arXiv subjects

Guosheng Hu

Publications and source records attributed to Guosheng Hu.

At least 19 recordsLinked to original sources

arXiv2026

MAPLE: MoE Adaptive Plug-and-play Layer-wise Expert allocation

Sparsely-activated Mixture-of-Experts (MoE) Transformers universally fix the same number of routed experts across all layers, a convention that ignores the well-documented heterogeneity in layer-wise redundancy. We demonstrate that this uniformity is systematically suboptimal and propose MAPLE, a plug-and-play framework that reallocates the routed-expert budget heterogeneously across layers of any pretrained MoE LLM, without modifying weights or requiring retraining. Our core contribution is a closed-form sensitivity-guided allocation: we probe each layer's response to variation in expert count, quantify sensitivity using three measures, and derive an analytically optimal budget assignment that directs capacity towards sensitive layers and absorbs reductions in redundant layers. This closed-form solution is further refined by a sensitivity-constrained genetic search that uses layer-wise sensitivity as a prior to guide exploration, yielding faster convergence and superior allocation quality. On four MoE models spanning different scales and architectures, MAPLE outperforms uniform and pruning-based baselines under a 75% routed-expert budget. Notably, on DeepSeek-MoE-16B, MAPLE uses only 75% of the experts yet surpasses the original 100% expert-uniform baseline on ARC-E, ARC-C, and BoolQ, improving accuracy from 65.09 to 71.40, 48.49 to 51.50, and 80.03 to 82.38, respectively. These accuracy gains translate into measured deployment efficiency: implementing MAPLE in SGLang reduces single-GPU end-to-end serving latency by 32.2% and improves throughput by 47.4%. These results show that well-designed heterogeneous allocation can be more effective than simply activating more experts, establishing it as a principled and practical axis for improving MoE efficiency.

cs.LG↗

arXiv2026

Adapting Vision Foundation Models with Cascaded Semantics

Prompt tuning, a leading parameter-efficient adaptation paradigm in NLP, has recently been extended to computer vision. Visual prompt tuning (VPT) adapts pre-trained vision transformers (ViTs) by updating a small set of additional prompt parameters. However, existing visual prompts are randomly initialized and do not exploit prior knowledge, such as instructions in NLP. We address this gap by injecting two complementary semantic priors into VPT. Fundamental image priors, including color, texture, and shape, are extracted with classical hand-crafted operators and injected into the input space, while self-attention maps provide instance-aware semantics in the feature space. We further propose a cascaded scheme that integrates both priors throughout ViT adaptation. Experiments on 34 challenging image classification datasets demonstrate superior downstream adaptation while tuning only 0.74% of ViT parameters. Project page: https://xixiaouab.github.io/Cascaded-Semantics/.

cs.CV↗

arXiv2026

SPEAR: A System for Post-Quantization Error-Adaptive Recovery Enabling Efficient Low-Bit LLM Serving

Efficient large language model (LLM) serving is increasingly constrained by deployment cost. Quantization is a key technique for reducing serving cost, yet even state-of-the-art 4-bit quantizers exhibit a noticeable quality gap from FP16, particularly for smaller models where low-bit serving is most beneficial. We identify a fundamental cause of this gap: quantization error is highly input-dependent and varies substantially across tokens, while existing post-quantization compensation methods are static and apply identical corrections to all inputs. As a result, easy tokens are over-corrected while hard tokens remain under-corrected. We present SPEAR, a system for post-quantization error-adaptive recovery that improves low-bit LLM serving. SPEAR introduces lightweight Error Compensators (ECs) modulated by per-token gates and places them only at the most error-sensitive layers identified through a CKA-guided entropy-aware diagnostic. This focuses a small parameter budget where it is most effective. Efficient deployment of ECs presents several systems challenges, including additional computation, tensor-parallel synchronization caused by input-dependent gating, and latency instability across configurations. SPEAR addresses these issues through adaptive kernel-fusion dispatch, combining an epilogue-integrated peer-reduction kernel with P2P dual-write to fuse the post-EC computation into low-bit GEMMs, and an SLO-constrained EC-aware scheduler for predictable serving performance. Across challenging per-channel quantization settings, SPEAR recovers 56-75% of the perplexity gap between W4 and FP16 while adding less than 1% model memory overhead and maintaining latency comparable to a widely used 4-bit serving deployment.

