Spiking Neural Networks (SNNs) mimic the information-processing mechanisms of the human brain and are highly energy-efficient, making them well-suited for low-power edge devices. However, the pursuit of accuracy in current studies leads to large, long-timestep SNNs, conflicting with the resource constraints of these devices. In order to design lightweight and efficient SNNs, we propose a new approach named LitE-SNN that incorporates both spatial and temporal compression into the automated network design process. Spatially, we present a novel Compressive Convolution block (CompConv) to expand the search space to support pruning and mixed-precision quantization. Temporally, we are the first to propose a compressive timestep search to identify the optimal number of timesteps under specific computation cost constraints. Finally, we formulate a joint optimization to simultaneously learn the architecture parameters and spatial-temporal compression strategies to achieve high performance while minimizing memory and computation costs. Experimental results on CIFAR-10, CIFAR-100, and Google Speech Command datasets demonstrate our proposed LitE-SNNs can achieve competitive or even higher accuracy with remarkably smaller model sizes and fewer computation costs.
Deploying neural networks on microcontroller units (MCUs) presents substantial challenges due to their constrained computation and memory resources. Previous researches have explored patch-based inference as a strategy to conserve memory without sacrificing model accuracy. However, this technique suffers from severe redundant computation overhead, leading to a substantial increase in execution latency. A feasible solution to address this issue is mixed-precision quantization, but it faces the challenges of accuracy degradation and a time-consuming search time. In this paper, we propose QuantMCU, a novel patch-based inference method that utilizes value-driven mixed-precision quantization to reduce redundant computation. We first utilize value-driven patch classification (VDPC) to maintain the model accuracy. VDPC classifies patches into two classes based on whether they contain outlier values. For patches containing outlier values, we apply 8-bit quantization to the feature maps on the dataflow branches that follow. In addition, for patches without outlier values, we utilize value-driven quantization search (VDQS) on the feature maps of their following dataflow branches to reduce search time. Specifically, VDQS introduces a novel quantization search metric that takes into account both computation and accuracy, and it employs entropy as an accuracy representation to avoid additional training. VDQS also adopts an iterative approach to determine the bitwidth of each feature map to further accelerate the search process. Experimental results on real-world MCU devices show that QuantMCU can reduce computation by 2.2x on average while maintaining comparable model accuracy compared to the state-of-the-art patch-based inference methods.
Six-bit quantization (FP6) can effectively reduce the size of large language models (LLMs) and preserve the model quality consistently across varied applications. However, existing systems do not provide Tensor Core support for FP6 quantization and struggle to achieve practical performance improvements during LLM inference. It is challenging to support FP6 quantization on GPUs due to (1) unfriendly memory access of model weights with irregular bit-width and (2) high runtime overhead of weight de-quantization. To address these problems, we propose TC-FPx, the first full-stack GPU kernel design scheme with unified Tensor Core support of float-point weights for various quantization bit-width. We integrate TC-FPx kernel into an existing inference system, providing new end-to-end support (called FP6-LLM) for quantized LLM inference, where better trade-offs between inference cost and model quality are achieved. Experiments show that FP6-LLM enables the inference of LLaMA-70b using only a single GPU, achieving 1.69x-2.65x higher normalized inference throughput than the FP16 baseline. The source code is publicly available at https://github.com/usyd-fsalab/fp6_llm.
The majority of the research on the quantization of Deep Neural Networks (DNNs) is focused on reducing the precision of tensors visible by high-level frameworks (e.g., weights, activations, and gradients). However, current hardware still relies on high-accuracy core operations. Most significant is the operation of accumulating products. This high-precision accumulation operation is gradually becoming the main computational bottleneck. This is because, so far, the usage of low-precision accumulators led to a significant degradation in performance. In this work, we present a simple method to train and fine-tune high-end DNNs, to allow, for the first time, utilization of cheaper, $12$-bits accumulators, with no significant degradation in accuracy. Lastly, we show that as we decrease the accumulation precision further, using fine-grained gradient approximations can improve the DNN accuracy.
