arXiv subjects

Hanyu Zhang

Publications and source records attributed to Hanyu Zhang.

At least 37 records · Page 2Linked to original sources

arXiv2025

Non-Uniform Class-Wise Coreset Selection for Vision Model Fine-tuning

Coreset selection aims to identify a small yet highly informative subset of data, thereby enabling more efficient model training while reducing storage overhead. Recently, this capability has been leveraged to tackle the challenges of fine-tuning large foundation models, offering a direct pathway to their efficient and practical deployment. However, most existing methods are class-agnostic, causing them to overlook significant difficulty variations among classes. This leads them to disproportionately prune samples from either overly easy or hard classes, resulting in a suboptimal allocation of the data budget that ultimately degrades the final coreset performance. To address this limitation, we propose Non-Uniform Class-Wise Coreset Selection (NUCS), a novel framework that both integrates class-level and sample-level difficulty. We propose a robust metric for global class difficulty, quantified as the winsorized average of per-sample difficulty scores. Guided by this metric, our method performs a theoretically-grounded, non-uniform allocation of data selection budgets inter-class, while adaptively selecting samples intra-class with optimal difficulty ranges. Extensive experiments on a wide range of visual classification tasks demonstrate that NUCS consistently outperforms state-of-the-art methods across 10 diverse datasets and pre-trained models, achieving both superior accuracy and computational efficiency, highlighting the promise of non-uniform class-wise selection strategy for advancing the efficient fine-tuning of large foundation models.

cs.LG↗

arXiv2025

Mind the Gaps: Auditing and Reducing Group Inequity in Large-Scale Mobility Prediction

Next location prediction underpins a growing number of mobility, retail, and public-health applications, yet its societal impacts remain largely unexplored. In this paper, we audit state-of-the-art mobility prediction models trained on a large-scale dataset, highlighting hidden disparities based on user demographics. Drawing from aggregate census data, we compute the difference in predictive performance on racial and ethnic user groups and show a systematic disparity resulting from the underlying dataset, resulting in large differences in accuracy based on location and user groups. To address this, we propose Fairness-Guided Incremental Sampling (FGIS), a group-aware sampling strategy designed for incremental data collection settings. Because individual-level demographic labels are unavailable, we introduce Size-Aware K-Means (SAKM), a clustering method that partitions users in latent mobility space while enforcing census-derived group proportions. This yields proxy racial labels for the four largest groups in the state: Asian, Black, Hispanic, and White. Built on these labels, our sampling algorithm prioritizes users based on expected performance gains and current group representation. This method incrementally constructs training datasets that reduce demographic performance gaps while preserving overall accuracy. Our method reduces total disparity between groups by up to 40\% with minimal accuracy trade-offs, as evaluated on a state-of-art MetaPath2Vec model and a transformer-encoder model. Improvements are most significant in early sampling stages, highlighting the potential for fairness-aware strategies to deliver meaningful gains even in low-resource settings. Our findings expose structural inequities in mobility prediction pipelines and demonstrate how lightweight, data-centric interventions can improve fairness with little added complexity, especially for low-data applications.

cs.LG↗

arXiv2025

Efficiently Generating Correlated Sample Paths from Multi-step Time Series Foundation Models

Many time series applications require access to multi-step forecast trajectories in the form of sample paths. Recently, time series foundation models have leveraged multi-step lookahead predictions to improve the quality and efficiency of multi-step forecasts. However, these models only predict independent marginal distributions for each time step, rather than a full joint predictive distribution. To generate forecast sample paths with realistic correlation structures, one typically resorts to autoregressive sampling, which can be extremely expensive. In this paper, we present a copula-based approach to efficiently generate accurate, correlated sample paths from existing multi-step time series foundation models in one forward pass. Our copula-based approach generates correlated sample paths orders of magnitude faster than autoregressive sampling, and it yields improved sample path quality by mitigating the snowballing error phenomenon.

cs.LG↗

arXiv2025

Classifier-Centric Adaptive Framework for Open-Vocabulary Camouflaged Object Segmentation

Open-vocabulary camouflaged object segmentation requires models to segment camouflaged objects of arbitrary categories unseen during training, placing extremely high demands on generalization capabilities. Through analysis of existing methods, it is observed that the classification component significantly affects overall segmentation performance. Accordingly, a classifier-centric adaptive framework is proposed to enhance segmentation performance by improving the classification component via a lightweight text adapter with a novel layered asymmetric initialization. Through the classification enhancement, the proposed method achieves substantial improvements in segmentation metrics compared to the OVCoser baseline on the OVCamo benchmark: cIoU increases from 0.443 to 0.493, cSm from 0.579 to 0.658, and cMAE reduces from 0.336 to 0.239. These results demonstrate that targeted classification enhancement provides an effective approach for advancing camouflaged object segmentation performance.

