arXiv subjects

Lu Yin

Publications and source records attributed to Lu Yin.

At least 73 records · Page 4Linked to original sources

arXiv2025

Chain-of-Experts: Unlocking the Communication Power of Mixture-of-Experts Models

We propose Chain-of-Experts (CoE), a new Mixture-of-Experts (MoE) architecture that introduces sequential expert communication within each layer. Unlike traditional MoE models, where experts operate independently in parallel, CoE processes tokens iteratively across a chain of experts inside a layer. To support dynamic expert selection across iterations, CoE employs a dedicated router at each iteration step within a layer. This design allows tokens to re-evaluate and select different experts during each iteration, rather than being statically assigned. As a result, CoE introduces a flexible routing mechanism that increases the diversity of expert combinations and enriches the model's representational capacity. CoE demonstrates improved performance under fixed compute: on math reasoning tasks, it reduces validation loss from 1.20 to 1.12 compared to a standard MoE. Beyond performance, CoE offers a new scaling axis: depth through expert iteration, which complements conventional width/depth scaling. For example, using 2x iterations matches the performance of 3x expert selections (in width), while reducing memory usage by 17.6-42% relative to other scaling strategies. Our analysis reveals that CoE's benefits stem from its iterative residual structure and enhanced expert specialization empowered by iterative routing, which together unlock more expressive representations. Code is available at https://github.com/ZihanWang314/coe.

cs.LG↗

arXiv2025

Into the Unknown: Applying Inductive Spatial-Semantic Location Embeddings for Predicting Individuals' Mobility Beyond Visited Places

Predicting individuals' next locations is a core task in human mobility modelling, with wide-ranging implications for urban planning, transportation, public policy and personalised mobility services. Traditional approaches largely depend on location embeddings learned from historical mobility patterns, limiting their ability to encode explicit spatial information, integrate rich urban semantic context, and accommodate previously unseen locations. To address these challenges, we explore the application of CaLLiPer -- a multimodal representation learning framework that fuses spatial coordinates and semantic features of points of interest through contrastive learning -- for location embedding in individual mobility prediction. CaLLiPer's embeddings are spatially explicit, semantically enriched, and inductive by design, enabling robust prediction performance even in scenarios involving emerging locations. Through extensive experiments on four public mobility datasets under both conventional and inductive settings, we demonstrate that CaLLiPer consistently outperforms strong baselines, particularly excelling in inductive scenarios. Our findings highlight the potential of multimodal, inductive location embeddings to advance the capabilities of human mobility prediction systems. We also release the code and data (https://github.com/xlwang233/Into-the-Unknown) to foster reproducibility and future research.

cs.AI↗

arXiv2025

Outlier-weighed Layerwise Sampling for LLM Fine-tuning

The rapid advancements in Large Language Models (LLMs) have revolutionized various natural language processing tasks. However, the substantial size of LLMs presents significant challenges in training or fine-tuning. While parameter-efficient approaches such as low-rank adaptation (LoRA) have gained popularity, they often compromise performance compared to full-rank fine-tuning. In this paper, we propose Outlier-weighed Layerwise Sampling (OWS), a new memory-efficient fine-tuning approach, inspired by the layerwise outlier distribution of LLMs. Unlike LoRA, which adds extra adapters to all layers, OWS strategically assigns higher sampling probabilities to layers with more outliers, selectively sampling only a few layers and fine-tuning their pre-trained weights. To further increase the number of fine-tuned layers without a proportional rise in memory costs, we incorporate gradient low-rank projection, further boosting the approach's performance. Our extensive experiments across various architectures, including LLaMa2 and Mistral, demonstrate that OWS consistently outperforms baseline approaches, including full fine-tuning. Specifically, it achieves up to a 1.1% average accuracy gain on the Commonsense Reasoning benchmark, a 3.0% improvement on MMLU, and a notable 10% boost on MT-Bench, while being more memory efficient. OWS allows us to fine-tune 7B LLMs with only 21GB of memory. Our code is available at https://github.com/pixeli99/OWS.

cs.LG↗

arXiv2025

From Low Rank Gradient Subspace Stabilization to Low-Rank Weights: Observations, Theories, and Applications