Large Language Models (LLMs) such as ChatGPT and LlaMA are advancing rapidly in generative Artificial Intelligence (AI), but their immense size poses significant challenges, such as huge training and inference costs, substantial energy demands, and limitations for on-site deployment. Traditional compression methods such as pruning, distillation, and low-rank approximation focus on reducing the effective number of neurons in the network, while quantization focuses on reducing the numerical precision of individual weights to reduce the model size while keeping the number of neurons fixed. While these compression methods have been relatively successful in practice, there is no compelling reason to believe that truncating the number of neurons is an optimal strategy. In this context, this paper introduces CompactifAI, an innovative LLM compression approach using quantum-inspired Tensor Networks that focuses on the model's correlation space instead, allowing for a more controlled, refined and interpretable model compression. Our method is versatile and can be implemented with - or on top of - other compression techniques. As a benchmark, we demonstrate that a combination of CompactifAI with quantization allows to reduce a 93% the memory size of LlaMA 7B, reducing also 70% the number of parameters, accelerating 50% the training and 25% the inference times of the model, and just with a small accuracy drop of 2% - 3%, going much beyond of what is achievable today by other compression techniques. Our methods also allow to perform a refined layer sensitivity profiling, showing that deeper layers tend to be more suitable for tensor network compression, which is compatible with recent observations on the ineffectiveness of those layers for LLM performance. Our results imply that standard LLMs are, in fact, heavily overparametrized, and do not need to be large at all.
We introduce and investigate the iterated application of Generalized Matrix Learning Vector Quantizaton for the analysis of feature relevances in classification problems, as well as for the construction of class-discriminative subspaces. The suggested Iterated Relevance Matrix Analysis (IRMA) identifies a linear subspace representing the classification specific information of the considered data sets using Generalized Matrix Learning Vector Quantization (GMLVQ). By iteratively determining a new discriminative subspace while projecting out all previously identified ones, a combined subspace carrying all class-specific information can be found. This facilitates a detailed analysis of feature relevances, and enables improved low-dimensional representations and visualizations of labeled data sets. Additionally, the IRMA-based class-discriminative subspace can be used for dimensionality reduction and the training of robust classifiers with potentially improved performance.
Neural Architecture Search (NAS) has become the de-facto approach for designing accurate and efficient networks for edge devices. Since models are typically quantized for edge deployment, recent work has investigated quantization-aware NAS (QA-NAS) to search for highly accurate and efficient quantized models. However, existing QA-NAS approaches, particularly few-bit mixed-precision (FB-MP) methods, do not scale to larger tasks. Consequently, QA-NAS has mostly been limited to low-scale tasks and tiny networks. In this work, we present an approach to enable QA-NAS (INT8 and FB-MP) on large-scale tasks by leveraging the block-wise formulation introduced by block-wise NAS. We demonstrate strong results for the semantic segmentation task on the Cityscapes dataset, finding FB-MP models 33% smaller and INT8 models 17.6% faster than DeepLabV3 (INT8) without compromising task performance.
Accurate and rapid railway track segmentation can assist automatic train driving and is a key step in early warning to fixed or moving obstacles on the railway track. However, certain existing algorithms tailored for track segmentation often struggle to meet the requirements of real-time and efficiency on resource-constrained edge devices. Considering this challenge, we propose an edge-enabled real-time railway track segmentation algorithm, which is optimized to be suitable for edge applications by optimizing the network structure and quantizing the model after training. Initially, Ghost convolution is introduced to reduce the complexity of the backbone, thereby achieving the extraction of key information of the interested region at a lower cost. To further reduce the model complexity and calculation, a new lightweight detection head is proposed to achieve the best balance between accuracy and efficiency. Subsequently, we introduce quantization techniques to map the model's floating-point weights and activation values into lower bit-width fixed-point representations, reducing computational demands and memory footprint, ultimately accelerating the model's inference. Finally, we draw inspiration from GPU parallel programming principles to expedite the pre-processing and post-processing stages of the algorithm by doing parallel processing. The approach is evaluated with public and challenging dataset RailSem19 and tested on Jetson Nano. Experimental results demonstrate that our enhanced algorithm achieves an accuracy level of 83.3% while achieving a real-time inference rate of 25 frames per second when the input size is 480x480, thereby effectively meeting the requirements for real-time and high-efficiency operation.
