arXiv subjects

Lu Yin

Publications and source records attributed to Lu Yin.

At least 55 records · Page 3Linked to original sources

arXiv2026

Demystifying the Roles of LLM Layers in Retrieval, Knowledge, and Reasoning

Recent studies suggest that the deeper layers of Large Language Models (LLMs) contribute little to representation learning and can often be removed without significant performance loss. However, such claims are typically drawn from narrow evaluations and may overlook important aspects of model behavior. In this work, we present a systematic study of depth utilization across diverse dimensions, including evaluation protocols, task categories, and model architectures. Our analysis confirms that very deep layers are generally less effective than earlier ones, but their contributions vary substantially with the evaluation setting. Under likelihood-based metrics without generation, pruning most layers preserves performance, with only the initial few being critical. By contrast, generation-based evaluation uncovers indispensable roles for middle and deeper layers in enabling reasoning and maintaining long-range coherence. We further find that knowledge and retrieval are concentrated in shallow components, whereas reasoning accuracy relies heavily on deeper layers -- yet can be reshaped through distillation. These results highlight that depth usage in LLMs is highly heterogeneous and context-dependent, underscoring the need for task-, metric-, and model-aware perspectives in both interpreting and compressing large models.

cs.AI↗

arXiv2026

SemPA: Improving Sentence Embeddings of Large Language Models through Semantic Preference Alignment

Traditional sentence embedding methods employ token-level contrastive learning on non-generative pre-trained models. Recently, there have emerged embedding methods based on generative large language models (LLMs). These methods either rely on fixed prompt templates or involve modifications to the model architecture. The former lacks further optimization of the model and results in limited performance, while the latter alters the internal computational mechanisms of the model, thereby compromising its generative capabilities. We propose SemPA, a novel approach that boosts the sentence representations while preserving the generative ability of LLMs via semantic preference alignment. We leverage sentence-level Direct Preference Optimization (DPO) to efficiently optimize LLMs on a paraphrase generation task, where the model learns to discriminate semantically equivalent sentences while preserving inherent generative capacity. Theoretically, we establish a formal connection between DPO and contrastive learning under the Plackett-Luce model framework. Empirically, experimental results on both semantic textual similarity tasks and various benchmarks for LLMs show that SemPA achieves better semantic representations without sacrificing the inherent generation capability of LLMs.

cs.CL↗

arXiv2025

Pushing the Limits of Sparsity: A Bag of Tricks for Extreme Pruning

Pruning of deep neural networks has been an effective technique for reducing model size while preserving most of the performance of dense networks, crucial for deploying models on memory and power-constrained devices. While recent sparse learning methods have shown promising performance up to moderate sparsity levels such as 95% and 98%, accuracy quickly deteriorates when pushing sparsities to extreme levels due to unique challenges such as fragile gradient flow. In this work, we explore network performance beyond the commonly studied sparsities, and develop techniques that encourage stable training without accuracy collapse even at extreme sparsities, including 99.90%, 99.95\% and 99.99% on ResNet architectures. We propose three complementary techniques that enhance sparse training through different mechanisms: 1) Dynamic ReLU phasing, where DyReLU initially allows for richer parameter exploration before being gradually replaced by standard ReLU, 2) weight sharing which reuses parameters within a residual layer while maintaining the same number of learnable parameters, and 3) cyclic sparsity, where both sparsity levels and sparsity patterns evolve dynamically throughout training to better encourage parameter exploration. We evaluate our method, which we term Extreme Adaptive Sparse Training (EAST) at extreme sparsities using ResNet-34 and ResNet-50 on CIFAR-10, CIFAR-100, and ImageNet, achieving competitive or improved performance compared to existing methods, with notable gains at extreme sparsity levels.

