Disaster recovery and management present significant challenges, particularly in unstable environments and hard-to-reach terrains. These difficulties can be overcome by employing unmanned aerial vehicles (UAVs) equipped with onboard embedded platforms and camera sensors. In this work, we address the critical need for accurate and timely disaster detection by enabling onboard aerial imagery processing and avoiding connectivity, privacy, and latency issues despite the challenges posed by limited onboard hardware resources. We propose a UAV-assisted edge framework for real-time disaster management, leveraging our proposed model optimized for real-time aerial image classification. The optimization of the model employs post-training quantization techniques. For real-world disaster scenarios, we introduce a novel dataset, DisasterEye, featuring UAV-captured disaster scenes as well as ground-level images taken by individuals on-site. Experimental results demonstrate the effectiveness of our model, achieving high accuracy with reduced inference latency and memory usage on resource-constrained devices. The framework's scalability and adaptability make it a robust solution for real-time disaster detection on resource-limited UAV platforms.
The emergence of neural network capabilities invariably leads to a significant surge in computational demands due to expanding model sizes and increased computational complexity. To reduce model size and lower inference costs, recent research has focused on simplifying models and designing hardware accelerators using low-bit quantization. However, due to numerical representation limits, scalar quantization cannot reduce bit width lower than 1-bit, diminishing its benefits. To break through these limitations, we introduce LUT-DLA, a Look-Up Table (LUT) Deep Learning Accelerator Framework that utilizes vector quantization to convert neural network models into LUTs, achieving extreme low-bit quantization. The LUT-DLA framework facilitates efficient and cost-effective hardware accelerator designs and supports the LUTBoost algorithm, which helps to transform various DNN models into LUT-based models via multistage training, drastically cutting both computational and hardware overhead. Additionally, through co-design space exploration, LUT-DLA assesses the impact of various model and hardware parameters to fine-tune hardware configurations for different application scenarios, optimizing performance and efficiency. Our comprehensive experiments show that LUT-DLA achieves improvements in power efficiency and area efficiency with gains of $1.4$~$7.0\times$ and $1.5$~$146.1\times$, respectively, while maintaining only a modest accuracy drop. For CNNs, accuracy decreases by $0.1\%$~$3.1\%$ using the $L_2$ distance similarity, $0.1\%$~$3.4\%$ with the $L_1$ distance similarity, and $0.1\%$~$3.8\%$ when employing the Chebyshev distance similarity. For transformer-based models, the accuracy drop ranges from $1.4\%$ to $3.0\%$.
Retrieval-augmented generation (RAG) is a promising technique that has shown great potential in addressing some of the limitations of large language models (LLMs). LLMs have two major limitations: they can contain outdated information due to their training data, and they can generate factually inaccurate responses, a phenomenon known as hallucinations. RAG aims to mitigate these issues by leveraging a database of relevant documents, which are stored as embedding vectors in a high-dimensional space. However, one of the challenges of using high-dimensional embeddings is that they require a significant amount of memory to store. This can be a major issue, especially when dealing with large databases of documents. To alleviate this problem, we propose the use of 4-bit quantization to store the embedding vectors. This involves reducing the precision of the vectors from 32-bit floating-point numbers to 4-bit integers, which can significantly reduce the memory requirements. Our approach has several benefits. Firstly, it significantly reduces the memory storage requirements of the high-dimensional vector database, making it more feasible to deploy RAG systems in resource-constrained environments. Secondly, it speeds up the searching process, as the reduced precision of the vectors allows for faster computation. Our code is available at https://github.com/taeheej/4bit-Quantization-in-Vector-Embedding-for-RAG
Point cloud completion aims to reconstruct complete 3D shapes from partial 3D point clouds. With advancements in deep learning techniques, various methods for point cloud completion have been developed. Despite achieving encouraging results, a significant issue remains: these methods often overlook the variability in point clouds sampled from a single 3D object surface. This variability can lead to ambiguity and hinder the achievement of more precise completion results. Therefore, in this study, we introduce a novel point cloud completion network, namely Dual-Codebook Point Completion Network (DC-PCN), following an encder-decoder pipeline. The primary objective of DC-PCN is to formulate a singular representation of sampled point clouds originating from the same 3D surface. DC-PCN introduces a dual-codebook design to quantize point-cloud representations from a multilevel perspective. It consists of an encoder-codebook and a decoder-codebook, designed to capture distinct point cloud patterns at shallow and deep levels. Additionally, to enhance the information flow between these two codebooks, we devise an information exchange mechanism. This approach ensures that crucial features and patterns from both shallow and deep levels are effectively utilized for completion. Extensive experiments on the PCN, ShapeNet\_Part, and ShapeNet34 datasets demonstrate the state-of-the-art performance of our method.