Vision transformers (ViTs) have demonstrated remarkable performance across various visual tasks. However, ViT models suffer from substantial computational and memory requirements, making it challenging to deploy them on resource-constrained platforms. Quantization is a popular approach for reducing model size, but most studies mainly focus on equal bit-width quantization for the entire network, resulting in sub-optimal solutions. While there are few works on mixed precision quantization (MPQ) for ViTs, they typically rely on search space-based methods or employ mixed precision arbitrarily. In this paper, we introduce LRP-QViT, an explainability-based method for assigning mixed-precision bit allocations to different layers based on their importance during classification. Specifically, to measure the contribution score of each layer in predicting the target class, we employ the Layer-wise Relevance Propagation (LRP) method. LRP assigns local relevance at the output layer and propagates it through all layers, distributing the relevance until it reaches the input layers. These relevance scores serve as indicators for computing the layer contribution score. Additionally, we have introduced a clipped channel-wise quantization aimed at eliminating outliers from post-LayerNorm activations to alleviate severe inter-channel variations. To validate and assess our approach, we employ LRP-QViT across ViT, DeiT, and Swin transformer models on various datasets. Our experimental findings demonstrate that both our fixed-bit and mixed-bit post-training quantization methods surpass existing models in the context of 4-bit and 6-bit quantization.
In disentangled representation learning, a model is asked to tease apart a dataset's underlying sources of variation and represent them independently of one another. Since the model is provided with no ground truth information about these sources, inductive biases take a paramount role in enabling disentanglement. In this work, we construct an inductive bias towards encoding to and decoding from an organized latent space. Concretely, we do this by (i) quantizing the latent space into discrete code vectors with a separate learnable scalar codebook per dimension and (ii) applying strong model regularization via an unusually high weight decay. Intuitively, the latent space design forces the encoder to combinatorially construct codes from a small number of distinct scalar values, which in turn enables the decoder to assign a consistent meaning to each value. Regularization then serves to drive the model towards this parsimonious strategy. We demonstrate the broad applicability of this approach by adding it to both basic data-reconstructing (vanilla autoencoder) and latent-reconstructing (InfoGAN) generative models. For reliable evaluation, we also propose InfoMEC, a new set of metrics for disentanglement that is cohesively grounded in information theory and fixes well-established shortcomings in previous metrics. Together with regularization, latent quantization dramatically improves the modularity and explicitness of learned representations on a representative suite of benchmark datasets. In particular, our quantized-latent autoencoder (QLAE) consistently outperforms strong methods from prior work in these key disentanglement properties without compromising data reconstruction.
While convolution and self-attention are extensively used in learned image compression (LIC) for transform coding, this paper proposes an alternative called Contextual Clustering based LIC (CLIC) which primarily relies on clustering operations and local attention for correlation characterization and compact representation of an image. As seen, CLIC expands the receptive field into the entire image for intra-cluster feature aggregation. Afterward, features are reordered to their original spatial positions to pass through the local attention units for inter-cluster embedding. Additionally, we introduce the Guided Post-Quantization Filtering (GuidedPQF) into CLIC, effectively mitigating the propagation and accumulation of quantization errors at the initial decoding stage. Extensive experiments demonstrate the superior performance of CLIC over state-of-the-art works: when optimized using MSE, it outperforms VVC by about 10% BD-Rate in three widely-used benchmark datasets; when optimized using MS-SSIM, it saves more than 50% BD-Rate over VVC. Our CLIC offers a new way to generate compact representations for image compression, which also provides a novel direction along the line of LIC development.
It is necessary to develop efficient DNNs deployed on edge devices with limited computation resources. However, the compressed networks often execute new tasks in the target domain, which is different from the source domain where the original network is trained. It is important to investigate the robustness of compressed networks in two types of data distribution shifts: domain shifts and adversarial perturbations. In this study, we discover that compressed models are less robust to distribution shifts than their original networks. Interestingly, larger networks are more vulnerable to losing robustness than smaller ones, even when they are compressed to a similar size as the smaller networks. Furthermore, compact networks obtained by knowledge distillation are much more robust to distribution shifts than pruned networks. Finally, post-training quantization is a reliable method for achieving significant robustness to distribution shifts, and it outperforms both pruned and distilled models in terms of robustness.