cs.CV↗

arXiv2025

Improving Sparse IMU-based Motion Capture with Motion Label Smoothing

Sparse Inertial Measurement Units (IMUs) based human motion capture has gained significant momentum, driven by the adaptation of fundamental AI tools such as recurrent neural networks (RNNs) and transformers that are tailored for temporal and spatial modeling. Despite these achievements, current research predominantly focuses on pipeline and architectural designs, with comparatively little attention given to regularization methods, highlighting a critical gap in developing a comprehensive AI toolkit for this task. To bridge this gap, we propose motion label smoothing, a novel method that adapts the classic label smoothing strategy from classification to the sparse IMU-based motion capture task. Specifically, we first demonstrate that a naive adaptation of label smoothing, including simply blending a uniform vector or a ``uniform'' motion representation (e.g., dataset-average motion or a canonical T-pose), is suboptimal; and argue that a proper adaptation requires increasing the entropy of the smoothed labels. Second, we conduct a thorough analysis of human motion labels, identifying three critical properties: 1) Temporal Smoothness, 2) Joint Correlation, and 3) Low-Frequency Dominance, and show that conventional approaches to entropy enhancement (e.g., blending Gaussian noise) are ineffective as they disrupt these properties. Finally, we propose the blend of a novel skeleton-based Perlin noise for motion label smoothing, designed to raise label entropy while satisfying motion properties. Extensive experiments applying our motion label smoothing to three state-of-the-art methods across four real-world IMU datasets demonstrate its effectiveness and robust generalization (plug-and-play) capability.

cs.GR↗

arXiv2025

CLIMB-3D: Continual Learning for Imbalanced 3D Instance Segmentation

While 3D instance segmentation (3DIS) has advanced significantly, most existing methods assume that all object classes are known in advance and uniformly distributed. However, this assumption is unrealistic in dynamic, real-world environments where new classes emerge gradually and exhibit natural imbalance. Although some approaches address the emergence of new classes, they often overlook class imbalance, which leads to suboptimal performance, particularly on rare categories. To tackle this, we propose \ourmethodbf, a unified framework for \textbf{CL}ass-incremental \textbf{Imb}alance-aware \textbf{3D}IS. Building upon established exemplar replay (ER) strategies, we show that ER alone is insufficient to achieve robust performance under memory constraints. To mitigate this, we introduce a novel pseudo-label generator (PLG) that extends supervision to previously learned categories by leveraging predictions from a frozen model trained on prior tasks. Despite its promise, PLG tends to be biased towards frequent classes. Therefore, we propose a class-balanced re-weighting (CBR) scheme that estimates object frequencies from pseudo-labels and dynamically adjusts training bias, without requiring access to past data. We design and evaluate three incremental scenarios for 3DIS on the challenging ScanNet200 dataset and additionally validate our method for semantic segmentation on ScanNetV2. Our approach achieves state-of-the-art results, surpassing prior work by up to 16.76\% mAP for instance segmentation and approximately 30\% mIoU for semantic segmentation, demonstrating strong generalisation across both frequent and rare classes. Code is available at: https://github.com/vgthengane/CLIMB3D

cs.CV↗

arXiv2025

Constraints on Cosmic Birefringence from SPIDER, Planck, and ACT observations

The Early Dark Energy (EDE) model has been proposed as a candidate mechanism to generate cosmic birefringence through a Chern-Simons coupling between a dynamical scalar field and the cosmic microwave background (CMB) photon. Such birefringence induces a nonzero cross-correlation between the CMB $E$- and $B$-modes, providing a direct observational signature of parity violation. Recent measurements of the $EB$ and $TB$ power spectra, however, cannot yet unambiguously separate instrumental miscalibration ($α$) from a true cosmic-rotation angle ($β$). For this reason, we perform a model-independent analysis in terms of the total effective rotation angle $α+β$. We analyze the latest $EB$ and $TB$ measurements from the SPIDER, Planck, and ACT experiments and derive constraints on the Chern-Simons coupling constant $gM_{Pl}$ and on the polarization rotation angle $α+β$. We find that the coupling $gM_{Pl}$ is not compatible with the SPIDER data, while it provides reasonable fits to the Planck and ACT measurements. The fits for $α+β$ prefer a value larger than zero: when combined, Planck+ACT yield a detection significance of approximately 7$σ$. We also find that ACT data alone do not provide sufficiently tight constraints on either $gM_{Pl}$ or $α+β$, whereas the combination Planck+ACT improves the statistical consistency of ACT's high-$\ell$ results and leads to a better PTE for those measurements.

