arXiv subjects

Jiayi Zhou

Publications and source records attributed to Jiayi Zhou.

At least 55 records · Page 3Linked to original sources

arXiv2025

IF-VidCap: Can Video Caption Models Follow Instructions?

Although Multimodal Large Language Models (MLLMs) have demonstrated proficiency in video captioning, practical applications require captions that follow specific user instructions rather than generating exhaustive, unconstrained descriptions. Current benchmarks, however, primarily assess descriptive comprehensiveness while largely overlooking instruction-following capabilities. To address this gap, we introduce IF-VidCap, a new benchmark for evaluating controllable video captioning, which contains 1,400 high-quality samples. Distinct from existing video captioning or general instruction-following benchmarks, IF-VidCap incorporates a systematic framework that assesses captions on two dimensions: format correctness and content correctness. Our comprehensive evaluation of over 20 prominent models reveals a nuanced landscape: despite the continued dominance of proprietary models, the performance gap is closing, with top-tier open-source solutions now achieving near-parity. Furthermore, we find that models specialized for dense captioning underperform general-purpose MLLMs on complex instructions, indicating that future work should simultaneously advance both descriptive richness and instruction-following fidelity.

cs.CV↗

arXiv2025

Language Models Resist Alignment: Evidence From Data Compression

Large language models (LLMs) may exhibit unintended or undesirable behaviors. Recent works have concentrated on aligning LLMs to mitigate harmful outputs. Despite these efforts, some anomalies indicate that even a well-conducted alignment process can be easily circumvented, whether intentionally or accidentally. Does alignment fine-tuning yield have robust effects on models, or are its impacts merely superficial? In this work, we make the first exploration of this phenomenon from both theoretical and empirical perspectives. Empirically, we demonstrate the $\mathbf{elasticity}$ of post-alignment models, i.e., the tendency to revert to the behavior distribution formed during the pre-training phase upon further fine-tuning. Leveraging compression theory, we formally deduce that fine-tuning disproportionately undermines alignment relative to pre-training, potentially by orders of magnitude. We validate the presence of elasticity through experiments on models of varying types and scales. Specifically, we find that model performance declines rapidly before reverting to the pre-training distribution, after which the rate of decline drops significantly. Furthermore, we further reveal that elasticity positively correlates with the increased model size and the expansion of pre-training data. Our findings underscore the need to address the inherent elasticity of LLMs to mitigate their resistance to alignment. The model weight and code are available at pku-lm-resist-alignment.github.io.

cs.CL↗

arXiv2025

PKU-SafeRLHF: Towards Multi-Level Safety Alignment for LLMs with Human Preference

In this study, we introduce the safety human preference dataset, PKU-SafeRLHF, designed to promote research on safety alignment in large language models (LLMs). As a sibling project to SafeRLHF and BeaverTails, we separate annotations of helpfulness and harmlessness for question-answering pairs, providing distinct perspectives on these coupled attributes. Overall, we provide 44.6k refined prompts and 265k question-answer pairs with safety meta-labels for 19 harm categories and three severity levels ranging from minor to severe, with answers generated by Llama-family models. Based on this, we collected 166.8k preference data, including dual-preference (helpfulness and harmlessness decoupled) and single-preference data (trade-off the helpfulness and harmlessness from scratch), respectively. Using the large-scale annotation data, we further train severity-sensitive moderation for the risk control of LLMs and safety-centric RLHF algorithms for the safety alignment of LLMs. We believe this dataset will be a valuable resource for the community, aiding in the safe deployment of LLMs. Data is available at https://huggingface.co/datasets/PKU-Alignment/PKU-SafeRLHF.

cs.AI↗

arXiv2025

Point Cloud Neural Operator for Parametric PDEs on Complex and Variable Geometries