cs.AR↗

arXiv2026

FFR: Forward-Forward Learning for Regression

The Forward-Forward (FF) algorithm offers a computationally efficient and biologically plausible alternative to backpropagation (BP) by training neural networks through purely local, layer-wise optimization. However, FF is inherently designed for classification via contrastive positive-negative sample pairs, and extending it to regression poses fundamental challenges: continuous target space lack natural "opposites" for contrastive learning, and the standard goodness function carries no information about target magnitude or ordering. We propose FFR (Forward-Forward for Regression), to our knowledge, the first framework to extend FF to real-world regression and demonstrate competitive performance across diverse real-world datasets. FFR introduces three key innovations: (1) an ordinal competitive goodness function that replaces contrastive pairs with competitive learning between partitioned neuron groups under distance-aware ordinal supervision; (2) a stratified ladder architecture where shallow layers learn coarse ordinal discrimination and deeper layers refine into fine-grained regression, with multi-scale feature aggregation for inter-layer collaboration; and (3) hierarchical prediction with uncertainty estimation, where multi-scale predictors jointly provide robust predictions and prediction confidence as a free-lunch. Extensive experimental results show FFR recovers on average 98.6% of BP's accuracy across five real-world regression benchmarks while reducing peak training memory to only 27% of BP's at depth 8 and 8% at depth 32, with per-iteration time around 72% of BP's, and substantially outperforms all BP-free competitors.

cs.LG↗

arXiv2026

ColorConceptBench: A Benchmark for Probabilistic Color-Concept Understanding in Text-to-Image Models

Text-to-image (T2I) models have advanced considerably in generating high-quality images from textual descriptions. However, their ability to associate colors with concepts remains largely constrained to explicit color names or codes, while their capacity to handle \emph{implicit concepts}, such as emotions and visual states, remains underexplored. To address this gap, we introduce ColorConceptBench, an expert-annotated benchmark that systematically evaluates color-concept associations through probabilistic color distributions. ColorConceptBench moves beyond explicit color specifications by examining how models interpret 1,281 implicit color concepts, grounded in 6,584 human annotations. Our evaluation of nine leading T2I models reveals that performance varies substantially across semantic categories, and models exhibit a significant lack of sensitivity to abstract semantics. These limitations persist even when applying classifier-free guidance scaling at inference time, suggesting that achieving human-like color understanding demands a shift in how models learn and represent implicit semantic meaning.

cs.CV↗

arXiv2026

Not All Directions Matter: Towards Structured and Task-Aware Low-Rank Model Adaptation

Low-Rank Adaptation (LoRA) has become a cornerstone of parameter-efficient fine-tuning (PEFT). Yet, its efficacy is hampered by two fundamental limitations: semantic drift, by treating all update directions with equal importance, and structural incoherence, from adapting layers independently, resulting in suboptimal, uncoordinated updates. To remedy these, we propose StructLoRA, a framework that addresses both limitations through a principled, dual-component design: (1) an Information Bottleneck-guided filter that prunes task-irrelevant directions to mitigate semantic drift, and (2) a lightweight, training-only graph-based coordinator that enforces inter-layer consistency to resolve structural incoherence. Extensive experiments across large language model , vision language model, and vision model (including LLaMA, LLaVA, and ViT) demonstrate that StructLoRA consistently establishes a new state-of-the-art, outperforming not only vanilla LoRA but also advanced dynamic rank allocation and sparsity-based methods. Notably, the benefits are particularly pronounced in challenging low-rank and low-data regimes. Crucially, since our proposed modules operate only during training, StructLoRA enhances performance with zero additional inference cost, advancing the focus of PEFT -- from mere parameter compression to a more holistic optimization of information quality and structural integrity.

cs.CV↗

arXiv2026

Trade in Minutes! Rationality-Driven Agentic System for Quantitative Financial Trading

Recent advancements in large language models (LLMs) and agentic systems have shown exceptional decision-making capabilities, revealing significant potential for autonomic finance. Current financial trading agents predominantly simulate anthropomorphic roles that inadvertently introduce emotional biases and rely on peripheral information, while being constrained by the necessity for continuous inference during deployment. In this paper, we pioneer the harmonization of strategic depth in agents with the mechanical rationality essential for quantitative trading. Consequently, we present TiMi (Trade in Minutes), a rationality-driven multi-agent system that architecturally decouples strategy development from minute-level deployment. TiMi leverages specialized LLM capabilities of semantic analysis, code programming, and mathematical reasoning within a comprehensive policy-optimization-deployment chain. Specifically, we propose a two-tier analytical paradigm from macro patterns to micro customization, layered programming design for trading bot implementation, and closed-loop optimization driven by mathematical reflection. Extensive evaluations across 200+ trading pairs in stock and cryptocurrency markets empirically validate the efficacy of TiMi in stable profitability, action efficiency, and risk control under volatile market dynamics.