astro-ph.CO↗

arXiv2025

AlphaDecay: Module-wise Weight Decay for Heavy-Tailed Balancing in LLMs

Weight decay is a standard regularization technique for training large language models (LLMs). While it is common to assign a uniform decay rate to every layer, this approach overlooks the structural diversity of LLMs and the varying spectral properties across modules. In this paper, we introduce AlphaDecay, a simple yet effective method that adaptively assigns different weight decay strengths to each module of an LLM. Our approach is guided by Heavy-Tailed Self-Regularization (HT-SR) theory, which analyzes the empirical spectral density (ESD) of weight correlation matrices to quantify "heavy-tailedness." Modules exhibiting more pronounced heavy-tailed ESDs, reflecting stronger feature learning, are assigned weaker decay, while modules with lighter-tailed spectra receive stronger decay. Our method leverages tailored weight decay assignments to balance the module-wise differences in spectral properties, leading to improved performance. Extensive pre-training tasks with various model sizes from 60M to 1B demonstrate that AlphaDecay achieves better perplexity and generalization than conventional uniform decay and other adaptive decay baselines. Our code is available at https://github.com/hed-ucas/AlphaDecay.

cs.CL↗

arXiv2025

Shape-aware Inertial Poser: Motion Tracking for Humans with Diverse Shapes Using Sparse Inertial Sensors

Human motion capture with sparse inertial sensors has gained significant attention recently. However, existing methods almost exclusively rely on a template adult body shape to model the training data, which poses challenges when generalizing to individuals with largely different body shapes (such as a child). This is primarily due to the variation in IMU-measured acceleration caused by changes in body shape. To fill this gap, we propose Shape-aware Inertial Poser (SAIP), the first solution considering body shape differences in sparse inertial-based motion capture. Specifically, we decompose the sensor measurements related to shape and pose in order to effectively model their joint correlations. Firstly, we train a regression model to transfer the IMU-measured accelerations of a real body to match the template adult body model, compensating for the shape-related sensor measurements. Then, we can easily follow the state-of-the-art methods to estimate the full body motions of the template-shaped body. Finally, we utilize a second regression model to map the joint velocities back to the real body, combined with a shape-aware physical optimization strategy to calculate global motions on the subject. Furthermore, our method relies on body shape awareness, introducing the first inertial shape estimation scheme. This is accomplished by modeling the shape-conditioned IMU-pose correlation using an MLP-based network. To validate the effectiveness of SAIP, we also present the first IMU motion capture dataset containing individuals of different body sizes. This dataset features 10 children and 10 adults, with heights ranging from 110 cm to 190 cm, and a total of 400 minutes of paired IMU-Motion samples. Extensive experimental results demonstrate that SAIP can effectively handle motion capture tasks for diverse body shapes. The code and dataset are available at https://github.com/yinlu5942/SAIP.

cs.GR↗

arXiv2025

Rewiring Experts on the Fly:Continuous Rerouting for Better Online Adaptation in Mixture-of-Expert models

Mixture-of-Experts (MoE) models achieve efficient scaling through sparse expert activation, but often suffer from suboptimal routing decisions due to distribution shifts in deployment. While existing test-time adaptation methods could potentially address these issues, they primarily focus on dense models and require access to external data, limiting their practical applicability to MoE architectures. However, we find that, instead of relying on reference data, we can optimize MoE expert selection on-the-fly based only on input context. As such, we propose \textit{a data-free, online test-time framework} that continuously adapts MoE routing decisions during text generation without external supervision or data. Our method cycles between two phases: During the prefill stage, and later in regular intervals, we optimize the routing decisions of the model using self-supervision based on the already generated sequence. Then, we generate text as normal, maintaining the modified router until the next adaption. We implement this through lightweight additive vectors that only update router logits in selected layers, maintaining computational efficiency while preventing over-adaptation. The experimental results show consistent performance gains on challenging reasoning tasks while maintaining robustness to context shifts. For example, our method achieves a 5.5\% improvement on HumanEval with OLMoE. Furthermore, owing to its plug-and-play property, our method naturally complements existing test-time scaling techniques, e.g., achieving 6\% average gains when incorporated with self-consistency on DeepSeek-V2-Lite.

