Due to the scarcity and unpredictable nature of defect samples, industrial anomaly detection (IAD) predominantly employs unsupervised learning. However, all unsupervised IAD methods face a common challenge: the inherent bias in normal samples, which causes models to focus on variable regions while overlooking potential defects in invariant areas. To effectively overcome this, it is essential to decompose and recalibrate attention, guiding the model to suppress irrelevant variations and concentrate on subtle, defect-susceptible areas. In this paper, we propose Recalibrating Attention of Industrial Anomaly Detection (RAAD), a framework that systematically decomposes and recalibrates attention maps. RAAD employs a two-stage process: first, it reduces attention bias through quantization, and second, it fine-tunes defect-prone regions for improved sensitivity. Central to this framework is Hierarchical Quantization Scoring (HQS), which dynamically allocates bit-widths across layers based on their anomaly detection contributions. HQS dynamically adjusts bit-widths based on the hierarchical nature of attention maps, compressing lower layers that produce coarse and noisy attention while preserving deeper layers with sharper, defect-focused attention. This approach optimizes both computational efficiency and the model' s sensitivity to anomalies. We validate the effectiveness of RAAD on 32 datasets using a single 3090ti. Experiments demonstrate that RAAD, balances the complexity and expressive power of the model, enhancing its anomaly detection capability.
We present a novel machine learning (ML) method to accelerate conservative-to-primitive inversion, focusing on hybrid piecewise polytropic and tabulated equations of state. Traditional root-finding techniques are computationally expensive, particularly for large-scale relativistic hydrodynamics simulations. To address this, we employ feedforward neural networks (NNC2PS and NNC2PL), trained in PyTorch and optimized for GPU inference using NVIDIA TensorRT, achieving significant speedups with minimal accuracy loss. The NNC2PS model achieves $ L_1 $ and $ L_\infty $ errors of $ 4.54 \times 10^{-7} $ and $ 3.44 \times 10^{-6} $, respectively, while the NNC2PL model exhibits even lower error values. TensorRT optimization with mixed-precision deployment substantially accelerates performance compared to traditional root-finding methods. Specifically, the mixed-precision TensorRT engine for NNC2PS achieves inference speeds approximately 400 times faster than a traditional single-threaded CPU implementation for a dataset size of 1,000,000 points. Ideal parallelization across an entire compute node in the Delta supercomputer (Dual AMD 64 core 2.45 GHz Milan processors; and 8 NVIDIA A100 GPUs with 40 GB HBM2 RAM and NVLink) predicts a 25-fold speedup for TensorRT over an optimally-parallelized numerical method when processing 8 million data points. Moreover, the ML method exhibits sub-linear scaling with increasing dataset sizes. We release the scientific software developed, enabling further validation and extension of our findings. This work underscores the potential of ML, combined with GPU optimization and model quantization, to accelerate conservative-to-primitive inversion in relativistic hydrodynamics simulations.
Large language model (LLM) inference demands significant amount of computation and memory, especially in the key attention mechanism. While techniques, such as quantization and acceleration algorithms, like FlashAttention, have improved efficiency of the overall inference, they address different aspects of the problem: quantization focuses on weight-activation operations, while FlashAttention improves execution but requires high-precision formats. Recent Key-value (KV) cache quantization reduces memory bandwidth but still needs floating-point dequantization for attention operation. We present TurboAttention, a comprehensive approach to enable quantized execution of attention that simultaneously addresses both memory and computational efficiency. Our solution introduces two key innovations: FlashQ, a headwise attention quantization technique that enables both compression of KV cache and quantized execution of activation-activation multiplication, and Sparsity-based Softmax Approximation (SAS), which eliminates the need for dequantization to FP32 during exponentiation operation in attention. Experimental results demonstrate that TurboAttention achieves 1.2-1.8x speedup in attention, reduces the KV cache size by over 4.4x, and enables up to 2.37x maximum throughput over the FP16 baseline while outperforming state-of-the-art quantization and compression techniques across various datasets and models.