Electronic health records (EHRs) house crucial patient data in clinical notes. As these notes grow in volume and complexity, manual extraction becomes challenging. This work introduces a natural language interface using large language models (LLMs) for dynamic question-answering on clinical notes. Our chatbot, powered by Langchain and transformer-based LLMs, allows users to query in natural language, receiving relevant answers from clinical notes. Experiments, utilizing various embedding models and advanced LLMs, show Wizard Vicuna's superior accuracy, albeit with high compute demands. Model optimization, including weight quantization, improves latency by approximately 48 times. Promising results indicate potential, yet challenges such as model hallucinations and limited diverse medical case evaluations remain. Addressing these gaps is crucial for unlocking the value in clinical notes and advancing AI-driven clinical decision-making.
Quantization techniques commonly reduce the inference costs of neural networks by restricting the precision of weights and activations. Recent studies show that also reducing the precision of the accumulator can further improve hardware efficiency at the risk of numerical overflow, which introduces arithmetic errors that can degrade model accuracy. To avoid numerical overflow while maintaining accuracy, recent work proposed accumulator-aware quantization (A2Q), a quantization-aware training method that constrains model weights during training to safely use a target accumulator bit width during inference. Although this shows promise, we demonstrate that A2Q relies on an overly restrictive constraint and a sub-optimal weight initialization strategy that each introduce superfluous quantization error. To address these shortcomings, we introduce: (1) an improved bound that alleviates accumulator constraints without compromising overflow avoidance; and (2) a new strategy for initializing quantized weights from pre-trained floating-point checkpoints. We combine these contributions with weight normalization to introduce A2Q+. We support our analysis with experiments that show A2Q+ significantly improves the trade-off between accumulator bit width and model accuracy and characterize new trade-offs that arise as a consequence of accumulator constraints.
Sleep Stage Classification (SSC) is a labor-intensive task, requiring experts to examine hours of electrophysiological recordings for manual classification. This is a limiting factor when it comes to leveraging sleep stages for therapeutic purposes. With increasing affordability and expansion of wearable devices, automating SSC may enable deployment of sleep-based therapies at scale. Deep Learning has gained increasing attention as a potential method to automate this process. Previous research has shown accuracy comparable to manual expert scores. However, previous approaches require sizable amount of memory and computational resources. This constrains the ability to classify in real time and deploy models on the edge. To address this gap, we aim to provide a model capable of predicting sleep stages in real-time, without requiring access to external computational sources (e.g., mobile phone, cloud). The algorithm is power efficient to enable use on embedded battery powered systems. Our compact sleep stage classifier can be deployed on most off-the-shelf microcontrollers (MCU) with constrained hardware settings. This is due to the memory footprint of our approach requiring significantly fewer operations. The model was tested on three publicly available data bases and achieved performance comparable to the state of the art, whilst reducing model complexity by orders of magnitude (up to 280 times smaller compared to state of the art). We further optimized the model with quantization of parameters to 8 bits with only an average drop of 0.95% in accuracy. When implemented in firmware, the quantized model achieves a latency of 1.6 seconds on an Arm CortexM4 processor, allowing its use for on-line SSC-based therapies.
Many common types of data can be represented as functions that map coordinates to signal values, such as pixel locations to RGB values in the case of an image. Based on this view, data can be compressed by overfitting a compact neural network to its functional representation and then encoding the network weights. However, most current solutions for this are inefficient, as quantization to low-bit precision substantially degrades the reconstruction quality. To address this issue, we propose overfitting variational Bayesian neural networks to the data and compressing an approximate posterior weight sample using relative entropy coding instead of quantizing and entropy coding it. This strategy enables direct optimization of the rate-distortion performance by minimizing the $\beta$-ELBO, and target different rate-distortion trade-offs for a given network architecture by adjusting $\beta$. Moreover, we introduce an iterative algorithm for learning prior weight distributions and employ a progressive refinement process for the variational posterior that significantly enhances performance. Experiments show that our method achieves strong performance on image and audio compression while retaining simplicity.