cs.CL↗

arXiv2025

Double-Checker: Enhancing Reasoning of Slow-Thinking LLMs via Self-Critical Fine-Tuning

While slow-thinking large language models (LLMs) exhibit reflection-like reasoning, commonly referred to as the "aha moment:, their ability to generate informative critiques and refine prior solutions remains limited. In this paper, we introduce Double-Checker, a principled framework designed to enhance the reasoning capabilities of slow-thinking LLMs by fostering explicit self-critique and iterative refinement of their previous solutions. By fine-tuning on our curated 1,730 self-critical instances, Double-Checker empowers long-CoT LLMs to iteratively critique and refine their outputs during inference until they evaluate their solutions as correct under self-generated critiques. We validate the efficacy of Double-Checker across a comprehensive suite of reasoning benchmarks, demonstrating that iterative self-critique significantly enhances the reasoning capabilities of long-CoT LLMs. Notably, our Double-Checker increases the pass@1 performance on challenging AIME benchmarks from 4.4% to 18.2% compared to the original long-CoT LLMs. These results highlight a promising direction for developing more trustworthy and effective LLMs capable of structured self-critique. Our codes and data are available at https://github.com/XinXU-USTC/DoubleChecker

cs.CL↗

arXiv2025

Model Unmerging: Making Your Models Unmergeable for Secure Model Sharing

Model merging leverages multiple finetuned expert models to construct a multi-task model with low cost, and is gaining increasing attention. However, as a growing number of finetuned models become publicly available, concerns about the safety of model merging have emerged. Unauthorized merging may infringe on developers' rights and risk leaking sensitive personal information. Most existing methods focus on detecting whether a merged model originates from a specific source model, but fail to effectively prevent illegal merging. In this paper, we propose MergeLock, an active protection mechanism that disrupts model parameters to render them unmergeable, thereby directly preventing unauthorized model merging. Specifically, leveraging the inherent symmetry of the attention mechanism in Transformer-based models, we randomly sample two pairs of invertible matrices and apply them to the Query-Key (QK) and Value-Output (VO) branches. This transformation keeps the model's output unchanged while pushing it away from the shared parameter space of other finetuned models. Extensive experiments across both vision and language tasks demonstrate that MergeLock can degrade the performance of merged models by over 95% when a protected model is involved in most cases, demonstrating its effectiveness. Moreover, we further demonstrate that merged models protected by MergeLock cannot be effectively recovered using low-cost restoration methods, further enhancing robustness against unauthorized merging. The code is available at https://github.com/hetailang/Merge-Lock.

cs.LG↗

arXiv2025

OWLed: Outlier-weighed Layerwise Pruning for Efficient Autonomous Driving Framework

The integration of Large Language Models (LLMs) into autonomous driving systems offers promising enhancements in environmental understanding and decision-making. However, the substantial computational demands of deploying LLMs locally on vehicles render this approach unfeasible for real-world automotive applications. To address this challenge, we introduce OWLed, the Outlier-Weighed Layerwise Pruning for Efficient Autonomous Driving Framework that leverages outlier-weighted layerwise sparsity for model compression. Our method assigns non-uniform sparsity ratios to different layers based on the distribution of outlier features, significantly reducing the model size without the need for fine-tuning. To ensure the compressed model adapts well to autonomous driving tasks, we incorporate driving environment data into both the calibration and pruning processes. Our empirical studies reveal that the encoder component is more sensitive to pruning than the LLM, highlighting its critical role in the system. Experimental results demonstrate that OWLed outperforms existing methods in perception, action prediction, and language understanding while substantially lowering computational requirements. These findings underscore the potential of combining advanced pruning techniques with LLMs to develop efficient and robust autonomous driving systems capable of handling complex scenarios. Code will be made publicly available.