Understanding complex animal behaviors hinges on deciphering the neural activity patterns within brain circuits, making the ability to forecast neural activity crucial for developing predictive models of brain dynamics. This capability holds immense value for neuroscience, particularly in applications such as real-time optogenetic interventions. While traditional encoding and decoding methods have been used to map external variables to neural activity and vice versa, they focus on interpreting past data. In contrast, neural forecasting aims to predict future neural activity, presenting a unique and challenging task due to the spatiotemporal sparsity and complex dependencies of neural signals. Existing transformer-based forecasting methods, while effective in many domains, struggle to capture the distinctiveness of neural signals characterized by spatiotemporal sparsity and intricate dependencies. To address this challenge, we here introduce QuantFormer, a transformer-based model specifically designed for forecasting neural activity from two-photon calcium imaging data. Unlike conventional regression-based approaches, QuantFormerreframes the forecasting task as a classification problem via dynamic signal quantization, enabling more effective learning of sparse neural activation patterns. Additionally, QuantFormer tackles the challenge of analyzing multivariate signals from an arbitrary number of neurons by incorporating neuron-specific tokens, allowing scalability across diverse neuronal populations. Trained with unsupervised quantization on the Allen dataset, QuantFormer sets a new benchmark in forecasting mouse visual cortex activity. It demonstrates robust performance and generalization across various stimuli and individuals, paving the way for a foundational model in neural signal prediction.
Despite the success of CNN models on a variety of Image classification and segmentation tasks, their extensive computational and storage demands pose considerable challenges for real-world deployment on resource constrained devices. Quantization is one technique that aims to alleviate these large storage requirements and speed up the inference process by reducing the precision of model parameters to lower-bit representations. In this paper, we introduce a novel post-training quantization method for model weights. Our method finds optimal clipping thresholds and scaling factors along with mathematical guarantees that our method minimizes quantization noise. Empirical results on Real World Datasets demonstrate that our quantization scheme significantly reduces model size and computational requirements while preserving model accuracy.
This work focus on how to stabilize and lossless model compression, aiming to reduce model complexity and enhance efficiency without sacrificing performance due to compression errors. A key challenge is effectively leveraging compression errors and defining the boundaries for lossless compression to minimize model loss. i.e., compression for better. Currently, there is no systematic approach to determining this error boundary or understanding its specific impact on model performance. We propose a general \textbf{L}oss\textbf{L}ess \textbf{C}ompression theoretical framework (\textbf{LLC}), which further delineates the compression neighborhood and higher-order analysis boundaries through the total differential, thereby specifying the error range within which a model can be compressed without loss. To verify the effectiveness of LLC, we apply various compression techniques, including quantization and decomposition. Specifically, for quantization, we reformulate the classic quantization search problem as a grouped knapsack problem within the lossless neighborhood, achieving lossless quantization while improving computational efficiency. For decomposition, LLC addresses the approximation problem under low-rank constraints, automatically determining the rank for each layer and producing lossless low-rank models. We conduct extensive experiments on multiple neural network architectures on different datasets. The results show that without fancy tricks, LLC can effectively achieve lossless model compression. Our code will be made publicly.
Post-Training Quantization (PTQ) converts pre-trained Full-Precision (FP) models into quantized versions without training. While existing methods reduce size and computational costs, they also significantly degrade performance and quantization efficiency at extremely low settings due to quantization noise. We introduce a deep model series expansion framework to address this issue, enabling rapid and accurate approximation of unquantized models without calibration sets or fine-tuning. This is the first use of series expansion for neural network quantization. Specifically, our method expands the FP model into multiple low-bit basis models. To ensure accurate quantization, we develop low-bit basis model expansions at different granularities (tensor, layer, model), and theoretically confirm their convergence to the dense model, thus restoring FP model accuracy. Additionally, we design AbelianAdd/Mul operations between isomorphic models in the low-bit expansion, forming an Abelian group to ensure operation parallelism and commutativity. The experiments show that our algorithm achieves state-of-the-art performance in low-bit settings; for example, 4-bit quantization of ResNet-50 surpasses the original accuracy, reaching 77.03%. The code will be made public.