Diffusion models achieve superior performance in image generation tasks. However, it incurs significant computation overheads due to its iterative structure. To address these overheads, we analyze this iterative structure and observe that adjacent time steps in diffusion models exhibit high value similarity, leading to narrower differences between consecutive time steps. We adapt these characteristics to a quantized diffusion model and reveal that the majority of these differences can be represented with reduced bit-width, and even zero. Based on our observations, we propose the Ditto algorithm, a difference processing algorithm that leverages temporal similarity with quantization to enhance the efficiency of diffusion models. By exploiting the narrower differences and the distributive property of layer operations, it performs full bit-width operations for the initial time step and processes subsequent steps with temporal differences. In addition, Ditto execution flow optimization is designed to mitigate the memory overhead of temporal difference processing, further boosting the efficiency of the Ditto algorithm. We also design the Ditto hardware, a specialized hardware accelerator, fully exploiting the dynamic characteristics of the proposed algorithm. As a result, the Ditto hardware achieves up to 1.5x speedup and 17.74% energy saving compared to other accelerators.
We consider the problem of Multiple-Input Single- Output (MISO) communication with limited feedback, where the transmitter relies on a limited number of bits associated with the channel state information (CSI), available at the receiver (CSIR) but not at the transmitter (no CSIT), sent via the feedback link. We demonstrate how character-polynomial (CP) codes, a class of analog subspace codes (also, referred to as Grassmann codes) can be used for the corresponding quantization problem in the Grassmann space. The proposed CP codebook-based precoding design allows for a smooth trade-off between the number of feedback bits and the beamforming gain, by simply adjusting the rate of the underlying CP code. We present a theoretical upper bound on the mean squared quantization error of the CP codebook, and utilize it to upper bound the resulting distortion as the normalized gap between the CP codebook beamforming gain and the baseline equal gain transmission (EGT) with perfect CSIT. We further show that the distortion vanishes asymptotically. The results are also confirmed via simulations for different types of fading models in the MISO system and various parameters.
Koopman-based lifted linear identification have been widely used for data-driven prediction and model predictive control (MPC) of nonlinear systems. It has found applications in flow-control, soft robotics, and unmanned aerial vehicles (UAV). For autonomous systems, this system identification method works by embedding the nonlinear system in a higher-dimensional linear space and computing a finite-dimensional approximation of the corresponding Koopman operator with the Extended Dynamic Mode Decomposition (EDMD) algorithm. EDMD is a data-driven algorithm that estimates an approximate linear system by lifting the state data-snapshots via nonlinear dictionary functions. For control systems, EDMD is further modified to utilize both state and control data-snapshots to estimate a lifted linear predictor with control input. This article investigates how the estimation process is affected when the data is quantized. Specifically, we examine the fundamental connection between estimates of the linear predictor matrices obtained from unquantized data and those from quantized data via modified EDMD. Furthermore, using the law of large numbers, we demonstrate that, under a large data regime, the quantized estimate can be considered a regularized version of the unquantized estimate. We also explore the relationship between the two estimates in the finite data regime. We further analyze the effect of nonlinear lifting functions on this regularization due to quantization. The theory is validated through repeated numerical experiments conducted on several control systems. The effect of quantization on the MPC performance is also demonstrated.