Large Language Models' (LLMs) weight matrices can often be expressed in low-rank form with potential to relax memory and compute resource requirements. Unlike prior efforts that focus on developing novel matrix decompositions, in this work we study the non-uniform low-rank properties of weight matrices in LLMs through the lens of stabilizing gradient subspace. First, we provide a theoretical framework to understand the stabilization of gradient subspaces through Hessian analysis. Second, we empirically establish an important relationship between gradient dynamics and low-rank expressiveness of weight matrices. Our findings reveal that different LLM components exhibit varying levels of converged low-rank structures, necessitating variable rank reduction across them to minimize drop in performance due to compression. Drawing on this result, we present Weight Low-Rank Projection(WeLore) that unifies weight compression and memory-efficient fine-tuning into one, in a data-agnostic and one-shot manner. When used as a compression technique, WeLore categorizes weight matrices into Low-rank Components (LRCs) and Non-Low-rank Components (N-LRCs) and suitably encodes them for minimum performance loss. Our gradient dynamics perspective illustrates that LRCs tend to have better fine-tuning capabilities and their standalone fine-tuning can closely mimic and sometimes outperform the training loss trajectory and performance of full fine-tuning with notable memory and compute footprint reduction. Codes are available at https://github.com/VITA-Group/WeLore.

cs.LG↗

arXiv2025

Light quark energy loss in the flavor-dependent systems from holography

Using the holographic model of finite-endpoint-momentum shooting string approach, we study the instantaneous energy loss of light quarks for the flavor-dependent systems with $N_f = 0$, $N_f = 2$, and $N_f = 2+1$ in the Einstein-Maxwell-dilaton (EMD) framework. In particular, we investigate for the first time the impact of the flavor content of the strongly coupled QGP medium on the instantaneous energy loss of light quarks. It turns out that the instantaneous energy loss of light quarks is smallest for $N_f = 0$, and adding $u(d)$ quarks and $s$ quark in the system increases this energy loss. In addition, we found that the instantaneous energy loss in the strongly coupled plasma is minimal near the critical temperature, but it increases as the system moves away from the critical endpoint due to rising temperature or chemical potential.

hep-ph↗

arXiv2025

Do high redshift QSOs and GRBs corroborate JWST?

The James Webb Space Telescope (JWST) is reporting massive high redshift galaxies that appear challenging from the $Λ$CDM perspective. Interpreted as a cosmological problem, this necessitates the Planck collaboration underestimating either matter density $Ω_m$ or physical matter density $Ω_m h^2$ at higher redshifts. Through standard frequentist profile likelihoods, we identify corroborating quasar (QSO) and gamma-ray burst (GRB) data sets where $Ω_m$ increases with effective redshift $z_{\textrm{eff}}$ and remains anomalously large at higher redshifts. We relax the traditional priors by allowing for $Ω_m > 1$, consistent with negative dark energy density, to locate profile likelihood peaks where possible. While the variation of $Ω_m$ with $z_{\textrm{eff}}$ is at odds with the $Λ$CDM model, demarcating frequentist confidence intervals through differences in $χ^2$ in profile likelihoods, the prevailing technique in the literature, points to $3.9 σ$ and $7.9 σ$ tensions between GRBs and QSOs, respectively, and Planck-$Λ$CDM. We explain the approximations inherent in the existing profile likelihood literature, and highlight fresh methodology that generalises the prescription. We show that alternative methods, including Bayesian approaches, lead to similar tensions. Finally, in the large sample limit, we show that Feldman-Cousins prescription for frequentist confidence intervals in the presence of a boundary (prior) leads to confidence intervals that are bounded above by Wilks' theorem.

astro-ph.CO↗

arXiv2025

AgentAlign: Navigating Safety Alignment in the Shift from Informative to Agentic Large Language Models

The acquisition of agentic capabilities has transformed LLMs from "knowledge providers" to "action executors", a trend that while expanding LLMs' capability boundaries, significantly increases their susceptibility to malicious use. Previous work has shown that current LLM-based agents execute numerous malicious tasks even without being attacked, indicating a deficiency in agentic use safety alignment during the post-training phase. To address this gap, we propose AgentAlign, a novel framework that leverages abstract behavior chains as a medium for safety alignment data synthesis. By instantiating these behavior chains in simulated environments with diverse tool instances, our framework enables the generation of highly authentic and executable instructions while capturing complex multi-step dynamics. The framework further ensures model utility by proportionally synthesizing benign instructions through non-malicious interpretations of behavior chains, precisely calibrating the boundary between helpfulness and harmlessness. Evaluation results on AgentHarm demonstrate that fine-tuning three families of open-source models using our method substantially improves their safety (35.8% to 79.5% improvement) while minimally impacting or even positively enhancing their helpfulness, outperforming various prompting methods. The dataset and code have both been open-sourced.

