Implicit Neural Representations (INRs), also known as neural fields, have emerged as a powerful paradigm in deep learning, parameterizing continuous spatial fields using coordinate-based neural networks. In this paper, we propose \textbf{PICO}, an INR-based framework for static point cloud compression. Unlike prevailing encoder-decoder paradigms, we decompose the point cloud compression task into two separate stages: geometry compression and attribute compression, each with distinct INR optimization objectives. Inspired by Kolmogorov-Arnold Networks (KANs), we introduce a novel network architecture, \textbf{LeAFNet}, which leverages learnable activation functions in the latent space to better approximate the target signal's implicit function. By reformulating point cloud compression as neural parameter compression, we further improve compression efficiency through quantization and entropy coding. Experimental results demonstrate that \textbf{LeAFNet} outperforms conventional MLPs in INR-based point cloud compression. Furthermore, \textbf{PICO} achieves superior geometry compression performance compared to the current MPEG point cloud compression standard, yielding an average improvement of $4.92$ dB in D1 PSNR. In joint geometry and attribute compression, our approach exhibits highly competitive results, with an average PCQM gain of $2.7 \times 10^{-3}$.
Road environment segmentation plays a significant role in autonomous driving. Numerous works based on Fully Convolutional Networks (FCNs) and Transformer architectures have been proposed to leverage local and global contextual learning for efficient and accurate semantic segmentation. In both architectures, the encoder often relies heavily on extracting continuous representations from the image, which limits the ability to represent meaningful discrete information. To address this limitation, we propose segmentation of the autonomous driving environment using vector quantization. Vector quantization offers three primary advantages for road environment segmentation. (1) Each continuous feature from the encoder is mapped to a discrete vector from the codebook, helping the model discover distinct features more easily than with complex continuous features. (2) Since a discrete feature acts as compressed versions of the encoder's continuous features, they also compress noise or outliers, enhancing the image segmentation task. (3) Vector quantization encourages the latent space to form coarse clusters of continuous features, forcing the model to group similar features, making the learned representations more structured for the decoding process. In this work, we combined vector quantization with the lightweight image segmentation model MobileUNETR and used it as a baseline model for comparison to demonstrate its efficiency. Through experiments, we achieved 77.0 % mIoU on Cityscapes, outperforming the baseline by 2.9 % without increasing the model's initial size or complexity.
Large language models (LLMs) have become pivotal in artificial intelligence, demonstrating strong capabilities in reasoning, understanding, and generating data. However, their deployment on edge devices is hindered by their substantial size, often reaching several billion parameters. Quantization is a widely used method to reduce memory usage and inference time, however LLMs present unique challenges due to the prevalence of outliers in their activations. In this work, we leverage the theoretical advantages of Hadamard matrices over random rotation matrices to push the boundaries of quantization in LLMs. We demonstrate that Hadamard matrices are more effective in reducing outliers, which are a significant obstacle in achieving low-bit quantization. Our method based on a gradual binary search enables 3-bit quantization for weights, activations, and key-value (KV) caches, resulting in a 40% increase in accuracy on common benchmarks compared to SoTA methods. We extend the use of rotation matrices to support non-power-of-2 embedding dimensions, similar to the Qwen architecture, by employing the Paley algorithm. We theoretically demonstrates the superiority of Hadamard matrices in reducing outliers.We achieved 3-bit quantization for weights, activations, and KV cache, significantly enhancing model performance. Our experimental results on multiple models family like Mistral, LLaMA, and Qwen demonstrate the effectiveness of our approach, outperforming existing methods and enabling practical 3-bit quantization.