The restoration of nonlinearly distorted audio signals, alongside the identification of the applied memoryless nonlinear operation, is studied. The paper focuses on the difficult but practically important case in which both the nonlinearity and the original input signal are unknown. The proposed method uses a generative diffusion model trained unconditionally on guitar or speech signals to jointly model and invert the nonlinear system at inference time. Both the memoryless nonlinear function model and the restored audio signal are obtained as output. Successful example case studies are presented including inversion of hard and soft clipping, digital quantization, half-wave rectification, and wavefolding nonlinearities. Our results suggest that, out of the nonlinear functions tested here, the cubic Catmull-Rom spline is best suited to approximating these nonlinearities. In the case of guitar recordings, comparisons with informed and supervised methods show that the proposed blind method is at least as good as they are in terms of objective metrics. Experiments on distorted speech show that the proposed blind method outperforms general-purpose speech enhancement techniques and restores the original voice quality. The proposed method can be applied to audio effects modeling, restoration of music and speech recordings, and characterization of analog recording media.
As Large Language Models (LLMs) become increasingly computationally complex, developing efficient deployment strategies, such as quantization, becomes crucial. State-of-the-art Post-training Quantization (PTQ) techniques often rely on calibration processes to maintain the accuracy of these models. However, while these calibration techniques can enhance performance in certain domains, they may not be as effective in others. This paper aims to draw attention to robust statistical approaches that can mitigate such issues. We propose a weight-adaptive PTQ method that can be considered a precursor to calibration-based PTQ methods, guiding the quantization process to preserve the distribution of weights by minimizing the Kullback-Leibler divergence between the quantized weights and the originally trained weights. This minimization ensures that the quantized model retains the Shannon information content of the original model to a great extent, guaranteeing robust and efficient deployment across many tasks. As such, our proposed approach can perform on par with most common calibration-based PTQ methods, establishing a new pre-calibration step for further adjusting the quantized weights with calibration. We show that our pre-calibration results achieve the same accuracy as some existing calibration-based PTQ methods on various LLMs.
Transformer architectures have become the standard neural network model for various machine learning applications including natural language processing and computer vision. However, the compute and memory requirements introduced by transformer models make them challenging to adopt for edge applications. Furthermore, fine-tuning pre-trained transformers (e.g., foundation models) is a common task to enhance the model's predictive performance on specific tasks/applications. Existing transformer accelerators are oblivious to complexities introduced by fine-tuning. In this paper, we propose the design of a three-dimensional (3D) heterogeneous architecture referred to as Atleus that incorporates heterogeneous computing resources specifically optimized to accelerate transformer models for the dual purposes of fine-tuning and inference. Specifically, Atleus utilizes non-volatile memory and systolic array for accelerating transformer computational kernels using an integrated 3D platform. Moreover, we design a suitable NoC to achieve high performance and energy efficiency. Finally, Atleus adopts an effective quantization scheme to support model compression. Experimental results demonstrate that Atleus outperforms existing state-of-the-art by up to 56x and 64.5x in terms of performance and energy efficiency respectively
LiDAR is a crucial sensor in autonomous driving, commonly used alongside cameras. By exploiting this camera-LiDAR setup and recent advances in image representation learning, prior studies have shown the promising potential of image-to-LiDAR distillation. These prior arts focus on the designs of their own losses to effectively distill the pre-trained 2D image representations into a 3D model. However, the other parts of the designs have been surprisingly unexplored. We find that fundamental design elements, e.g., the LiDAR coordinate system, quantization according to the existing input interface, and data utilization, are more critical than developing loss functions, which have been overlooked in prior works. In this work, we show that simple fixes to these designs notably outperform existing methods by 16% in 3D semantic segmentation on the nuScenes dataset and 13% in 3D object detection on the KITTI dataset in downstream task performance. We focus on overlooked design choices along the spatial and temporal axes. Spatially, prior work has used cylindrical coordinate and voxel sizes without considering their side effects yielded with a commonly deployed sparse convolution layer input interface, leading to spatial quantization errors in 3D models. Temporally, existing work has avoided cumbersome data curation by discarding unsynced data, limiting the use to only the small portion of data that is temporally synced across sensors. We analyze these effects and propose simple solutions for each overlooked aspect.