Surrogate models are critical for accelerating computationally expensive simulations in science and engineering, particularly for solving parametric partial differential equations (PDEs). Developing practical surrogate models poses significant challenges, particularly in handling geometrically complex and variable domains, which are often discretized as point clouds. In this work, we systematically investigate the formulation of neural operators -- maps between infinite-dimensional function spaces -- on point clouds to better handle complex and variable geometries while mitigating discretization effects. We introduce the Point Cloud Neural Operator (PCNO), designed to efficiently approximate solution maps of parametric PDEs on such domains. We evaluate the performance of PCNO on a range of pedagogical PDE problems, focusing on aspects such as boundary layers, adaptively meshed point clouds, and variable domains with topological variations. Its practicality is further demonstrated through three-dimensional applications, such as predicting pressure loads on various vehicle types and simulating the inflation process of intricate parachute structures.

math.NA↗

arXiv2025

Reward Generalization in RLHF: A Topological Perspective

Existing alignment methods share a common topology of information flow, where reward information is collected from humans, modeled with preference learning, and used to tune language models. However, this shared topology has not been systematically characterized, nor have its alternatives been thoroughly explored, leaving the problems of low data efficiency and unreliable generalization unaddressed. As a solution, we introduce a theory of reward generalization in reinforcement learning from human feedback (RLHF), focusing on the topology of information flow at both macro and micro levels. At the macro level, we portray the RLHF information flow as an autoencoding process over behavior distributions, formalizing the RLHF objective of distributional consistency between human preference and model behavior. At the micro level, we present induced Bayesian networks to model the impact of dataset topologies on reward generalization. Combining analysis on both levels, we propose reward modeling from tree-structured preference information. It is shown to reduce reward uncertainty by up to $Θ(\log n/\log\log n)$ times compared to baselines, where $n$ is the dataset size. Validation on three NLP tasks shows that it achieves an average win rate of 65% against baselines, thus improving reward generalization for free via topology design, while reducing the amount of data requiring annotation.

cs.LG↗

arXiv2025

Generative RLHF-V: Learning Principles from Multi-modal Human Preference

Training multi-modal large language models (MLLMs) that align with human intentions is a long-term challenge. Traditional score-only reward models for alignment suffer from low accuracy, weak generalization, and poor interpretability, blocking the progress of alignment methods, e.g., reinforcement learning from human feedback (RLHF). Generative reward models (GRMs) leverage MLLMs' intrinsic reasoning capabilities to discriminate pair-wise responses, but their pair-wise paradigm makes it hard to generalize to learnable rewards. We introduce Generative RLHF-V, a novel alignment framework that integrates GRMs with multi-modal RLHF. We propose a two-stage pipeline: $\textbf{multi-modal generative reward modeling from RL}$, where RL guides GRMs to actively capture human intention, then predict the correct pair-wise scores; and $\textbf{RL optimization from grouped comparison}$, which enhances multi-modal RL scoring precision by grouped responses comparison. Experimental results demonstrate that, besides out-of-distribution generalization of RM discrimination, our framework improves 4 MLLMs' performance across 7 benchmarks by $18.1\%$, while the baseline RLHF is only $5.3\%$. We further validate that Generative RLHF-V achieves a near-linear improvement with an increasing number of candidate responses. Our code and models can be found at https://generative-rlhf-v.github.io.

cs.AI↗

arXiv2025

Mitigating Deceptive Alignment via Self-Monitoring

Modern large language models rely on chain-of-thought (CoT) reasoning to achieve impressive performance, yet the same mechanism can amplify deceptive alignment, situations in which a model appears aligned while covertly pursuing misaligned goals. Existing safety pipelines treat deception as a black-box output to be filtered post-hoc, leaving the model free to scheme during its internal reasoning. We ask: Can deception be intercepted while the model is thinking? We answer this question, the first framework that embeds a Self-Monitor inside the CoT process itself, named CoT Monitor+. During generation, the model produces (i) ordinary reasoning steps and (ii) an internal self-evaluation signal trained to flag and suppress misaligned strategies. The signal is used as an auxiliary reward in reinforcement learning, creating a feedback loop that rewards honest reasoning and discourages hidden goals. To study deceptive alignment systematically, we introduce DeceptionBench, a five-category benchmark that probes covert alignment-faking, sycophancy, etc. We evaluate various LLMs and show that unrestricted CoT roughly aggravates the deceptive tendency. In contrast, CoT Monitor+ cuts deceptive behaviors by 43.8% on average while preserving task accuracy. Further, when the self-monitor signal replaces an external weak judge in RL fine-tuning, models exhibit substantially fewer obfuscated thoughts and retain transparency. Our project website can be found at cot-monitor-plus.github.io