Quantization is a critical step to enable efficient LLM serving under limited resource. However, previous research observes that certain weights in the LLM, known as outliers, are significantly sensitive to quantization noises. Existing quantization methods leave these outliers as floating points or higher precisions to retain performance, posting challenges on the efficient hardware deployment of the mixed-precision model. This work investigates an alternative way to tame the sensitive weights' impact on the quantization error, by reducing the loss Hessian trace with respect to outliers through an efficient fine-tuning process. We propose Noise Perturbation Fine-tuning (NPFT), which identifies outlier weights and add random weight perturbations on the outliers as the model going through a PEFT optimization. NPFT tames the sensitivity of outlier weights so that the quantized model performance can be improved without special treatment to the outliers. When applied to OPT and LLaMA models, our NPFT method achieves stable performance improvements for both uniform and non-uniform quantizers, while also offering better inference efficiency. Notably, the simplest RTN can achieve performance on par with GPTQ using our NPFT on LLaMA2-7B-4bits benchmark.
Recent advances in Large Language Models (LLMs) have demonstrated promising performance in sequential recommendation tasks, leveraging their superior language understanding capabilities. However, existing LLM-based recommendation approaches predominantly focus on modeling item-level co-occurrence patterns while failing to adequately capture user-level personalized preferences. This is problematic since even users who display similar behavioral patterns (e.g., clicking or purchasing similar items) may have fundamentally different underlying interests. To alleviate this problem, in this paper, we propose ULMRec, a framework that effectively integrates user personalized preferences into LLMs for sequential recommendation. Considering there has the semantic gap between item IDs and LLMs, we replace item IDs with their corresponding titles in user historical behaviors, enabling the model to capture the item semantics. For integrating the user personalized preference, we design two key components: (1) user indexing: a personalized user indexing mechanism that leverages vector quantization on user reviews and user IDs to generate meaningful and unique user representations, and (2) alignment tuning: an alignment-based tuning stage that employs comprehensive preference alignment tasks to enhance the model's capability in capturing personalized information. Through this design, ULMRec achieves deep integration of language semantics with user personalized preferences, facilitating effective adaptation to recommendation. Extensive experiments on two public datasets demonstrate that ULMRec significantly outperforms existing methods, validating the effectiveness of our approach.
As a paradigm of distributed machine learning, federated learning typically requires all edge devices to train a complete model locally. However, with the increasing scale of artificial intelligence models, the limited resources on edge devices often become a bottleneck for efficient fine-tuning. To address this challenge, federated split learning (FedSL) implements collaborative training across the edge devices and the server through model splitting. In this paper, we propose a lightweight FedSL scheme, that further alleviates the training burden on resource-constrained edge devices by pruning the client-side model dynamicly and using quantized gradient updates to reduce computation overhead. Additionally, we apply random dropout to the activation values at the split layer to reduce communication overhead. We conduct theoretical analysis to quantify the convergence performance of the proposed scheme. Finally, simulation results verify the effectiveness and advantages of the proposed lightweight FedSL in wireless network environments.
Effective compression technology is crucial for 3DGS to adapt to varying storage and transmission conditions. However, existing methods fail to address size constraints while maintaining optimal quality. In this paper, we introduce SizeGS, a framework that compresses 3DGS within a specified size budget while optimizing visual quality. We start with a size estimator to establish a clear relationship between file size and hyperparameters. Leveraging this estimator, we incorporate mixed precision quantization (MPQ) into 3DGS attributes, structuring MPQ in two hierarchical level -- inter-attribute and intra-attribute -- to optimize visual quality under the size constraint. At the inter-attribute level, we assign bit-widths to each attribute channel by formulating the combinatorial optimization as a 0-1 integer linear program, which can be efficiently solved. At the intra-attribute level, we divide each attribute channel into blocks of vectors, quantizing each vector based on the optimal bit-width derived at the inter-attribute level. Dynamic programming determines block lengths. Using the size estimator and MPQ, we develop a calibrated algorithm to identify optimal hyperparameters in just 10 minutes, achieving a 1.69$\times$ efficiency increase with quality comparable to state-of-the-art methods.