Retrievers, which form one of the most important recommendation stages, are responsible for efficiently selecting possible positive samples to the later stages under strict latency limitations. Because of this, large-scale systems always rely on approximate calculations and indexes to roughly shrink candidate scale, with a simple ranking model. Considering simple models lack the ability to produce precise predictions, most of the existing methods mainly focus on incorporating complicated ranking models. However, another fundamental problem of index effectiveness remains unresolved, which also bottlenecks complication. In this paper, we propose a novel index structure: streaming Vector Quantization model, as a new generation of retrieval paradigm. Streaming VQ attaches items with indexes in real time, granting it immediacy. Moreover, through meticulous verification of possible variants, it achieves additional benefits like index balancing and reparability, enabling it to support complicated ranking models as existing approaches. As a lightweight and implementation-friendly architecture, streaming VQ has been deployed and replaced all major retrievers in Douyin and Douyin Lite, resulting in remarkable user engagement gain.
Unlocking the potential of Large Language Models (LLMs) in data classification represents a promising frontier in natural language processing. In this work, we evaluate the performance of different LLMs in comparison with state-of-the-art deep-learning and machine-learning models, in two different classification scenarios: i) the classification of employees' working locations based on job reviews posted online (multiclass classification), and 2) the classification of news articles as fake or not (binary classification). Our analysis encompasses a diverse range of language models differentiating in size, quantization, and architecture. We explore the impact of alternative prompting techniques and evaluate the models based on the weighted F1-score. Also, we examine the trade-off between performance (F1-score) and time (inference response time) for each language model to provide a more nuanced understanding of each model's practical applicability. Our work reveals significant variations in model responses based on the prompting strategies. We find that LLMs, particularly Llama3 and GPT-4, can outperform traditional methods in complex classification tasks, such as multiclass classification, though at the cost of longer inference times. In contrast, simpler ML models offer better performance-to-time trade-offs in simpler binary classification tasks.
Diffusion models have achieved cutting-edge performance in image generation. However, their lengthy denoising process and computationally intensive score estimation network impede their scalability in low-latency and resource-constrained scenarios. Post-training quantization (PTQ) compresses and accelerates diffusion models without retraining, but it inevitably introduces additional quantization noise, resulting in mean and variance deviations. In this work, we propose D2-DPM, a dual denoising mechanism aimed at precisely mitigating the adverse effects of quantization noise on the noise estimation network. Specifically, we first unravel the impact of quantization noise on the sampling equation into two components: the mean deviation and the variance deviation. The mean deviation alters the drift coefficient of the sampling equation, influencing the trajectory trend, while the variance deviation magnifies the diffusion coefficient, impacting the convergence of the sampling trajectory. The proposed D2-DPM is thus devised to denoise the quantization noise at each time step, and then denoise the noisy sample through the inverse diffusion iterations. Experimental results demonstrate that D2-DPM achieves superior generation quality, yielding a 1.42 lower FID than the full-precision model while achieving 3.99x compression and 11.67x bit-operation acceleration.
Dataset distillation aims to find a synthetic training set such that training on the synthetic data achieves similar performance to training on real data, with orders of magnitude less computational requirements. Existing methods can be broadly categorized as either bi-level optimization problems that have neural network training heuristics as the lower level problem, or disentangled methods that bypass the bi-level optimization by matching distributions of data. The latter method has the major advantages of speed and scalability in terms of size of both training and distilled datasets. We demonstrate that when equipped with an encoder-decoder structure, the empirically successful disentangled methods can be reformulated as an optimal quantization problem, where a finite set of points is found to approximate the underlying probability measure by minimizing the expected projection distance. In particular, we link existing disentangled dataset distillation methods to the classical optimal quantization and Wasserstein barycenter problems, demonstrating consistency of distilled datasets for diffusion-based generative priors. We propose a simple extension of the state-of-the-art data distillation method D4M, achieving better performance on the ImageNet-1K dataset with trivial additional computation, and state-of-the-art performance in higher image-per-class settings.
The increasing computational and memory demands in deep learning present significant challenges, especially in resource-constrained environments. We introduce a zero-order quantized optimization (ZOQO) method designed for training models with quantized parameters and operations. Our approach leverages zero-order approximations of the gradient sign and adapts the learning process to maintain the parameters' quantization without the need for full-precision gradient calculations. We demonstrate the effectiveness of ZOQO through experiments in fine-tuning of large language models and black-box adversarial attacks. Despite the limitations of zero-order and quantized operations training, our method achieves competitive performance compared to full-precision methods, highlighting its potential for low-resource environments.