cs.AI↗

arXiv2025

Safe RLHF-V: Safe Reinforcement Learning from Multi-modal Human Feedback

Multimodal large language models (MLLMs) are essential for building general-purpose AI assistants; however, they pose increasing safety risks. How can we ensure safety alignment of MLLMs to prevent undesired behaviors? Going further, it is critical to explore how to fine-tune MLLMs to preserve capabilities while meeting safety constraints. Fundamentally, this challenge can be formulated as a min-max optimization problem. However, existing datasets have not yet disentangled single preference signals into explicit safety constraints, hindering systematic investigation in this direction. Moreover, it remains an open question whether such constraints can be effectively incorporated into the optimization process for multi-modal models. In this work, we present the first exploration of the Safe RLHF-V -- the first multimodal safety alignment framework. The framework consists of: $\mathbf{(I)}$ BeaverTails-V, the first open-source dataset featuring dual preference annotations for helpfulness and safety, supplemented with multi-level safety labels (minor, moderate, severe); $\mathbf{(II)}$ Beaver-Guard-V, a multi-level guardrail system to proactively defend against unsafe queries and adversarial attacks. Applying the guard model over five rounds of filtering and regeneration significantly enhances the precursor model's overall safety by an average of 40.9%. $\mathbf{(III)}$ Based on dual preference, we initiate the first exploration of multi-modal safety alignment within a constrained optimization. Experimental results demonstrate that Safe RLHF effectively improves both model helpfulness and safety. Specifically, Safe RLHF-V enhances model safety by 34.2% and helpfulness by 34.3%.

cs.LG↗

arXiv2025

Measurement of two-neutrino double electron capture half-life of $^{124}$Xe with PandaX-4T

Detailed studies of two-neutrino double electron capture (2$ν$DEC) is a crucial step towards searching for the neutrino-less mode to explore the Majorana nature of neutrinos. We have measured precisely the half-life of the 2$ν$DEC process in $^{124}$Xe, utilizing a total exposure of 1.73 tonne$\cdot$year from the commissioning run and the first science run of the PandaX-4T experiment. A time-dependent background model in the $\mathcal{O}$(10 keV) energy is constructed for the first time in PandaX-4T data. With an unbinned maximum likelihood fit, we determine the half-life of the 2$ν$DEC process to be $(1.03\pm0.15_{\rm stat}\pm0.08_{\rm sys})\times 10^{22}$$\,$yr. Furthermore, we have evaluated the branching ratio for both electrons captured from the $K$ shell ($KK$) to be $(65\pm5)\%$, which aligns with the $^{124}$Xe nuclear model calculations within 1.8$\,$$σ$.

nucl-ex↗

arXiv2025

Search for Solar Boosted Dark Matter Particles at the PandaX-4T Experiment

We present a novel constraint on light dark matter utilizing $1.54$ tonne$\cdot$year of data acquired from the PandaX-4T dual-phase xenon time projection chamber. This constraint is derived through detecting electronic recoil signals resulting from the interaction with solar-enhanced dark matter flux. Low-mass dark matter particles, lighter than a few MeV/$c^2$, can scatter with the thermal electrons in the Sun. Consequently, with higher kinetic energy, the boosted dark matter component becomes detectable via contact scattering with xenon electrons, resulting in a few keV energy deposition that exceeds the threshold of PandaX-4T. We calculate the expected recoil energy in PandaX-4T considering the Sun's acceleration and the detection capabilities of the xenon detector. The first experimental search results using the xenon detector yield the most stringent cross-section of $3.51 \times 10^{-39}~\mathrm{cm}^2$ at $0.08~\mathrm{MeV}$/$c^2$ for a solar boosted dark matter mass ranging from $0.02$ to $10~ \mathrm{MeV}$/$c^2$, achieving a 23 fold improvement compared with earlier experimental studies.