Large language models (LLMs) exhibit remarkable performance across various natural language processing tasks but suffer from immense computational and memory demands, limiting their deployment in resource-constrained environments. To address this challenge, we propose NoWag: (Normalized Weight and Activation Guided Compression), a unified framework for zero-shot shape preserving compression algorithms. We compressed Llama-2 7B/13B/70B and Llama-3 8/70BB models, using two popular forms of shape-preserving compression, vector quantization NoWag-VQ (NoWag for Vector Quantization), and unstructured/semi-structured pruning NoWag-P (NoWag for Pruning). We found that NoWag-VQ significantly outperforms state-of-the-art zero shot VQ, and that NoWag-P performs competitively against state-of-the-art methods. These results suggest commonalities between these compression paradigms that could inspire future work. Our code is available at https://github.com/LawrenceRLiu/NoWag
In this paper, we propose StableQuant, a novel adaptive post-training quantization (PTQ) algorithm for widely used speech foundation models (SFMs). While PTQ has been successfully employed for compressing large language models (LLMs) due to its ability to bypass additional fine-tuning, directly applying these techniques to SFMs may not yield optimal results, as SFMs utilize distinct network architecture for feature extraction. StableQuant demonstrates optimal quantization performance regardless of the network architecture type, as it adaptively determines the quantization range for each layer by analyzing both the scale distributions and overall performance. We evaluate our algorithm on two SFMs, HuBERT and wav2vec2.0, for an automatic speech recognition (ASR) task, and achieve superior performance compared to traditional PTQ methods. StableQuant successfully reduces the sizes of SFM models to a quarter and doubles the inference speed while limiting the word error rate (WER) performance drop to less than 0.3% with 8-bit quantization.
Why do larger language models generalize better? To investigate this question, we develop generalization bounds on the pretraining objective of large language models (LLMs) in the compute-optimal regime, as described by the Chinchilla scaling laws. We introduce a novel, fully empirical Freedman-type martingale concentration inequality that tightens existing bounds by accounting for the variance of the loss function. This generalization bound can be decomposed into three interpretable components: the number of parameters per token, the loss variance, and the quantization error at a fixed bitrate. As compute-optimal language models are scaled up, the number of parameters per data point remains constant; however, both the loss variance and the quantization error decrease, implying that larger models should have smaller generalization gaps. We examine why larger models tend to be more quantizable from an information theoretic perspective, showing that the rate at which they can integrate new information grows more slowly than their capacity on the compute-optimal frontier. From these findings we produce a scaling law for the generalization gap, with bounds that become predictably stronger with scale.
Currently, large language models are gaining popularity, their achievements are used in many areas, ranging from text translation to generating answers to queries. However, the main problem with these new machine learning algorithms is that training such models requires large computing resources that only large IT companies have. To avoid this problem, a number of methods (LoRA, quantization) have been proposed so that existing models can be effectively fine-tuned for specific tasks. In this paper, we propose an E2E (end to end) audio summarization model using these techniques. In addition, this paper examines the effectiveness of these approaches to the problem under consideration and draws conclusions about the applicability of these methods.
Vision Transformer (ViT)-based models have shown state-of-the-art performance (e.g., accuracy) in vision-based AI tasks. However, realizing their capability in resource-constrained embedded AI systems is challenging due to their inherent large memory footprints and complex computations, thereby incurring high power/energy consumption. Recently, Spiking Vision Transformer (SViT)-based models have emerged as alternate low-power ViT networks. However, their large memory footprints still hinder their applicability for resource-constrained embedded AI systems. Therefore, there is a need for a methodology to compress SViT models without degrading the accuracy significantly. To address this, we propose QSViT, a novel design methodology to compress the SViT models through a systematic quantization strategy across different network layers. To do this, our QSViT employs several key steps: (1) investigating the impact of different precision levels in different network layers, (2) identifying the appropriate base quantization settings for guiding bit precision reduction, (3) performing a guided quantization strategy based on the base settings to select the appropriate quantization setting, and (4) developing an efficient quantized network based on the selected quantization setting. The experimental results demonstrate that, our QSViT methodology achieves 22.75% memory saving and 21.33% power saving, while also maintaining high accuracy within 2.1% from that of the original non-quantized SViT model on the ImageNet dataset. These results highlight the potential of QSViT methodology to pave the way toward the efficient SViT deployments on resource-constrained embedded AI systems.
In machine learning, quantization is widely used to simplify data representation and facilitate algorithm deployment on hardware. Given the fundamental role of classification in machine learning, it is crucial to investigate the impact of quantization on classification. Current research primarily focuses on quantization errors, operating under the premise that higher quantization errors generally result in lower classification performance. However, this premise lacks a solid theoretical foundation and often contradicts empirical findings. For instance, certain extremely low bit-width quantization methods, such as $\{0,1\}$-binary quantization and $\{0, \pm1\}$-ternary quantization, can achieve comparable or even superior classification accuracy compared to the original non-quantized data, despite exhibiting high quantization errors. To more accurately evaluate classification performance, we propose to directly investigate the feature discrimination of quantized data, instead of analyzing its quantization error. Interestingly, it is found that both binary and ternary quantization methods can improve, rather than degrade, the feature discrimination of the original data. This remarkable performance is validated through classification experiments across various data types, including images, speech, and texts.