Recent advancements in high-fidelity dynamic scene reconstruction have leveraged dynamic 3D Gaussians and 4D Gaussian Splatting for realistic scene representation. However, to make these methods viable for real-time applications such as AR/VR, gaming, and rendering on low-power devices, substantial reductions in memory usage and improvements in rendering efficiency are required. While many state-of-the-art methods prioritize lightweight implementations, they struggle in handling scenes with complex motions or long sequences. In this work, we introduce Temporally Compressed 3D Gaussian Splatting (TC3DGS), a novel technique designed specifically to effectively compress dynamic 3D Gaussian representations. TC3DGS selectively prunes Gaussians based on their temporal relevance and employs gradient-aware mixed-precision quantization to dynamically compress Gaussian parameters. It additionally relies on a variation of the Ramer-Douglas-Peucker algorithm in a post-processing step to further reduce storage by interpolating Gaussian trajectories across frames. Our experiments across multiple datasets demonstrate that TC3DGS achieves up to 67$\times$ compression with minimal or no degradation in visual quality.
Research on loss surface geometry, such as Sharpness-Aware Minimization (SAM), shows that flatter minima improve generalization. Recent studies further reveal that flatter minima can also reduce the domain generalization (DG) gap. However, existing flatness-based DG techniques predominantly operate within a full-precision training process, which is impractical for deployment on resource-constrained edge devices that typically rely on lower bit-width representations (e.g., 4 bits, 3 bits). Consequently, low-precision quantization-aware training is critical for optimizing these techniques in real-world applications. In this paper, we observe a significant degradation in performance when applying state-of-the-art DG-SAM methods to quantized models, suggesting that current approaches fail to preserve generalizability during the low-precision training process. To address this limitation, we propose a novel Gradient-Adaptive Quantization-Aware Training (GAQAT) framework for DG. Our approach begins by identifying the scale-gradient conflict problem in low-precision quantization, where the task loss and smoothness loss induce conflicting gradients for the scaling factors of quantizers, with certain layers exhibiting opposing gradient directions. This conflict renders the optimization of quantized weights highly unstable. To mitigate this, we further introduce a mechanism to quantify gradient inconsistencies and selectively freeze the gradients of scaling factors, thereby stabilizing the training process and enhancing out-of-domain generalization. Extensive experiments validate the effectiveness of the proposed GAQAT framework. On PACS, our 3-bit and 4-bit models outperform direct DG-QAT integration by up to 4.5%. On DomainNet, the 4-bit model achieves near-lossless performance compared to full precision, with improvements of 1.39% (4-bit) and 1.06% (3-bit) over the SOTA QAT baseline.
Semantic communications provide significant performance gains over traditional communications by transmitting task-relevant semantic features through wireless channels. However, most existing studies rely on end-to-end (E2E) training of neural-type encoders and decoders to ensure effective transmission of these semantic features. To enable semantic communications without relying on E2E training, this paper presents a vision transformer (ViT)-based semantic communication system with importance-aware quantization (IAQ) for wireless image transmission. The core idea of the presented system is to leverage the attention scores of a pretrained ViT model to quantify the importance levels of image patches. Based on this idea, our IAQ framework assigns different quantization bits to image patches based on their importance levels. This is achieved by formulating a weighted quantization error minimization problem, where the weight is set to be an increasing function of the attention score. Then, an optimal incremental allocation method and a low-complexity water-filling method are devised to solve the formulated problem. Our framework is further extended for realistic digital communication systems by modifying the bit allocation problem and the corresponding allocation methods based on an equivalent binary symmetric channel (BSC) model. Simulations on single-view and multi-view image classification tasks show that our IAQ framework outperforms conventional image compression methods in both error-free and realistic communication scenarios.