cs.MA↗

arXiv2026

A Comprehensive Evaluation on Quantization Techniques for Large Language Models

For large language models (LLMs), post-training quantization (PTQ) can significantly reduce memory footprint and computational overhead. Model quantization is rapidly evolving. Though many papers report breakthrough results, they are often evaluated under different settings because a method typically contains multiple components. Analyzing connections among existing methods is important for deeper understanding. To bridge these gaps, we conduct an extensive review of state-of-the-art methods and perform comprehensive evaluations under the same conditions for fair comparison. To our knowledge, such a fair and extensive investigation remains critically underexplored. To better understand connections, first, we decouple published quantization methods into two steps: pre-quantization transformation and quantization error mitigation. The former is a preprocessing step that reduces outlier impact by flattening the data distribution; the latter offsets quantization errors to improve performance. Second, we evaluate and analyze the impact of different settings, including granularity and symmetry. Third, we analyze and evaluate the latest MXFP4 and NVFP4 data formats and their performance. Our experiments first demonstrate that optimized rotation and scaling yield the best pre-quantization performance, and that combining low-rank compensation with GPTQ can occasionally outperform GPTQ alone for error mitigation. Second, finer granularity improves performance but increases storage overhead. Third, we find that scaling-factor format and precision greatly affect FP4 performance, and that rotation-based strategies effective for INT4 offer limited gains for MXFP4 and NVFP4, motivating further study.

cs.LG↗

arXiv2026

Cap2Sum: Learning to Summarize Videos by Generating Captions

With the rapid growth of video data on the internet, video summarization is becoming a very important AI technology. However, due to the high labelling cost of video summarization, existing studies have to be conducted on small-scale datasets, leading to limited performance and generalization capacity. In this work, we introduce the use of dense video captions as a supervision signal to train video summarization models. Motivated by this, we propose Cap2Sum, a model that learns to summarize videos by generating captions, to exploit dense video caption annotations. This weakly-supervised approach allows us to train the models on large-scale dense video caption datasets to achieve better performance and generalization capacity. To further improve the generalization capacity, we introduce a CLIP (a strong vision-language model) Prior mechanism to enhance the learning of important objects that captions may ignore in the videos. In practice, Cap2Sum can perform zero-shot video summarization or be fine-tuned by the ground-truth summary or video caption of the target dataset. To examine the performance of Cap2Sum after weakly-supervised fine-tuning by the video captions, we propose two new datasets, TVSum-Caption and SumMe-Caption, which are derived from two common video summarization datasets and will be publicly released. We conduct extensive experiments and the results demonstrate that our method achieves significant improvements in performance and generalization capacity compared with previous methods.

cs.MM↗

arXiv2025

Metrics and evaluations for computational and sustainable AI efficiency

The rapid advancement of Artificial Intelligence (AI) has created unprecedented demands for computational power, yet methods for evaluating the performance, efficiency, and environmental impact of deployed models remain fragmented. Current approaches often fail to provide a holistic view, making it difficult to compare and optimise systems across heterogeneous hardware, software stacks, and numeric precisions. To address this gap, we propose a unified and reproducible methodology for AI model inference that integrates computational and environmental metrics under realistic serving conditions. Our framework provides a pragmatic, carbon-aware evaluation by systematically measuring latency and throughput distributions, energy consumption, and location-adjusted carbon emissions, all while maintaining matched accuracy constraints for valid comparisons. We apply this methodology to multi-precision models across diverse hardware platforms, from data-centre accelerators like the GH200 to consumer-level GPUs such as the RTX 4090, running on mainstream software stacks including PyTorch, TensorRT, and ONNX Runtime. By systematically categorising these factors, our work establishes a rigorous benchmarking framework that produces decision-ready Pareto frontiers, clarifying the trade-offs between accuracy, latency, energy, and carbon. The accompanying open-source code enables independent verification and facilitates adoption, empowering researchers and practitioners to make evidence-based decisions for sustainable AI deployment.