Large Language Models (LLMs) have significantly advanced artificial intelligence by optimizing traditional Natural Language Processing (NLP) workflows, facilitating their integration into various systems. Many such NLP systems, including ours, directly incorporate LLMs. However, this approach either results in expensive costs or yields suboptimal performance after fine-tuning. In this paper, we introduce a three-stage cost-efficient end-to-end LLM deployment pipeline, comprising prototyping, knowledge transfer, and model compression, to effectively tackle the cost-performance dilemma in LLM-based frameworks. Its high cost-efficiency is manifested not only in simplifying system complexity and producing super-tiny online models with enhanced performance and reduced costs in the results, but also in addressing development cycle constraints, the lack of extensive high-quality data, and limited computational resources during the project development process. In the first stage, we construct an optimal performance prototype system by transforming complex tasks into a function call-based LLM-driven pipeline, which serves as a teacher model to generate high-quality data. In the second stage, we combine techniques like rejection sampling fine-tuning, reinforcement learning, and knowledge distillation to transfer knowledge to 0.5B student models, delivering effective performance at minimal cost. In the final stage, we further compress models to 0.4B via quantization and pruning, achieving ultra-low latency and cost. Extensive experimental results and the framework's modular design suggest cross-domain capabilities and potential applicability in other NLP areas.
With the proliferation of task-specific large language models, delta compression has emerged as a method to mitigate the resource challenges of deploying numerous such models by effectively compressing the delta model parameters. Previous delta-sparsification methods either remove parameters randomly or truncate singular vectors directly after singular value decomposition (SVD). However, these methods either disregard parameter importance entirely or evaluate it with too coarse a granularity. In this work, we introduce ImPart, a novel importance-aware delta sparsification approach. Leveraging SVD, it dynamically adjusts sparsity ratios of different singular vectors based on their importance, effectively retaining crucial task-specific knowledge even at high sparsity ratios. Experiments show that ImPart achieves state-of-the-art delta sparsification performance, demonstrating $2\times$ higher compression ratio than baselines at the same performance level. When integrated with existing methods, ImPart sets a new state-of-the-art on delta quantization and model merging.
Graph Transformers (GTs), which simultaneously integrate message-passing and self-attention mechanisms, have achieved promising empirical results in some graph prediction tasks. Although these approaches show the potential of Transformers in capturing long-range graph topology information, issues concerning the quadratic complexity and high computing energy consumption severely limit the scalability of GTs on large-scale graphs. Recently, as brain-inspired neural networks, Spiking Neural Networks (SNNs), facilitate the development of graph representation learning methods with lower computational and storage overhead through the unique event-driven spiking neurons. Inspired by these characteristics, we propose a linear-time Graph Transformer using Spiking Vector Quantization (GT-SVQ) for node classification. GT-SVQ reconstructs codebooks based on rate coding outputs from spiking neurons, and injects the codebooks into self-attention blocks to aggregate global information in linear complexity. Besides, spiking vector quantization effectively alleviates codebook collapse and the reliance on complex machinery (distance measure, auxiliary loss, etc.) present in previous vector quantization-based graph learning methods. In experiments, we compare GT-SVQ with other state-of-the-art baselines on node classification datasets ranging from small to large. Experimental results show that GT-SVQ has achieved competitive performances on most datasets while maintaining up to 130x faster inference speed compared to other GTs.
Federated Learning (FL) allows collaborative training among multiple devices without data sharing, thus enabling privacy-sensitive applications on mobile or Internet of Things (IoT) devices, such as mobile health and asset tracking. However, designing an FL system with good model utility that works with low computation/communication overhead on heterogeneous, resource-constrained mobile/IoT devices is challenging. To address this problem, this paper proposes FedX, a novel adaptive model decomposition and quantization FL system for IoT. To balance utility with resource constraints on IoT devices, FedX decomposes a global FL model into different sub-networks with adaptive numbers of quantized bits for different devices. The key idea is that a device with fewer resources receives a smaller sub-network for lower overhead but utilizes a larger number of quantized bits for higher model utility, and vice versa. The quantization operations in FedX are done at the server to reduce the computational load on devices. FedX iteratively minimizes the losses in the devices' local data and in the server's public data using quantized sub-networks under a regularization term, and thus it maximizes the benefits of combining FL with model quantization through knowledge sharing among the server and devices in a cost-effective training process. Extensive experiments show that FedX significantly improves quantization times by up to 8.43X, on-device computation time by 1.5X, and total end-to-end training time by 1.36X, compared with baseline FL systems. We guarantee the global model convergence theoretically and validate local model convergence empirically, highlighting FedX's optimization efficiency.