cs.CR↗

arXiv2025

NeuroTrails: Training with Dynamic Sparse Heads as the Key to Effective Ensembling

Model ensembles have long been a cornerstone for improving generalization and robustness in deep learning. However, their effectiveness often comes at the cost of substantial computational overhead. To address this issue, state-of-the-art methods aim to replicate ensemble-class performance without requiring multiple independently trained networks. Unfortunately, these algorithms often still demand considerable compute at inference. In response to these limitations, we introduce $\textbf{NeuroTrails}$, a sparse multi-head architecture with dynamically evolving topology. This unexplored model-agnostic training paradigm improves ensemble performance while reducing the required resources. We analyze the underlying reason for its effectiveness and observe that the various neural trails induced by dynamic sparsity attain a $\textit{Goldilocks zone}$ of prediction diversity. NeuroTrails displays efficacy with convolutional and transformer-based architectures on computer vision and language tasks. Experiments on ResNet-50/ImageNet, LLaMA-350M/C4, among many others, demonstrate increased accuracy and stronger robustness in zero-shot generalization, while requiring significantly fewer parameters.

cs.LG↗

arXiv2025

SMILE: a universal tool for modulated-enhanced localization microscopy to achieve minimal three-dimensional resolution

Modulation-enhanced localization microscopy (MELM) has demonstrated significant improvements in both lateral and axial localization precision compared to conventional single-molecule localization microscopy (SMLM). However, lateral modulated illumination based MELM (MELMxy) remains fundamentally limited to two-dimensional imaging. Here we present three-dimensional Single-Molecule Modulated Illumination Localization Estimator (SMILE) that synergistically integrates lateral illumination modulation with point spread function engineering. By simultaneously exploiting lateral modulation patterns and an accurate point spread function (PSF) model for 3D localization, SMILE achieves near-theoretical-minimum localization uncertainty, demonstrating an average 4-fold enhancement in lateral precision compared to conventional 3D-SMLM. Crucially, SMILE exhibits exceptional compatibility with diverse PSFs and different illumination patterns with various structures including 4Pi configurations, making it a versatile tool that can be easily adapted for different experimental setups. When integrated with 4Pi microscopy, 4Pi-SMILE shows particular promise for achieving sub-10 nm axial resolution and approaching isotropic resolution. From the simulations and proof-of-concept experiments, we verified the superiority of SMILE over 3D-SMLM and ordinary MELM. We highlight SMILE as a novel methodology and robust framework that holds great potential to significantly promote the development of MELM.

physics.optics↗

arXiv2025

Does Gauss-Bonnet Inflationary Gravitational Waves satisfy the Pulsar Timing Arrays observations?

The observations from pulsar timing arrays (PTAs), led by the North American Nanohertz Observatory for Gravitational Waves (NANOGrav), have provided opportunities to constrain primordial gravitational waves at low frequencies. In this paper, we analyze the best-fit parameter values for different Gauss-Bonnet Inflationary Gravitational Wave (GB-IGW) models with the PTArcade program, and we compare the results with the observations of NANOGrav, European Pulsar Timing Array (EPTA), Parkes Pulsar Timing Array (PPTA), and International Pulsar Timing Array (IPTA). We find the potential parameter $n$ derived from the GB-IGW model is not necessarily positive. Instead, PTA data suggest the possibility of new parameter ranges. Meanwhile, the other parameters' results, such as the spectral indices of the scalar and tensor perturbation also have differences with traditional Cosmic Microwave Background analyses from Planck 2018, showing the reason for the different phenomena of Gravitational Wave power spectra expected. Additionally, the GB-IGW models we calculated are closer to the central values of multiple PTA datasets compared to the standard inflation model, making the GB-IGW model more likely to be detected by future space-based gravitational wave observatories.