cs.CV↗

arXiv2025

KnowMT-Bench: Benchmarking Knowledge-Intensive Long-Form Question Answering in Multi-Turn Dialogues

Multi-Turn Long-Form Question Answering (MT-LFQA) is a key application paradigm of Large Language Models (LLMs) in knowledge-intensive domains. However, existing benchmarks are limited to single-turn dialogue, while multi-turn dialogue benchmarks typically assess other orthogonal capabilities rather than knowledge-intensive factuality. To bridge this critical gap, we introduce \textbf{KnowMT-Bench}, the \textit{first-ever} benchmark designed to systematically evaluate MT-LFQA for LLMs across knowledge-intensive fields, including medicine, finance, and law. To faithfully assess the model's real-world performance, KnowMT-Bench employs a dynamic evaluation setting where models generate their own multi-turn dialogue histories given logically progressive question sequences. The factual capability and information delivery efficiency of the \textit{final-turn} answer are then evaluated using a human-validated automated pipeline. Our experiments reveal that multi-turn contexts degrade performance: factual capability declines due to the contextual noise from self-generated histories, while information efficiency drops as models become more verbose with increasing dialogue length. We then investigate mitigation strategies, demonstrating that retrieval-augmented generation (RAG) can effectively alleviate and even reverse this factual degradation. These findings underscore the importance of our benchmark in evaluating and enhancing the conversational factual capabilities of LLMs in real-world knowledge-intensive applications. Code is available at \href{https://github.com/hardenyu21/KnowMT-Bench}{\textcolor{cyan}{\texttt{KnowMT-Bench}}}.

cs.CL↗

arXiv2025

GLM-4.5: Agentic, Reasoning, and Coding (ARC) Foundation Models

We present GLM-4.5, an open-source Mixture-of-Experts (MoE) large language model with 355B total parameters and 32B activated parameters, featuring a hybrid reasoning method that supports both thinking and direct response modes. Through multi-stage training on 23T tokens and comprehensive post-training with expert model iteration and reinforcement learning, GLM-4.5 achieves strong performance across agentic, reasoning, and coding (ARC) tasks, scoring 70.1% on TAU-Bench, 91.0% on AIME 24, and 64.2% on SWE-bench Verified. With much fewer parameters than several competitors, GLM-4.5 ranks 3rd overall among all evaluated models and 2nd on agentic benchmarks. We release both GLM-4.5 (355B parameters) and a compact version, GLM-4.5-Air (106B parameters), to advance research in reasoning and agentic AI systems. Code, models, and more information are available at https://github.com/zai-org/GLM-4.5.

cs.CL↗

arXiv2025

Gradient-Adaptive Policy Optimization: Towards Multi-Objective Alignment of Large Language Models

Reinforcement Learning from Human Feedback (RLHF) has emerged as a powerful technique for aligning large language models (LLMs) with human preferences. However, effectively aligning LLMs with diverse human preferences remains a significant challenge, particularly when they are conflict. To address this issue, we frame human value alignment as a multi-objective optimization problem, aiming to maximize a set of potentially conflicting objectives. We introduce Gradient-Adaptive Policy Optimization (GAPO), a novel fine-tuning paradigm that employs multiple-gradient descent to align LLMs with diverse preference distributions. GAPO adaptively rescales the gradients for each objective to determine an update direction that optimally balances the trade-offs between objectives. Additionally, we introduce P-GAPO, which incorporates user preferences across different objectives and achieves Pareto solutions that better align with the user's specific needs. Our theoretical analysis demonstrates that GAPO converges towards a Pareto optimal solution for multiple objectives. Empirical results on Mistral-7B show that GAPO outperforms current state-of-the-art methods, achieving superior performance in both helpfulness and harmlessness.