cs.LG↗

arXiv2025

Mix-LN: Unleashing the Power of Deeper Layers by Combining Pre-LN and Post-LN

Large Language Models (LLMs) have achieved remarkable success, yet recent findings reveal that their deeper layers often contribute minimally and can be pruned without affecting overall performance. While some view this as an opportunity for model compression, we identify it as a training shortfall rooted in the widespread use of Pre-Layer Normalization (Pre-LN). We demonstrate that Pre-LN, commonly employed in models like GPT and LLaMA, leads to diminished gradient norms in its deeper layers, reducing their effectiveness. In contrast, Post-Layer Normalization (Post-LN) preserves larger gradient norms in deeper layers but suffers from vanishing gradients in earlier layers. To address this, we introduce Mix-LN, a novel normalization technique that combines the strengths of Pre-LN and Post-LN within the same model. Mix-LN applies Post-LN to the earlier layers and Pre-LN to the deeper layers, ensuring more uniform gradients across layers. This allows all parts of the network--both shallow and deep layers--to contribute effectively to training. Extensive experiments with various model sizes from 70M to 7B demonstrate that Mix-LN consistently outperforms both Pre-LN and Post-LN, promoting more balanced, healthier gradient norms throughout the network, and enhancing the overall quality of LLM pre-training. Furthermore, we demonstrate that models pre-trained with Mix-LN learn better compared to those using Pre-LN or Post-LN during supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF), highlighting the critical importance of high-quality deep layers. By effectively addressing the inefficiencies of deep layers in current LLMs, Mix-LN unlocks their potential, enhancing model capacity without increasing model size. Our code is available at https://github.com/pixeli99/MixLN.

cs.LG↗

arXiv2025

LOST: Low-rank and Sparse Pre-training for Large Language Models

While large language models (LLMs) have achieved remarkable performance across a wide range of tasks, their massive scale incurs prohibitive computational and memory costs for pre-training from scratch. Recent studies have investigated the use of low-rank parameterization as a means of reducing model size and training cost. In this context, sparsity is often employed as a complementary technique to recover important information lost in low-rank compression by capturing salient features in the residual space. However, existing approaches typically combine low-rank and sparse components in a simplistic or ad hoc manner, often resulting in undesirable performance degradation compared to full-rank training. In this paper, we propose \textbf{LO}w-rank and \textbf{S}parse pre-\textbf{T}raining (\textbf{LOST}) for LLMs, a novel method that ingeniously integrates low-rank and sparse structures to enable effective training of LLMs from scratch under strict efficiency constraints. LOST applies singular value decomposition to weight matrices, preserving the dominant low-rank components, while allocating the remaining singular values to construct channel-wise sparse components to complement the expressiveness of low-rank training. We evaluate LOST on LLM pretraining ranging from 60M to 7B parameters. Our experiments show that LOST achieves competitive or superior performance compared to full-rank models, while significantly reducing both memory and compute overhead. Moreover, Code is available at \href{https://github.com/JiaxiLi1/LOST-Low-rank-and-Sparse-Training-for-Large-Language-Models}{LOST Repo}

cs.LG↗

arXiv2025

GPAS: Accelerating Convergence of LLM Pretraining via Gradient-Preserving Activation Scaling

Modern Large Language Models, such as the LLaMA, Qwen and DeepSeek series, predominantly adopt the Pre-LayerNorm (Pre-LN) Transformer architecture. While being stable during pretraining and scalable to large model sizes, Pre-LN suffers from an exponential growth in activation variance across layers, causing the shortcut to dominate over sub-layer outputs in the residual connection and limiting the learning capacity of deeper layers. To mitigate this issue, we propose Gradient-Preserving Activation Scaling (GPAS), a simple technique that can be used in combination with existing approaches. GPAS works by scaling down the intermediate activations while keeping their gradients unchanged. This leaves information in the activations intact, and avoids the gradient vanishing problem associated with gradient downscaling. Extensive experiments across various model sizes from 71M to 1B show that GPAS achieves consistent performance gains. Beyond enhancing Pre-LN Transformers, GPAS also shows promise in improving alternative architectures such as Sandwich-LN and DeepNorm, demonstrating its versatility and potential for improving training dynamics in a wide range of settings. Our code is available at https://github.com/dandingsky/GPAS.