cs.PF↗

arXiv2025

Cross-Modal Distillation For Widely Differing Modalities

Deep learning achieved great progress recently, however, it is not easy or efficient to further improve its performance by increasing the size of the model. Multi-modal learning can mitigate this challenge by introducing richer and more discriminative information as input. To solve the problem of limited access to multi-modal data at the time of use, we conduct multi-modal learning by introducing a teacher model to transfer discriminative knowledge to a student model during training. However, this knowledge transfer via distillation is not trivial because the big domain gap between the widely differing modalities can easily lead to overfitting. In this work, we introduce a cross-modal distillation framework. Specifically, we find hard constrained loss, e.g. l2 loss forcing the student being exact the same as the teacher, can easily lead to overfitting in cross-modality distillation. To address this, we propose two soft constrained knowledge distillation strategies at the feature level and classifier level respectively. In addition, we propose a quality-based adaptive weights module to weigh input samples via quantified data quality, leading to robust model training. We conducted experiments on speaker recognition and image classification tasks, and the results show that our approach is able to effectively achieve knowledge transfer between the commonly used and widely differing modalities of image, text, and speech.

cs.AI↗

arXiv2025

The Chaotic Art: Quantum Representation and Manipulation of Color

Due to its unique computing principles, quantum computing technology will profoundly change the spectacle of color art. Focusing on experimental exploration of color qubit representation, color channel processing, and color image generation via quantum computing, this article proposes a new technical path for color computing in quantum computing environment, by which digital color is represented, operated, and measured in quantum bits, and then restored for classical computers as computing results. This method has been proved practicable as an artistic technique of color qubit representation and quantum computing via programming experiments in Qiskit and IBM Q. By building a bridge between classical chromatics and quantum graphics, quantum computers can be used for information visualization, image processing, and more color computing tasks. Furthermore, quantum computing can be expected to facilitate new color theories and artistic concepts.

quant-ph↗

arXiv2025

Unsupervised Exposure Correction

Current exposure correction methods have three challenges, labor-intensive paired data annotation, limited generalizability, and performance degradation in low-level computer vision tasks. In this work, we introduce an innovative Unsupervised Exposure Correction (UEC) method that eliminates the need for manual annotations, offers improved generalizability, and enhances performance in low-level downstream tasks. Our model is trained using freely available paired data from an emulated Image Signal Processing (ISP) pipeline. This approach does not need expensive manual annotations, thereby minimizing individual style biases from the annotation and consequently improving its generalizability. Furthermore, we present a large-scale Radiometry Correction Dataset, specifically designed to emphasize exposure variations, to facilitate unsupervised learning. In addition, we develop a transformation function that preserves image details and outperforms state-of-the-art supervised methods [12], while utilizing only 0.01% of their parameters. Our work further investigates the broader impact of exposure correction on downstream tasks, including edge detection, demonstrating its effectiveness in mitigating the adverse effects of poor exposure on low-level features. The source code and dataset are publicly available at https://github.com/BeyondHeaven/uec_code.

cs.CV↗

arXiv2025

Towards Extreme Pruning of LLMs with Plug-and-Play Mixed Sparsity

N:M structured pruning is essential for large language models (LLMs) because it can remove less important network weights and reduce the memory and computation requirements. Existing pruning methods mainly focus on designing metrics to measure the importance of network components to guide pruning. Apart from the impact of these metrics, we observe that different layers have different sensitivities over the network performance. Thus, we propose an efficient method based on the trace of Fisher Information Matrix (FIM) to quantitatively measure and verify the different sensitivities across layers. Based on this, we propose Mixed Sparsity Pruning (MSP) which uses a pruning-oriented evolutionary algorithm (EA) to determine the optimal sparsity levels for different layers. To guarantee fast convergence and achieve promising performance, we utilize efficient FIM-inspired layer-wise sensitivity to initialize the population of EA. In addition, our MSP can work as a plug-and-play module, ready to be integrated into existing pruning methods. Extensive experiments on LLaMA and LLaMA-2 on language modeling and zero-shot tasks demonstrate our superior performance. In particular, in extreme pruning ratio (e.g. 75%), our method significantly outperforms existing methods in terms of perplexity (PPL) by orders of magnitude (Figure 1).