Graph self-supervised learning has gained significant attention recently. However, many existing approaches heavily depend on perturbations, and inappropriate perturbations may corrupt the graph's inherent information. The Vector Quantized Variational Autoencoder (VQ-VAE) is a powerful autoencoder extensively used in fields such as computer vision; however, its application to graph data remains underexplored. In this paper, we provide an empirical analysis of vector quantization in the context of graph autoencoders, demonstrating its significant enhancement of the model's capacity to capture graph topology. Furthermore, we identify two key challenges associated with vector quantization when applying in graph data: codebook underutilization and codebook space sparsity. For the first challenge, we propose an annealing-based encoding strategy that promotes broad code utilization in the early stages of training, gradually shifting focus toward the most effective codes as training progresses. For the second challenge, we introduce a hierarchical two-layer codebook that captures relationships between embeddings through clustering. The second layer codebook links similar codes, encouraging the model to learn closer embeddings for nodes with similar features and structural topology in the graph. Our proposed model outperforms 16 representative baseline methods in self-supervised link prediction and node classification tasks across multiple datasets.
Recent advancements in reasoning language models have demonstrated remarkable performance in complex tasks, but their extended chain-of-thought reasoning process increases inference overhead. While quantization has been widely adopted to reduce the inference cost of large language models, its impact on reasoning models remains understudied. In this study, we conduct the first systematic study on quantized reasoning models, evaluating the open-sourced DeepSeek-R1-Distilled Qwen and LLaMA families ranging from 1.5B to 70B parameters, and QwQ-32B. Our investigation covers weight, KV cache, and activation quantization using state-of-the-art algorithms at varying bit-widths, with extensive evaluation across mathematical (AIME, MATH-500), scientific (GPQA), and programming (LiveCodeBench) reasoning benchmarks. Our findings reveal that while lossless quantization can be achieved with W8A8 or W4A16 quantization, lower bit-widths introduce significant accuracy risks. We further identify model size, model origin, and task difficulty as critical determinants of performance. Contrary to expectations, quantized models do not exhibit increased output lengths. In addition, strategically scaling the model sizes or reasoning steps can effectively enhance the performance. All quantized models and codes will be open-sourced in https://github.com/ruikangliu/Quantized-Reasoning-Models.
Chinese-Vicuna is an open-source, resource-efficient language model designed to bridge the gap in Chinese instruction-following capabilities by fine-tuning Meta's LLaMA architecture using Low-Rank Adaptation (LoRA). Targeting low-resource environments, it enables cost-effective deployment on consumer GPUs (e.g., RTX-2080Ti for 7B models) and supports domain-specific adaptation in fields like healthcare and law. By integrating hybrid datasets (BELLE and Guanaco) and 4-bit quantization (QLoRA), the model achieves competitive performance in tasks such as translation, code generation, and domain-specific Q\&A. The project provides a comprehensive toolkit for model conversion, CPU inference, and multi-turn dialogue interfaces, emphasizing accessibility for researchers and developers. Evaluations indicate competitive performance across medical tasks, multi-turn dialogue coherence, and real-time legal updates. Chinese-Vicuna's modular design, open-source ecosystem, and community-driven enhancements position it as a versatile foundation for Chinese LLM applications.