Diffusion models (DMs) have been significantly developed and widely used in various applications due to their excellent generative qualities. However, the expensive computation and massive parameters of DMs hinder their practical use in resource-constrained scenarios. As one of the effective compression approaches, quantization allows DMs to achieve storage saving and inference acceleration by reducing bit-width while maintaining generation performance. However, as the most extreme quantization form, 1-bit binarization causes the generation performance of DMs to face severe degradation or even collapse. This paper proposes a novel method, namely BiDM, for fully binarizing weights and activations of DMs, pushing quantization to the 1-bit limit. From a temporal perspective, we introduce the Timestep-friendly Binary Structure (TBS), which uses learnable activation binarizers and cross-timestep feature connections to address the highly timestep-correlated activation features of DMs. From a spatial perspective, we propose Space Patched Distillation (SPD) to address the difficulty of matching binary features during distillation, focusing on the spatial locality of image generation tasks and noise estimation networks. As the first work to fully binarize DMs, the W1A1 BiDM on the LDM-4 model for LSUN-Bedrooms 256$\times$256 achieves a remarkable FID of 22.74, significantly outperforming the current state-of-the-art general binarization methods with an FID of 59.44 and invalid generative samples, and achieves up to excellent 28.0 times storage and 52.7 times OPs savings. The code is available at https://github.com/Xingyu-Zheng/BiDM .
Text-to-image generation via Stable Diffusion models (SDM) have demonstrated remarkable capabilities. However, their computational intensity, particularly in the iterative denoising process, hinders real-time deployment in latency-sensitive applications. While Recent studies have explored post-training quantization (PTQ) and quantization-aware training (QAT) methods to compress Diffusion models, existing methods often overlook the consistency between results generated by quantized models and those from floating-point models. This consistency is paramount for professional applications where both efficiency and output reliability are essential. To ensure that quantized SDM generates high-quality and consistent images, we propose an efficient quantization framework for SDM. Our framework introduces a Serial-to-Parallel pipeline that simultaneously maintains training-inference consistency and ensures optimization stability. Building upon this foundation, we further develop several techniques including multi-timestep activation quantization, time information precalculation, inter-layer distillation, and selective freezing, to achieve high-fidelity generation in comparison to floating-point models while maintaining quantization efficiency. Through comprehensive evaluation across multiple Stable Diffusion variants (v1-4, v2-1, XL 1.0, and v3), our method demonstrates superior performance over state-of-the-art approaches with shorter training times. Under W4A8 quantization settings, we achieve significant improvements in both distribution similarity and visual fidelity, while preserving a high image quality.
Spiking Neural Networks (SNNs) are amenable to deployment on edge devices and neuromorphic hardware due to their lower dissipation. Recently, SNN-based transformers have garnered significant interest, incorporating attention mechanisms akin to their counterparts in Artificial Neural Networks (ANNs) while demonstrating excellent performance. However, deploying large spiking transformer models on resource-constrained edge devices such as mobile phones, still poses significant challenges resulted from the high computational demands of large uncompressed high-precision models. In this work, we introduce a novel heterogeneous quantization method for compressing spiking transformers through layer-wise quantization. Our approach optimizes the quantization of each layer using one of two distinct quantization schemes, i.e., uniform or power-of-two quantification, with mixed bit resolutions. Our heterogeneous quantization demonstrates the feasibility of maintaining high performance for spiking transformers while utilizing an average effective resolution of 3.14-3.67 bits with less than a 1% accuracy drop on DVS Gesture and CIFAR10-DVS datasets. It attains a model compression rate of 8.71x-10.19x for standard floating-point spiking transformers. Moreover, the proposed approach achieves a significant energy reduction of 5.69x, 8.72x, and 10.2x while maintaining high accuracy levels of 85.3%, 97.57%, and 80.4% on N-Caltech101, DVS-Gesture, and CIFAR10-DVS datasets, respectively.
As the data resources grow, providing recommendations that best meet the demands has become a vital requirement in business and life to overcome the information overload problem. However, building a system suggesting relevant recommendations has always been a point of debate. One of the most cost-efficient techniques in terms of producing relevant recommendations at a low complexity is Product Quantization (PQ). PQ approaches have continued developing in recent years. This system's crucial challenge is improving product quantization performance in terms of recall measures without compromising its complexity. This makes the algorithm suitable for problems that require a greater number of potentially relevant items without disregarding others, at high-speed and low-cost to keep up with traffic. This is the case of online shops where the recommendations for the purpose are important, although customers can be susceptible to scoping other products. This research proposes a fuzzy approach to perform norm-based product quantization. Type-2 Fuzzy sets (T2FSs) define the codebook allowing sub-vectors (T2FSs) to be associated with more than one element of the codebook, and next, its norm calculus is resolved by means of integration. Our method finesses the recall measure up, making the algorithm suitable for problems that require querying at most possible potential relevant items without disregarding others. The proposed method outperforms all PQ approaches such as NEQ, PQ, and RQ up to +6%, +5%, and +8% by achieving a recall of 94%, 69%, 59% in Netflix, Audio, Cifar60k datasets, respectively. More and over, computing time and complexity nearly equals the most computationally efficient existing PQ method in the state-of-the-art.