cs.CL↗

arXiv2025

Enhancing DESI DR1 Full-Shape analyses using HOD-informed priors

We present an analysis of DESI Data Release 1 (DR1) that incorporates Halo Occupation Distribution (HOD)-informed priors into Full-Shape (FS) modeling of the power spectrum based on cosmological perturbation theory (PT). By leveraging physical insights from the galaxy-halo connection, these HOD-informed priors on nuisance parameters substantially mitigate projection effects in extended cosmological models that allow for dynamical dark energy. The resulting credible intervals now encompass the posterior maximum from the baseline analysis using gaussian priors, eliminating a significant posterior shift observed in baseline studies. In the $Λ$CDM framework, a combined DESI DR1 FS information and constraints from the DESI DR1 baryon acoustic oscillations (BAO)-including Big Bang Nucleosynthesis (BBN) constraints and a weak prior on the scalar spectral index-yields $Ω_{\rm m} = 0.2994\pm 0.0090$ and $σ_8 = 0.836^{+0.024}_{-0.027}$, representing improvements of approximately 4% and 23% over the baseline analysis, respectively. For the $w_0w_a$CDM model, our results from various data combinations are highly consistent, with all configurations converging to a region with $w_0 > -1$ and $w_a < 0$. This convergence not only suggests intriguing hints of dynamical dark energy but also underscores the robustness of our HOD-informed prior approach in delivering reliable cosmological constraints.

astro-ph.CO↗

arXiv2025

A Self-Supervised Paradigm for Data-Efficient Medical Foundation Model Pre-training: V-information Optimization Framework

Self-supervised pre-training medical foundation models on large-scale datasets demonstrate exceptional performance. Recent research challenges this common paradigm by introducing data-effective learning approaches, demonstrating that merely increasing pre-training data volume does not necessarily improve model performance. However, current methods still have unclear standards and the underlying theoretical foundation remains unknown. In this paper, as the first attempt to address this limitation, we introduce V-information into self-supervised pre-training of foundation models to provide a theoretical foundation for sample selection. Our derivation confirms that by optimizing V-information, sample selection can be framed as an optimization problem where choosing diverse and challenging samples enhances model performance even under limited training data. Under this guidance, we develop an optimized data-effective learning method (OptiDEL) to optimize V-information in real-world medical domains by generating more diverse and harder samples. We compare the OptiDEL method with state-of-the-art approaches finding that OptiDEL consistently outperforms existing approaches across eight different datasets, with foundation models trained on only 5% of the pre-training data achieving up to 6.2% higher mIoU than those trained on the full dataset. Remarkably, OptiDEL demonstrates an average improvement of 4.7% mIoU over competing methods while using 20x less training data.

cs.LG↗

arXiv2025

HOD-informed prior for EFT-based full-shape analyses of LSS

To improve the performance of full-shape analyses of large-scale structure, we consider using a halo occupation distribution (HOD)-informed prior for the effective field theory (EFT) nuisance parameters. We generate 320 000 mock galaxy catalogs using 10 000 sets of HOD parameters across 32 simulation boxes with different cosmologies. We measure and fit the redshift-space power spectra using a fast emulator of the EFT model, and the resulting best-fit EFT parameter distributions are used to create the prior. This prior effectively constrains the EFT nuisance parameter space, limiting it to the space of HOD-mocks that can be well fit by a EFT model. We have tested the stability of the prior under different configurations, including the effect of varying the HOD sample distribution and the inclusion of the hexadecapole moment. We find that our HOD-informed prior and the cosmological parameter constraints derived using it are robust. While cosmological fits using the standard EFT prior suffer from prior effects, sometimes failing to recover the true cosmology within Bayesian credible intervals, the HOD-informed prior mitigates these issues and significantly improves cosmological parameter recovery for $Λ$CDM and beyond. This work lays the foundation for better full-shape large-scale structure analyses in current and upcoming galaxy surveys, making it a valuable tool for addressing key questions in cosmology.

astro-ph.CO↗

arXiv2025

Coherence Properties of Rare-Earth Spins in Micrometer-Thin Films

Rare-earth ions in bulk crystals are excellent solid-state quantum systems in quantum information science, owing to the exceptional optical and spin coherence properties. However, the weak fluorescence of single rare-earth ions present a significant challenge for scalability, necessitating the integration into micro-cavities. Thin films serve as a promising material platform for the integration, yet the fabrication without compromising the properties of the materials and rare-earth ions remains challenging. In this work, we fabricate micrometer-thin yttrium aluminum garnet (YAG) films from bulk crystals using ion implantation techniques. The resulting films preserve the single-crystalline structure of the original bulk crystal. Notably, the embedded rare-earth ions are photo-stable and exhibit bulk-like spin coherence properties. Our results demonstrate the compatibility of bulk-like spin properties with the thin-film fabrication technique, facilitating the efficient integration of rare-earth ions into on-chip photonic devices and advancing the applications of rare-earth ions systems in quantum technologies.

