTQ for model inference. However, floating-point formats smaller than 8 bits and their relative comparison in terms of accuracy-hardware cost with integers remains unexplored on FPGAs. In this work, we present minifloats, which are reduced-precision floating-point formats capable of further reducing the memory footprint, latency, and energy cost of a model while approaching full-precision model accuracy. We implement a custom FPGA-based multiply-accumulate operator library and explore the vast design space, comparing minifloat and integer representations across 3 to 8 bits for both weights and activations. We also examine the applicability of various integerbased quantization techniques to minifloats. Our experiments show that minifloats offer a promising alternative for emerging workloads such as vision transformers.
Spiking Neural Networks (SNNs) have been widely praised for their high energy efficiency and immense potential. However, comprehensive research that critically contrasts and correlates SNNs with quantized Artificial Neural Networks (ANNs) remains scant, often leading to skewed comparisons lacking fairness towards ANNs. This paper introduces a unified perspective, illustrating that the time steps in SNNs and quantized bit-widths of activation values present analogous representations. Building on this, we present a more pragmatic and rational approach to estimating the energy consumption of SNNs. Diverging from the conventional Synaptic Operations (SynOps), we champion the "Bit Budget" concept. This notion permits an intricate discourse on strategically allocating computational and storage resources between weights, activation values, and temporal steps under stringent hardware constraints. Guided by the Bit Budget paradigm, we discern that pivoting efforts towards spike patterns and weight quantization, rather than temporal attributes, elicits profound implications for model performance. Utilizing the Bit Budget for holistic design consideration of SNNs elevates model performance across diverse data types, encompassing static imagery and neuromorphic datasets. Our revelations bridge the theoretical chasm between SNNs and quantized ANNs and illuminate a pragmatic trajectory for future endeavors in energy-efficient neural computations.
The field of Tiny Machine Learning (TinyML) has gained significant attention due to its potential to enable intelligent applications on resource-constrained devices. This review provides an in-depth analysis of the advancements in efficient neural networks and the deployment of deep learning models on ultra-low power microcontrollers (MCUs) for TinyML applications. It begins by introducing neural networks and discussing their architectures and resource requirements. It then explores MEMS-based applications on ultra-low power MCUs, highlighting their potential for enabling TinyML on resource-constrained devices. The core of the review centres on efficient neural networks for TinyML. It covers techniques such as model compression, quantization, and low-rank factorization, which optimize neural network architectures for minimal resource utilization on MCUs. The paper then delves into the deployment of deep learning models on ultra-low power MCUs, addressing challenges such as limited computational capabilities and memory resources. Techniques like model pruning, hardware acceleration, and algorithm-architecture co-design are discussed as strategies to enable efficient deployment. Lastly, the review provides an overview of current limitations in the field, including the trade-off between model complexity and resource constraints. Overall, this review paper presents a comprehensive analysis of efficient neural networks and deployment strategies for TinyML on ultra-low-power MCUs. It identifies future research directions for unlocking the full potential of TinyML applications on resource-constrained devices.
One of the major bottlenecks in high-resolution Earth Observation (EO) space systems is the downlink between the satellite and the ground. Due to hardware limitations, on-board power limitations or ground-station operation costs, there is a strong need to reduce the amount of data transmitted. Various processing methods can be used to compress the data. One of them is the use of on-board deep learning to extract relevant information in the data. However, most ground-based deep neural network parameters and computations are performed using single-precision floating-point arithmetic, which is not adapted to the context of on-board processing. We propose to rely on quantized neural networks and study how to combine low precision (mini) floating-point arithmetic with a Quantization-Aware Training methodology. We evaluate our approach with a semantic segmentation task for ship detection using satellite images from the Airbus Ship dataset. Our results show that 6-bit floating-point quantization for both weights and activations can compete with single-precision without significant accuracy degradation. Using a Thin U-Net 32 model, only a 0.3% accuracy degradation is observed with 6-bit minifloat quantization (a 6-bit equivalent integer-based approach leads to a 0.5% degradation). An initial hardware study also confirms the potential impact of such low-precision floating-point designs, but further investigation at the scale of a full inference accelerator is needed before concluding whether they are relevant in a practical on-board scenario.