astro-ph.CO↗

arXiv2025

TODO: Enhancing LLM Alignment with Ternary Preferences

Aligning large language models (LLMs) with human intent is critical for enhancing their performance across a variety of tasks. Standard alignment techniques, such as Direct Preference Optimization (DPO), often rely on the binary Bradley-Terry (BT) model, which can struggle to capture the complexities of human preferences -- particularly in the presence of noisy or inconsistent labels and frequent ties. To address these limitations, we introduce the Tie-rank Oriented Bradley-Terry model (TOBT), an extension of the BT model that explicitly incorporates ties, enabling more nuanced preference representation. Building on this, we propose Tie-rank Oriented Direct Preference Optimization (TODO), a novel alignment algorithm that leverages TOBT's ternary ranking system to improve preference alignment. In evaluations on Mistral-7B and Llama 3-8B models, TODO consistently outperforms DPO in modeling preferences across both in-distribution and out-of-distribution datasets. Additional assessments using MT Bench and benchmarks such as Piqa, ARC-c, and MMLU further demonstrate TODO's superior alignment performance. Notably, TODO also shows strong results in binary preference alignment, highlighting its versatility and potential for broader integration into LLM alignment. The implementation details can be found in https://github.com/XXares/TODO.

cs.CL↗

arXiv2025

Seeing and Reasoning with Confidence: Supercharging Multimodal LLMs with an Uncertainty-Aware Agentic Framework

Multimodal large language models (MLLMs) show promise in tasks like visual question answering (VQA) but still face challenges in multimodal reasoning. Recent works adapt agentic frameworks or chain-of-thought (CoT) reasoning to improve performance. However, CoT-based multimodal reasoning often demands costly data annotation and fine-tuning, while agentic approaches relying on external tools risk introducing unreliable output from these tools. In this paper, we propose Seeing and Reasoning with Confidence (SRICE), a training-free multimodal reasoning framework that integrates external vision models with uncertainty quantification (UQ) into an MLLM to address these challenges. Specifically, SRICE guides the inference process by allowing MLLM to autonomously select regions of interest through multi-stage interactions with the help of external tools. We propose to use a conformal prediction-based approach to calibrate the output of external tools and select the optimal tool by estimating the uncertainty of an MLLM's output. Our experiment shows that the average improvement of SRICE over the base MLLM is 4.6% on five datasets and the performance on some datasets even outperforms fine-tuning-based methods, revealing the significance of ensuring reliable tool use in an MLLM agent.

cs.AI↗

arXiv2025

E2ENet: Dynamic Sparse Feature Fusion for Accurate and Efficient 3D Medical Image Segmentation

Deep neural networks have evolved as the leading approach in 3D medical image segmentation due to their outstanding performance. However, the ever-increasing model size and computation cost of deep neural networks have become the primary barrier to deploying them on real-world resource-limited hardware. In pursuit of improving performance and efficiency, we propose a 3D medical image segmentation model, named Efficient to Efficient Network (E2ENet), incorporating two parametrically and computationally efficient designs. i. Dynamic sparse feature fusion (DSFF) mechanism: it adaptively learns to fuse informative multi-scale features while reducing redundancy. ii. Restricted depth-shift in 3D convolution: it leverages the 3D spatial information while keeping the model and computational complexity as 2D-based methods. We conduct extensive experiments on BTCV, AMOS-CT and Brain Tumor Segmentation Challenge, demonstrating that E2ENet consistently achieves a superior trade-off between accuracy and efficiency than prior arts across various resource constraints. E2ENet achieves comparable accuracy on the large-scale challenge AMOS-CT, while saving over 68\% parameter count and 29\% FLOPs in the inference phase, compared with the previous best-performing method. Our code has been made available at: https://github.com/boqian333/E2ENet-Medical.