quant-ph↗

arXiv2025

Controlling Large Language Models Through Concept Activation Vectors

As large language models (LLMs) are widely deployed across various domains, the ability to control their generated outputs has become more critical. This control involves aligning LLMs outputs with human values and ethical principles or customizing LLMs on specific topics or styles for individual users. Existing controlled generation methods either require significant computational resources and extensive trial-and-error or provide coarse-grained control. In this paper, we propose Generation with Concept Activation Vector (GCAV), a lightweight model control framework that ensures accurate control without requiring resource-extensive fine-tuning. Specifically, GCAV first trains a concept activation vector for specified concepts to be controlled, such as toxicity. During inference, GCAV steers the concept vector in LLMs, for example, by removing the toxicity concept vector from the activation layers. Control experiments from different perspectives, including toxicity reduction, sentiment control, linguistic style, and topic control, demonstrate that our framework achieves state-of-the-art performance with granular control, allowing for fine-grained adjustments of both the steering layers and the steering magnitudes for individual samples.

cs.CL↗

arXiv2025

Baichuan4-Finance Technical Report

Large language models (LLMs) have demonstrated strong capabilities in language understanding, generation, and reasoning, yet their potential in finance remains underexplored due to the complexity and specialization of financial knowledge. In this work, we report the development of the Baichuan4-Finance series, including a comprehensive suite of foundational Baichuan4-Finance-Base and an aligned language model Baichuan4-Finance, which are built upon Baichuan4-Turbo base model and tailored for finance domain. Firstly, we have dedicated significant effort to building a detailed pipeline for improving data quality. Moreover, in the continual pre-training phase, we propose a novel domain self-constraint training strategy, which enables Baichuan4-Finance-Base to acquire financial knowledge without losing general capabilities. After Supervised Fine-tuning and Reinforcement Learning from Human Feedback and AI Feedback, the chat model Baichuan4-Finance is able to tackle various financial certification questions and real-world scenario applications. We evaluate Baichuan4-Finance on many widely used general datasets and two holistic financial benchmarks. The evaluation results show that Baichuan4-Finance-Base surpasses almost all competitive baselines on financial tasks by significant margins without sacrificing performance on general LLM benchmarks. At the same time, Baichuan4-Finance demonstrates even more impressive performance on financial application scenarios, showcasing its potential to foster community innovation in the financial LLM field.

cs.CL↗

arXiv2024

LLMForecaster: Improving Seasonal Event Forecasts with Unstructured Textual Data

Modern time-series forecasting models often fail to make full use of rich unstructured information about the time series themselves. This lack of proper conditioning can lead to obvious model failures; for example, models may be unaware of the details of a particular product, and hence fail to anticipate seasonal surges in customer demand in the lead up to major exogenous events like holidays for clearly relevant products. To address this shortcoming, this paper introduces a novel forecast post-processor -- which we call LLMForecaster -- that fine-tunes large language models (LLMs) to incorporate unstructured semantic and contextual information and historical data to improve the forecasts from an existing demand forecasting pipeline. In an industry-scale retail application, we demonstrate that our technique yields statistically significantly forecast improvements across several sets of products subject to holiday-driven demand surges.

cs.LG↗

arXiv2024

Weather-Informed Probabilistic Forecasting and Scenario Generation in Power Systems

The integration of renewable energy sources (RES) into power grids presents significant challenges due to their intrinsic stochasticity and uncertainty, necessitating the development of new techniques for reliable and efficient forecasting. This paper proposes a method combining probabilistic forecasting and Gaussian copula for day-ahead prediction and scenario generation of load, wind, and solar power in high-dimensional contexts. By incorporating weather covariates and restoring spatio-temporal correlations, the proposed method enhances the reliability of probabilistic forecasts in RES. Extensive numerical experiments compare the effectiveness of different time series models, with performance evaluated using comprehensive metrics on a real-world and high-dimensional dataset from Midcontinent Independent System Operator (MISO). The results highlight the importance of weather information and demonstrate the efficacy of the Gaussian copula in generating realistic scenarios, with the proposed weather-informed Temporal Fusion Transformer (WI-TFT) model showing superior performance.

stat.ML↗

arXiv2024

Optimization-based Learning for Dynamic Load Planning in Trucking Service Networks