We propose a simple approach for memory-efficient adaptation of pretrained language models. Our approach uses an iterative algorithm to decompose each pretrained matrix into a high-precision low-rank component and a memory-efficient quantized component. During finetuning, the quantized component remains fixed and only the low-rank component is updated. We present an integer linear programming formulation of the quantization component which enables dynamic configuration of quantization parameters (e.g., bit-width, block size) for each matrix given an overall target memory budget. We further explore a data-aware version of the algorithm which uses an approximation of the Fisher information matrix to weight the reconstruction objective during matrix decomposition. Experiments on finetuning RoBERTa and LLaMA-2 (7B and 70B) demonstrate that our low-rank plus quantized matrix decomposition approach (LQ-LoRA) outperforms strong QLoRA and GPTQ-LoRA baselines and enables aggressive quantization to sub-3 bits with only minor performance degradations. When finetuned on a language modeling calibration dataset, LQ-LoRA can also be used for model compression; in this setting our 2.75-bit LLaMA-2-70B model (which has 2.85 bits on average when including the low-rank components and requires 27GB of GPU memory) performs respectably compared to the 16-bit baseline.
From classical HPC to deep learning, MatMul is at the heart of today's computing. The recent Maddness method approximates MatMul without the need for multiplication by using a hash-based version of product quantization (PQ) indexing into a look-up table (LUT). Stella Nera is the first Maddness accelerator and it achieves 15x higher area efficiency (GMAC/s/mm^2) and more than 25x higher energy efficiency (TMAC/s/W) than direct MatMul accelerators implemented in the same technology. The hash function is a decision tree, which allows for an efficient hardware implementation as the multiply-accumulate operations are replaced by decision tree passes and LUT lookups. The entire Maddness MatMul can be broken down into parts that allow an effective implementation with small computing units and memories, allowing it to reach extreme efficiency while remaining generically applicable for MatMul tasks. In a commercial 14nm technology and scaled to 3nm, we achieve an energy efficiency of 161 [email omitted] with a Top-1 accuracy on CIFAR-10 of more than 92.5% using ResNet9.
Albeit the scalable performance of vision transformers (ViTs), the dense computational costs (training & inference) undermine their position in industrial applications. Post-training quantization (PTQ), tuning ViTs with a tiny dataset and running in a low-bit format, well addresses the cost issue but unluckily bears more performance drops in lower-bit cases. In this paper, we introduce I&S-ViT, a novel method that regulates the PTQ of ViTs in an inclusive and stable fashion. I&S-ViT first identifies two issues in the PTQ of ViTs: (1) Quantization inefficiency in the prevalent log2 quantizer for post-Softmax activations; (2) Rugged and magnified loss landscape in coarse-grained quantization granularity for post-LayerNorm activations. Then, I&S-ViT addresses these issues by introducing: (1) A novel shift-uniform-log2 quantizer (SULQ) that incorporates a shift mechanism followed by uniform quantization to achieve both an inclusive domain representation and accurate distribution approximation; (2) A three-stage smooth optimization strategy (SOS) that amalgamates the strengths of channel-wise and layer-wise quantization to enable stable learning. Comprehensive evaluations across diverse vision tasks validate I&S-ViT' superiority over existing PTQ of ViTs methods, particularly in low-bit scenarios. For instance, I&S-ViT elevates the performance of 3-bit ViT-B by an impressive 50.68%.
Quantization and pruning form the foundation of compression for neural networks, enabling efficient inference for large language models (LLMs). Recently, various quantization and pruning techniques have demonstrated remarkable performance in a post-training setting. They rely upon calibration data, a small set of unlabeled examples that are used to generate layer activations. However, no prior work has systematically investigated how the calibration data impacts the effectiveness of model compression methods. In this paper, we present the first extensive empirical study on the effect of calibration data upon LLM performance. We trial a variety of quantization and pruning methods, datasets, tasks, and models. Surprisingly, we find substantial variations in downstream task performance, contrasting existing work that suggests a greater level of robustness to the calibration data. Finally, we make a series of recommendations for the effective use of calibration data in LLM quantization and pruning.