cs.LG↗

arXiv2025

Outlier Weighed Layerwise Sparsity (OWL): A Missing Secret Sauce for Pruning LLMs to High Sparsity

Large Language Models (LLMs), renowned for their remarkable performance across diverse domains, present a challenge when it comes to practical deployment due to their colossal model size. In response to this challenge, efforts have been directed toward the application of traditional network pruning techniques to LLMs, uncovering a massive number of parameters that can be pruned in one-shot without hurting performance. Prevailing LLM pruning strategies have consistently adhered to the practice of uniformly pruning all layers at equivalent sparsity, resulting in robust performance. However, this observation stands in contrast to the prevailing trends observed in the field of vision models, where non-uniform layerwise sparsity typically yields stronger results. To understand the underlying reasons for this disparity, we conduct a comprehensive study and discover a strong correlation with the emergence of activation outliers in LLMs. Inspired by this finding, we introduce a novel LLM pruning methodology that incorporates a tailored set of non-uniform layerwise sparsity ratios, termed as Outlier Weighed Layerwise sparsity (OWL). The sparsity ratio of OWL is proportional to the outlier ratio observed within each layer, facilitating a more effective alignment between layerwise weight sparsity and outlier ratios. Our empirical evaluation, conducted across the LLaMA-V1 family and OPT, spanning various benchmarks, demonstrates the distinct advantages offered by OWL over previous methods. For instance, OWL exhibits a remarkable performance gain, surpassing the state-of-the-art Wanda and SparseGPT by 61.22 and 6.80 perplexity at a high sparsity level of 70%, respectively, while delivering 2.6x end-to-end inference speed-up in the DeepSparse inference engine. Codes are available at https://github.com/luuyin/OWL.

cs.LG↗

arXiv2025

MagShield: Towards Better Robustness in Sparse Inertial Motion Capture Under Magnetic Disturbances

This paper proposes a novel method called MagShield, designed to address the issue of magnetic interference in sparse inertial motion capture (MoCap) systems. Existing Inertial Measurement Unit (IMU) systems are prone to orientation estimation errors in magnetically disturbed environments, limiting their practical application in real-world scenarios. To address this problem, MagShield employs a "detect-then-correct" strategy, first detecting magnetic disturbances through multi-IMU joint analysis, and then correcting orientation errors using human motion priors. MagShield can be integrated with most existing sparse inertial MoCap systems, improving their performance in magnetically disturbed environments. Experimental results demonstrate that MagShield significantly enhances the accuracy of motion capture under magnetic interference and exhibits good compatibility across different sparse inertial MoCap systems.

cs.CV↗

arXiv2025

GPTailor: Large Language Model Pruning Through Layer Cutting and Stitching

Large language models (LLMs) have shown remarkable capabilities in language understanding and generation. However, such impressive capability typically comes with a substantial model size, which presents significant challenges in deployment and inference. While structured pruning of model parameters offers a promising way to reduce computational costs at deployment time, current methods primarily focus on single model pruning. In this work, we develop a novel strategy to compress models by strategically combining or merging layers from finetuned model variants, which preserves the original model's abilities by aggregating capabilities accentuated in different finetunes. We pose the optimal tailoring of these LLMs as a zero-order optimization problem, adopting a search space that supports three different operations: (1) Layer removal, (2) Layer selection from different candidate models, and (3) Layer merging. Our experiments demonstrate that this approach leads to competitive model pruning, for example, for the Llama2-13B model families, our compressed models maintain approximately 97.3\% of the original performance while removing $\sim25\%$ of parameters, significantly outperforming previous state-of-the-art methods. The code is available at https://github.com/Guinan-Su/auto-merge-llm.

cs.CL↗