The load planning problem is a critical challenge in service network design for parcel carriers: it decides how many trailers to assign for dispatch over time between pairs of terminals. Another key challenge is to determine a flow plan, which specifies how parcel volumes are assigned to planned loads. This paper considers the Outbound Load Planning Problem (OLPP) that considers flow and load planning challenges jointly in order to adjust loads and flows as the demand forecast changes over time before the day of operations in a terminal. The paper aims at developing a decision-support tool to inform planners making these decisions at terminals across the network. The paper formulates the OLPP as a mixed-integer programming model and shows that it admits a large number of symmetries in a network where each commodity can be routed through primary and alternate terminals. As a result, an optimization solver may return fundamentally different solutions to closely related problems, confusing planners and reducing trust in optimization. To remedy this limitation, this paper proposes a lexicographical optimization approach that eliminates those symmetries by generating optimal solutions staying close to a reference plan. Moreover, this paper designs an optimization proxy that addresses the computational challenges of the optimization model. The optimization proxy combines a machine-learning model and a repair procedure to find near-optimal solutions that satisfy real-time constraints imposed by planners in the loop. An extensive computational study on industrial instances shows that the optimization proxy is orders of magnitude faster for generating solutions that are consistent with each other. The proposed approach also demonstrates the benefits of the OLPP for load consolidation and the significant savings obtained from combining machine learning and optimization.

cs.AI↗

arXiv2024

Emulating power spectra for pre- and post-reconstructed galaxy samples

The small-scale linear information in galaxy samples typically lost during non-linear growth can be restored to a certain level by the density field reconstruction, which has been demonstrated for improving the precision of the baryon acoustic oscillations (BAO) measurements. As proposed in the literature, a joint analysis of the power spectrum before and after the reconstruction enables an efficient extraction of information carried by high-order statistics. However, the statistics of the post-reconstruction density field are difficult to model. In this work, we circumvent this issue by developing an accurate emulator for the pre-reconstructed, post-reconstructed, and cross power spectra ($P_{\rm pre}$, $P_{\rm post}$, $P_{\rm cross}$) up to $k=0.5~h~{\rm Mpc^{-1}}$ based on the \textsc{Dark Quest} N-body simulations. The accuracy of the emulator is at percent level, namely, the error of the emulated monopole and quadrupole of the power spectra is less than $1\%$ and $10\%$ of the ground truth, respectively. A fit to an example power spectra using the emulator shows that the constraints on cosmological parameters get largely improved using $P_{\rm pre}$+$P_{\rm post}$+$P_{\rm cross}$ with $k_{\rm max}=0.25~h~{\rm Mpc^{-1}}$, compared to that derived from $P_{\rm pre}$ alone, namely, the constraints on ($Ω_m$, $H_0$, $σ_8$) are tightened by $\sim41 \%-55\%$, and the uncertainties of the derived BAO and RSD parameters ($α_{\perp}$, $α_{||}$, $fσ_8$) shrink by $\sim 28\%-54\%$, respectively. This highlights the complementarity among $P_{\rm pre}$, $P_{\rm post}$ and $P_{\rm cross}$, which demonstrates the efficiency and practicability of a joint $P_{\rm pre}$, $P_{\rm post}$ and $P_{\rm cross}$ analysis for cosmological implications.

astro-ph.CO↗

arXiv2024

The DESI One-Percent survey: exploring the Halo Occupation Distribution of Emission Line Galaxies with AbacusSummit simulations

The One-Percent survey of the Dark Energy Spectroscopic Instrument collected ~ 270k emission line galaxies (ELGs) at 0.8 < z < 1.6. The high completeness of the sample allowed the clustering to be measured down to scales never probed before, 0.04 Mpc/h in rp for the projected 2-point correlation function (2PCF) and 0.17 Mpc/h in galaxy pair separation s for the 2PCF monopole and quadrupole. The most striking feature of the measurements is a strong signal at the smallest scales, below 0.2 Mpc/h in rp and 1 Mpc/h in s. We analyze these data in the halo occupation distribution framework. We consider different distributions for central galaxies, a standard power law for satellites with no condition on the presence of a central galaxy and explore several extensions of these models. For all considered models, the mean halo mass of the sample is found to be log10 ~ 11.9. We obtain a satellite mean occupation function which agrees with physically motivated ELG models only if we introduce central-satellite conformity, meaning that the satellite occupation is conditioned by the presence of central galaxies of the same type. To achieve in addition a good modeling of the clustering between 0.1 and 1 Mpc/h in rp, we allow for ELG positioning outside of the halo virial radius and find 0.5% of ELGs residing in the outskirts of halos. Furthermore, the satellite velocity dispersion inside halos is found to be ~ 30% larger than that of the halo dark matter particles. These are the main findings of our work. We investigate assembly bias as a function of halo concentration, local density or local density anisotropies and observe no significant change in our results. We split the data sample in two redshift bins and report no significant evolution with redshift. Lastly, changing the cosmology in the modeling impacts only slightly our results.

astro-ph.CO↗