cs.CL↗

arXiv2025

Revisiting Large Language Model Pruning using Neuron Semantic Attribution

Model pruning technique is vital for accelerating large language models by reducing their size and computational requirements. However, the generalizability of existing pruning methods across diverse datasets and tasks remains unclear. Thus, we conduct extensive evaluations on 24 datasets and 4 tasks using popular pruning methods. Based on these evaluations, we find and then investigate that calibration set greatly affect the performance of pruning methods. In addition, we surprisingly find a significant performance drop of existing pruning methods in sentiment classification tasks. To understand the link between performance drop and pruned neurons, we propose Neuron Semantic Attribution, which learns to associate each neuron with specific semantics. This method first makes the unpruned neurons of LLMs explainable.

cs.CL↗

arXiv2024

Adapting Amidst Degradation: Cross Domain Li-ion Battery Health Estimation via Physics-Guided Test-Time Training

Health modeling of lithium-ion batteries (LIBs) is crucial for safe and efficient energy management and carries significant socio-economic implications. Although Machine Learning (ML)-based State of Health (SOH) estimation methods have made significant progress in accuracy, the scarcity of high-quality LIB data remains a major obstacle. Existing transfer learning methods for cross-domain LIB SOH estimation have significantly alleviated the labeling burden of target LIB data, however, they still require sufficient unlabeled target data (UTD) for effective adaptation to the target domain. Collecting this UTD is challenging due to the time-consuming nature of degradation experiments. To address this issue, we introduce a practical Test-Time Training framework, BatteryTTT, which adapts the model continually using each UTD collected amidst degradation, thereby significantly reducing data collection time. To fully utilize each UTD, BatteryTTT integrates the inherent physical laws of modern LIBs into self-supervised learning, termed Physcics-Guided Test-Time Training. Additionally, we explore the potential of large language models (LLMs) in battery sequence modeling by evaluating their performance in SOH estimation through model reprogramming and prefix prompt adaptation. The combination of BatteryTTT and LLM modeling, termed GPT4Battery, achieves state-of-the-art generalization results across current LIB benchmarks. Furthermore, we demonstrate the practical value and scalability of our approach by deploying it in our real-world battery management system (BMS) for 300Ah large-scale energy storage LIBs.

cs.LG↗

arXiv2024

Object Pose Estimation via the Aggregation of Diffusion Features

Estimating the pose of objects from images is a crucial task of 3D scene understanding, and recent approaches have shown promising results on very large benchmarks. However, these methods experience a significant performance drop when dealing with unseen objects. We believe that it results from the limited generalizability of image features. To address this problem, we have an in-depth analysis on the features of diffusion models, e.g. Stable Diffusion, which hold substantial potential for modeling unseen objects. Based on this analysis, we then innovatively introduce these diffusion features for object pose estimation. To achieve this, we propose three distinct architectures that can effectively capture and aggregate diffusion features of different granularity, greatly improving the generalizability of object pose estimation. Our approach outperforms the state-of-the-art methods by a considerable margin on three popular benchmark datasets, LM, O-LM, and T-LESS. In particular, our method achieves higher accuracy than the previous best arts on unseen objects: 97.9% vs. 93.5% on Unseen LM, 85.9% vs. 76.3% on Unseen O-LM, showing the strong generalizability of our method. Our code is released at https://github.com/Tianfu18/diff-feats-pose.

cs.CV↗

arXiv2024

Conv-Adapter: Exploring Parameter Efficient Transfer Learning for ConvNets

While parameter efficient tuning (PET) methods have shown great potential with transformer architecture on Natural Language Processing (NLP) tasks, their effectiveness with large-scale ConvNets is still under-studied on Computer Vision (CV) tasks. This paper proposes Conv-Adapter, a PET module designed for ConvNets. Conv-Adapter is light-weight, domain-transferable, and architecture-agnostic with generalized performance on different tasks. When transferring on downstream tasks, Conv-Adapter learns tasks-specific feature modulation to the intermediate representations of backbones while keeping the pre-trained parameters frozen. By introducing only a tiny amount of learnable parameters, e.g., only 3.5% full fine-tuning parameters of ResNet50. It can also be applied for transformer-based backbones. Conv-Adapter outperforms previous PET baseline methods and achieves comparable or surpasses the performance of full fine-tuning on 23 classification tasks of various domains. It also presents superior performance on the few-shot classification with an average margin of 3.39%. Beyond classification, Conv-Adapter can generalize to detection and segmentation tasks with more than 50% reduction of parameters but comparable performance to the traditional full fine-tuning.

cs.CV↗