Objective: This paper aims to prompt large language models (LLMs) for clinical temporal relation extraction (CTRE) in both few-shot and fully supervised settings. Materials and Methods: This study utilizes four LLMs: Encoder-based GatorTron-Base (345M)/Large (8.9B); Decoder-based LLaMA3-8B/MeLLaMA-13B. We developed full (FFT) and parameter-efficient (PEFT) fine-tuning strategies and evaluated these strategies on the 2012 i2b2 CTRE task. We explored four fine-tuning strategies for GatorTron-Base: (1) Standard Fine-Tuning, (2) Hard-Prompting with Unfrozen LLMs, (3) Soft-Prompting with Frozen LLMs, and (4) Low-Rank Adaptation (LoRA) with Frozen LLMs. For GatorTron-Large, we assessed two PEFT strategies-Soft-Prompting and LoRA with Frozen LLMs-leveraging Quantization techniques. Additionally, LLaMA3-8B and MeLLaMA-13B employed two PEFT strategies: LoRA strategy with Quantization (QLoRA) applied to Frozen LLMs using instruction tuning and standard fine-tuning. Results: Under fully supervised settings, Hard-Prompting with Unfrozen GatorTron-Base achieved the highest F1 score (89.54%), surpassing the SOTA model (85.70%) by 3.74%. Additionally, two variants of QLoRA adapted to GatorTron-Large and Standard Fine-Tuning of GatorTron-Base exceeded the SOTA model by 2.36%, 1.88%, and 0.25%, respectively. Decoder-based models with frozen parameters outperformed their Encoder-based counterparts in this setting; however, the trend reversed in few-shot scenarios. Discussions and Conclusions: This study presented new methods that significantly improved CTRE performance, benefiting downstream tasks reliant on CTRE systems. The findings underscore the importance of selecting appropriate models and fine-tuning strategies based on task requirements and data availability. Future work will explore larger models and broader CTRE applications.
Large language models (LLMs) are notoriously memory-intensive during training, particularly with the popular AdamW optimizer. This memory burden necessitates using more or higher-end GPUs or reducing batch sizes, limiting training scalability and throughput. To address this, various memory-efficient optimizers have been proposed to reduce optimizer memory usage. However, they face critical challenges: (i) reliance on costly SVD operations; (ii) significant performance trade-offs compared to AdamW; and (iii) still substantial optimizer memory overhead to maintain competitive performance. In this work, we identify that AdamW's learning rate adaptation rule can be effectively coarsened as a structured learning rate update. Based on this insight, we propose Approximated Gradient Scaling for Memory-Efficient LLM Optimization (APOLLO), which approximates learning rate scaling using an auxiliary low-rank optimizer state based on pure random projection. This structured learning rate update rule makes APOLLO highly tolerant to further memory reductions while delivering comparable pre-training performance. Even its rank-1 variant, APOLLO-Mini, achieves superior pre-training performance compared to AdamW with SGD-level memory costs. Extensive experiments demonstrate that the APOLLO series performs on-par with or better than AdamW, while achieving greater memory savings by nearly eliminating the optimization states of AdamW. These savings provide significant system-level benefits: (1) Enhanced Throughput: 3x throughput on an 8xA100-80GB setup compared to AdamW by supporting 4x larger batch sizes. (2) Improved Model Scalability: Pre-training LLaMA-13B with naive DDP on A100-80GB GPUs without system-level optimizations. (3) Low-End GPU Friendly Pre-training: Pre-training LLaMA-7B on a single GPU using less than 12 GB of memory with weight quantization.