In this paper, we propose Mix-QViT, an explainability-driven MPQ framework that systematically allocates bit-widths to each layer based on two criteria: layer importance, assessed via Layer-wise Relevance Propagation (LRP), which identifies how much each layer contributes to the final classification, and quantization sensitivity, determined by evaluating the performance impact of quantizing each layer at various precision levels while keeping others layers at a baseline. Additionally, for post-training quantization (PTQ), we introduce a clipped channel-wise quantization method designed to reduce the effects of extreme outliers in post-LayerNorm activations by removing severe inter-channel variations. We validate our approach by applying Mix-QViT to ViT, DeiT, and Swin Transformer models across multiple datasets. Our experimental results for PTQ demonstrate that both fixed-bit and mixed-bit methods outperform existing techniques, particularly at 3-bit, 4-bit, and 6-bit precision. Furthermore, in quantization-aware training, Mix-QViT achieves superior performance with 2-bit mixed-precision.
Quantizing the weights of a neural network has two steps: (1) Finding a good low bit-complexity representation for weights (which we call the quantization grid) and (2) Rounding the original weights to values in the quantization grid. In this paper, we study the problem of rounding optimally given any quantization grid. The simplest and most commonly used way to round is Round-to-Nearest (RTN). By rounding in a data-dependent way instead, one can improve the quality of the quantized model significantly. We study the rounding problem from the lens of \emph{discrepancy theory}, which studies how well we can round a continuous solution to a discrete solution without affecting solution quality too much. We prove that given $m=\mathrm{poly}(1/\epsilon)$ samples from the data distribution, we can round all but $O(m)$ model weights such that the expected approximation error of the quantized model on the true data distribution is $\le \epsilon$ as long as the space of gradients of the original model is approximately low rank (which we empirically validate). Our proof, which is algorithmic, inspired a simple and practical rounding algorithm called \emph{DiscQuant}. In our experiments, we demonstrate that DiscQuant significantly improves over the prior state-of-the-art rounding method called GPTQ and the baseline RTN over a range of benchmarks on Phi3mini-3.8B and Llama3.1-8B. For example, rounding Phi3mini-3.8B to a fixed quantization grid with 3.25 bits per parameter using DiscQuant gets 64\% accuracy on the GSM8k dataset, whereas GPTQ achieves 54\% and RTN achieves 31\% (the original model achieves 84\%). We make our code available at https://github.com/jerry-chee/DiscQuant.
Mixed-precision quantization methods have been proposed to reduce model size while minimizing accuracy degradation. However, existing studies require retraining and do not consider the computational overhead and intermediate representations (IR) generated during the compilation process, limiting their application at the compiler level. This computational overhead refers to the runtime latency caused by frequent quantization and dequantization operations during inference. Performing these operations at the individual operator level causes significant runtime delays. To address these issues, we propose QuantuneV2, a compiler-based mixed-precision quantization method designed for practical embedded AI applications. QuantuneV2 performs inference only twice, once before quantization and once after quantization, and operates with a computational complexity of O(n) that increases linearly with the number of model parameters. We also made the sensitivity analysis more stable by using local metrics like weights, activation values, the Signal to Quantization Noise Ratio, and the Mean Squared Error. We also cut down on computational overhead by choosing the best IR and using operator fusion. Experimental results show that QuantuneV2 achieved up to a 10.28 percent improvement in accuracy and a 12.52 percent increase in speed compared to existing methods across five models: ResNet18v1, ResNet50v1, SqueezeNetv1, VGGNet, and MobileNetv2. This demonstrates that QuantuneV2 enhances model performance while maintaining computational efficiency, making it suitable for deployment in embedded AI environments.
Deploying LLMs on edge devices presents serious technical challenges. Memory elasticity is crucial for edge devices with unified memory, where memory is shared and fluctuates dynamically. Existing solutions suffer from either poor transition granularity or high storage costs. We propose FlexQuant, a novel elasticity framework that generates an ensemble of quantized models, providing an elastic hosting solution with 15x granularity improvement and 10x storage reduction compared to SoTA methods. FlexQuant works with most quantization methods and creates a family of trade-off options under various storage limits through our pruning method. It brings great performance and flexibility to the edge deployment of LLMs.