This paper proposes a novel end-to-end digital semantic communication framework based on multi-codebook vector quantization (VQ), referred to as ESC-MVQ. Unlike prior approaches that rely on end-to-end training with a specific power or modulation scheme, often under a particular channel condition, ESC-MVQ models a channel transfer function as parallel binary symmetric channels (BSCs) with trainable bit-flip probabilities. Building on this model, ESC-MVQ jointly trains multiple VQ codebooks and their associated bit-flip probabilities with a single encoder-decoder pair. To maximize inference performance when deploying ESC-MVQ in digital communication systems, we devise an optimal communication strategy that jointly optimizes codebook assignment, adaptive modulation, and power allocation. To this end, we develop an iterative algorithm that selects the most suitable VQ codebook for semantic features and flexibly allocates power and modulation schemes across the transmitted symbols. Simulation results demonstrate that ESC-MVQ, using a single encoder-decoder pair, outperforms existing digital semantic communication methods in both performance and memory efficiency, offering a scalable and adaptive solution for realizing digital semantic communication in diverse channel conditions.
Neural Networks (NNs) trained through supervised learning struggle with managing edge-case scenarios common in real-world driving due to the intractability of exhaustive datasets covering all edge-cases, making knowledge-driven approaches, akin to how humans intuitively detect unexpected driving behavior, a suitable complement to data-driven methods. This work proposes a hybrid architecture combining low-level Model Predictive Controller (MPC) with locally deployed Large Language Models (LLMs) to enhance decision-making and Human Machine Interaction (HMI). The DecisionxLLM module evaluates robotic state information against natural language instructions to ensure adherence to desired driving behavior. The MPCxLLM module then adjusts MPC parameters based on LLM-generated insights, achieving control adaptability while preserving the safety and constraint guarantees of traditional MPC systems. Further, to enable efficient on-board deployment and to eliminate dependency on cloud connectivity, we shift processing to the on-board computing platform: We propose an approach that exploits Retrieval Augmented Generation (RAG), Low Rank Adaptation (LoRA) fine-tuning, and quantization. Experimental results demonstrate that these enhancements yield significant improvements in reasoning accuracy by up to 10.45%, control adaptability by as much as 52.2%, and up to 10.5x increase in computational efficiency (tokens/s), validating the proposed framework's practicality for real-time deployment even on down-scaled robotic platforms. This work bridges high-level decision-making with low-level control adaptability, offering a synergistic framework for knowledge-driven and adaptive Autonomous Driving Systems (ADS).
While large language models (LLMs) have revolutionized text-to-speech (TTS) synthesis through discrete tokenization paradigms, current architectures exhibit fundamental tensions between three critical dimensions: 1) irreversible loss of acoustic characteristics caused by quantization of speech prompts; 2) stringent dependence on precisely aligned prompt speech-text pairs that limit real-world deployment; and 3) catastrophic forgetting of the LLM's native text comprehension during optimization for speech token generation. To address these challenges, we propose an LLM-based text-to-speech Generation approach Optimized via a novel dual-branch ArchiTecture (GOAT-TTS). Our framework introduces two key innovations: (1) The modality-alignment branch combines a speech encoder and projector to capture continuous acoustic embeddings, enabling bidirectional correlation between paralinguistic features (language, timbre, emotion) and semantic text representations without transcript dependency; (2) The speech-generation branch employs modular fine-tuning on top-k layers of an LLM for speech token prediction while freezing the bottom-n layers to preserve foundational linguistic knowledge. Moreover, multi-token prediction is introduced to support real-time streaming TTS synthesis. Experimental results demonstrate that our GOAT-TTS achieves performance comparable to state-of-the-art TTS models while validating the efficacy of synthesized dialect speech data.
Large Artificial Intelligence Models (LAMs) powered by massive datasets, extensive parameter scales, and extensive computational resources, leading to significant transformations across various industries. Yet, their practical deployment on resource-limited mobile edge devices is hindered by critical challenges such as data privacy, constrained resources, and high overhead costs. Addressing this gap, this paper proposes a novel framework, named Quantized Split Federated Fine-Tuning Large AI Model (SFLAM). By partitioning the training load between edge devices and servers using a split learning paradigm, SFLAM can facilitate the operation of large models on devices and significantly lowers the memory requirements on edge devices. Additionally, SFLAM incorporates quantization management, power control, and bandwidth allocation strategies to enhance training efficiency while concurrently reducing energy consumption and communication latency. A theoretical analysis exploring the latency-energy trade-off is presented, and the framework's efficacy is validated via comprehensive simulations. The findings indicate that SFLAM achieves superior performance in terms of learning efficiency and scalability compared to conventional methods, thereby providing a valuable approach for enabling advanced AI services in resource-constrained scenarios.