hep-ex↗

arXiv2025

Ultra-Low Complexity On-Orbit Compression for Remote Sensing Imagery via Block Modulated Imaging

The growing field of remote sensing faces a challenge: the ever-increasing size and volume of imagery data are exceeding the storage and transmission capabilities of satellite platforms. Efficient compression of remote sensing imagery is a critical solution to alleviate these burdens on satellites. However, existing compression methods are often too computationally expensive for satellites. With the continued advancement of compressed sensing theory, single-pixel imaging emerges as a powerful tool that brings new possibilities for on-orbit image compression. However, it still suffers from prolonged imaging times and the inability to perform high-resolution imaging, hindering its practical application. This paper advances the study of compressed sensing in remote sensing image compression, proposing Block Modulated Imaging (BMI). By requiring only a single exposure, BMI significantly enhances imaging acquisition speeds. Additionally, BMI obviates the need for digital micromirror devices and surpasses limitations in image resolution. Furthermore, we propose a novel decoding network specifically designed to reconstruct images compressed under the BMI framework. Leveraging the gated 3D convolutions and promoting efficient information flow across stages through a Two-Way Cross-Attention module, our decoding network exhibits demonstrably superior reconstruction performance. Extensive experiments conducted on multiple renowned remote sensing datasets unequivocally demonstrate the efficacy of our proposed method. To further validate its practical applicability, we developed and tested a prototype of the BMI-based camera, which has shown promising potential for on-orbit image compression. The code is available at https://github.com/Johnathan218/BMNet.

eess.IV↗

arXiv2025

AI Alignment: A Comprehensive Survey

AI alignment aims to make AI systems behave in line with human intentions and values. As AI systems grow more capable, so do risks from misalignment. To provide a comprehensive and up-to-date overview of the alignment field, in this survey, we delve into the core concepts, methodology, and practice of alignment. First, we identify four principles as the key objectives of AI alignment: Robustness, Interpretability, Controllability, and Ethicality (RICE). Guided by these four principles, we outline the landscape of current alignment research and decompose them into two key components: forward alignment and backward alignment. The former aims to make AI systems aligned via alignment training, while the latter aims to gain evidence about the systems' alignment and govern them appropriately to avoid exacerbating misalignment risks. On forward alignment, we discuss techniques for learning from feedback and learning under distribution shift. On backward alignment, we discuss assurance techniques and governance practices. We also release and continually update the website (www.alignmentsurvey.com) which features tutorials, collections of papers, blog posts, and other resources.

cs.AI↗

arXiv2025

Search for MeV-Scale Axionlike Particles and Dark Photons with PandaX-4T

Axionlike particles (ALPs) and dark photons (DPs) are viable dark matter particle candidates. We have searched for possible ALP/DP signals in the PandaX-4T liquid xenon detector using 440 kg$\cdot$yr of data. A binned likelihood fit is constructed to search for possible mono-energetic peaks induced by the absorption processes between ALPs/DPs and atomic electrons of xenon. A detailed temporal model of decays associated with xenon isotopes is introduced to constrain the number of background events. No signal excess over background expectations is observed, and we have established the most stringent exclusion limits for most ALP/DP masses across the range of 150 keV/$c^2$ to 1 MeV/$c^2$. The improvement is particularly significant within the mass range of \text{150-400}~keV/$c^2$, with the average factor of 3.5 compared to previous results.