cs.CV↗

arXiv2025

Sebra: Debiasing Through Self-Guided Bias Ranking

Ranking samples by fine-grained estimates of spuriosity (the degree to which spurious cues are present) has recently been shown to significantly benefit bias mitigation, over the traditional binary biased-\textit{vs}-unbiased partitioning of train sets. However, this spuriosity ranking comes with the requirement of human supervision. In this paper, we propose a debiasing framework based on our novel \ul{Se}lf-Guided \ul{B}ias \ul{Ra}nking (\emph{Sebra}), that mitigates biases (spurious correlations) via an automatic ranking of data points by spuriosity within their respective classes. Sebra leverages a key local symmetry in Empirical Risk Minimization (ERM) training -- the ease of learning a sample via ERM inversely correlates with its spuriousity; the fewer spurious correlations a sample exhibits, the harder it is to learn, and vice versa. However, globally across iterations, ERM tends to deviate from this symmetry. Sebra dynamically steers ERM to correct this deviation, facilitating the sequential learning of attributes in increasing order of difficulty, \ie, decreasing order of spuriosity. As a result, the sequence in which Sebra learns samples naturally provides spuriousity rankings. We use the resulting fine-grained bias characterization in a contrastive learning framework to mitigate biases from multiple sources. Extensive experiments show that Sebra consistently outperforms previous state-of-the-art unsupervised debiasing techniques across multiple standard benchmarks, including UrbanCars, BAR, CelebA, and ImageNet-1K. Code, pre-trained models, and training logs are available at https://kadarsh22.github.io/sebra_iclr25/.

cs.LG↗

arXiv2024

Observational evidence for Early Dark Energy as a unified explanation for Cosmic Birefringence and the Hubble tension

We test the $n$=3 Ultralight Axion-like model of Early Dark Energy (EDE) with the observationsof the $EB$ mode of the cosmic microwave background (CMB) radiation, and local expansion rate measurements. Our results show that the shape of the CMB $EB$ angular power spectrum is sensitive to the background cosmological parameters. We run Markov chain Monte Carlo (MCMC) simulations to fit the $Λ$CDM + EDE parameters simultaneously, and find that the EDE model with $n$=3 can provide a good fit to the observed CMB $EB$ spectra, consistent with the locally measured value of the Hubble constant. Our result is the first to show that axion-like EDE can provide a unified explanation for the observed cosmic birefringence and the Hubble tension.

astro-ph.CO↗

arXiv2024

Are Sparse Neural Networks Better Hard Sample Learners?

While deep learning has demonstrated impressive progress, it remains a daunting challenge to learn from hard samples as these samples are usually noisy and intricate. These hard samples play a crucial role in the optimal performance of deep neural networks. Most research on Sparse Neural Networks (SNNs) has focused on standard training data, leaving gaps in understanding their effectiveness on complex and challenging data. This paper's extensive investigation across scenarios reveals that most SNNs trained on challenging samples can often match or surpass dense models in accuracy at certain sparsity levels, especially with limited data. We observe that layer-wise density ratios tend to play an important role in SNN performance, particularly for methods that train from scratch without pre-trained initialization. These insights enhance our understanding of SNNs' behavior and potential for efficient learning approaches in data-centric AI. Our code is publicly available at: \url{https://github.com/QiaoXiao7282/hard_sample_learners}.

cs.CV↗

arXiv2024

Late-time Cosmology without Dark Sector but with Closed String Massless Sector

We explore the possibility of solving the dark energy and the coincidence problems by postulating the massless sector of closed strings. This sector constitutes the gravitational multiplet of string theory and, when applied to four-dimensional cosmology, predicts that \textit{the expansion of an open Universe defined in string frame is readily accelerating}. We confront the prediction with the late-time cosmological data of Type Ia supernovae and quasar absorption spectrum, which probe the evolutions of the Hubble parameter and possibly the fine-structure constant. We report that these observations are in admirable agreement with the prediction without any dark sector or coincidence problem. We estimate the Hubble constant, $H_{0}\simeq 71.2\pm 0.2\,\mathrm{km/s/Mpc}$.

hep-th↗

arXiv2024

Aspect-Based Few-Shot Learning

We generalize the formulation of few-shot learning by introducing the concept of an aspect. In the traditional formulation of few-shot learning, there is an underlying assumption that a single "true" label defines the content of each data point. This label serves as a basis for the comparison between the query object and the objects in the support set. However, when a human expert is asked to execute the same task without a predefined set of labels, they typically consider the rest of the data points in the support set as context. This context specifies the level of abstraction and the aspect from which the comparison can be made. In this work, we introduce a novel architecture and training procedure that develops a context given the query and support set and implements aspect-based few-shot learning that is not limited to a predetermined set of classes. We demonstrate that our method is capable of forming and using an aspect for few-shot learning on the Geometric Shapes and Sprites dataset. The results validate the feasibility of our approach compared to traditional few-shot learning.

cs.CV↗