On-device learning remains a formidable challenge, especially when dealing with resource-constrained devices that have limited computational capabilities. This challenge is primarily rooted in two key issues: first, the memory available on embedded devices is typically insufficient to accommodate the memory-intensive back-propagation algorithm, which often relies on floating-point precision. Second, the development of learning algorithms on models with extreme quantization levels, such as Binary Neural Networks (BNNs), is critical due to the drastic reduction in bit representation. In this study, we propose a solution that combines recent advancements in the field of Continual Learning (CL) and Binary Neural Networks to enable on-device training while maintaining competitive performance. Specifically, our approach leverages binary latent replay (LR) activations and a novel quantization scheme that significantly reduces the number of bits required for gradient computation. The experimental validation demonstrates a significant accuracy improvement in combination with a noticeable reduction in memory requirement, confirming the suitability of our approach in expanding the practical applications of deep learning in real-world scenarios.
The stringent requirements for the Deep Neural Networks (DNNs) accelerator's reliability stand along with the need for reducing the computational burden on the hardware platforms, i.e. reducing the energy consumption and execution time as well as increasing the efficiency of DNN accelerators. Moreover, the growing demand for specialized DNN accelerators with tailored requirements, particularly for safety-critical applications, necessitates a comprehensive design space exploration to enable the development of efficient and robust accelerators that meet those requirements. Therefore, the trade-off between hardware performance, i.e. area and delay, and the reliability of the DNN accelerator implementation becomes critical and requires tools for analysis. This paper presents a comprehensive methodology for exploring and enabling a holistic assessment of the trilateral impact of quantization on model accuracy, activation fault reliability, and hardware efficiency. A fully automated framework is introduced that is capable of applying various quantization-aware techniques, fault injection, and hardware implementation, thus enabling the measurement of hardware parameters. Moreover, this paper proposes a novel lightweight protection technique integrated within the framework to ensure the dependable deployment of the final systolic-array-based FPGA implementation. The experiments on established benchmarks demonstrate the analysis flow and the profound implications of quantization on reliability, hardware performance, and network accuracy, particularly concerning the transient faults in the network's activations.
The development of deep learning algorithms has extensively empowered humanity's task automatization capacity. However, the huge improvement in the performance of these models is highly correlated with their increasing level of complexity, limiting their usefulness in human-oriented applications, which are usually deployed in resource-constrained devices. This led to the development of compression techniques that drastically reduce the computational and memory costs of deep learning models without significant performance degradation. This paper aims to systematize the current literature on this topic by presenting a comprehensive survey of model compression techniques in biometrics applications, namely quantization, knowledge distillation and pruning. We conduct a critical analysis of the comparative value of these techniques, focusing on their advantages and disadvantages and presenting suggestions for future work directions that can potentially improve the current methods. Additionally, we discuss and analyze the link between model bias and model compression, highlighting the need to direct compression research toward model fairness in future works.
Stride determines the distance between adjacent filter positions as the filter moves across the input. A fixed stride causes important information contained in the image can not be captured, so that important information is not classified. Therefore, in previous research, the DiffStride Method was applied, namely the Strided Convolution Method with which it can learn its own stride value. Severe Quantization and a constraining lower bound on preserved information are arises with Max Pooling Downsampling Method. Spectral Pooling reduce the constraint lower bound on preserved information by cutting off the representation in the frequency domain. In this research a CNN Model is proposed with the Downsampling Learnable Stride Technique performed by Backpropagation combined with the Spectral Pooling Technique. Diffstride and Spectral Pooling techniques are expected to maintain most of the information contained in the image. In this study, we compare the Hybrid Method, which is a combined implementation of Spectral Pooling and DiffStride against the Baseline Method, which is the DiffStride implementation on ResNet 18. The accuracy result of the DiffStride combination with Spectral Pooling improves over DiffStride which is baseline method by 0.0094. This shows that the Hybrid Method can maintain most of the information by cutting of the representation in the frequency domain and determine the stride of the learning result through Backpropagation.