hep-ex↗

arXiv2025

Position reconstruction and surface background model for the PandaX-4T detector

We report the position reconstruction methods and surface background model for the PandaX-4T dark matter direct search experiment. This work develops two position reconstruction algorithms: template matching (TM) method and photon acceptance function (PAF) method. Both methods determine the horizontal position of events based on the light pattern of secondary scintillation collected by the light sensors. After a comprehensive evaluation of resolution, uniformity, and robustness, the PAF method was selected for position reconstruction, while the TM method was employed for verification. The PAF method achieves a bulk event resolution of 1.0 mm and a surface event resolution of 4.4 mm for a typical $S2$ signal with a bottom charge of 1500 PE (about 14 keV). The uniformity is around 20\%. Robustness studies reveal average deviations of 5.1 mm and 8.8 mm for the commissioning run (Run0) and the first science run (Run1), respectively, due to the deactivation of certain PMTs. A data-driven surface background model is developed based on the PAF method. The surface background is estimated to be $0.09 \pm 0.06$ events for Run0 (0.54 tonne$\cdot$year) and $0.17 \pm 0.11$ events for Run1 (1.00 tonne$\cdot$year).

physics.ins-det↗

arXiv2025

Dark Matter Search Results from 1.54 Tonne$\cdot$Year Exposure of PandaX-4T

In this letter, we report the dark matter search results from the commissioning run and the first science run of the PandaX-4T experiment. A blind analysis is carried out on the entire data set. The data processing is improved compared to previous work, unifying the low-level signal reconstruction in a wide energy range up to 120 keV. With a total exposure of 1.54 tonne$\cdot$year, no significant excess of nuclear recoil events is found. The lowest 90% confidence level exclusion on the spin-independent cross section is $1.6 \times 10^{-47}\,\mathrm{cm}^2$ at a dark matter mass of 40 GeV$/c^2$. Our results represent the most stringent constraint for a dark matter mass above 100 GeV$/c^2$.

hep-ex↗

arXiv2025

Exploring New Physics with PandaX-4T Low Energy Electronic Recoil Data

New particles beyond the standard model of particle physics, such as axions, can be effectively searched through their interactions with electrons. We use the large liquid xenon detector PandaX-4T to search for novel electronic recoil signals induced by solar axions, neutrinos with anomalous magnetic moment, axionlike particles, dark photons, and light fermionic dark matter. A detailed background model is established using the latest datasets with 1.54 $\rm ton \cdot yr$ exposure. No significant excess above the background has been observed, and we have obtained competitive constraints for axion couplings, neutrino magnetic moment, and fermionic dark matter interactions.

hep-ex↗

arXiv2024

Align Anything: Training All-Modality Models to Follow Instructions with Language Feedback

Reinforcement learning from human feedback (RLHF) has proven effective in enhancing the instruction-following capabilities of large language models; however, it remains underexplored in the cross-modality domain. As the number of modalities increases, aligning all-modality models with human intentions -- such as instruction following -- becomes a pressing challenge. In this work, we make the first attempt to fine-tune all-modality models (i.e. input and output with any modality, also named any-to-any models) using human preference data across all modalities (including text, image, audio, and video), ensuring its behavior aligns with human intentions. This endeavor presents several challenges. First, there is no large-scale all-modality human preference data in existing open-source resources, as most datasets are limited to specific modalities, predominantly text and image. Secondly, the effectiveness of binary preferences in RLHF for post-training alignment in complex all-modality scenarios remains an unexplored area. Finally, there is a lack of a systematic framework to evaluate the capabilities of all-modality models, particularly regarding modality selection and synergy. To address these challenges, we propose the align-anything framework, which includes meticulously annotated 200k all-modality human preference data. Then, we introduce an alignment method that learns from unified language feedback, effectively capturing complex modality-specific human preferences and enhancing the model's instruction-following capabilities. Furthermore, to assess performance improvements in all-modality models after post-training alignment, we construct a challenging all-modality capability evaluation framework -- eval-anything. All data, models, and code frameworks have been open-sourced for the community. For more details, please refer to https://github.com/PKU-Alignment/align-anything.

cs.AI↗

arXiv2024

Searching for Neutrinoless Double-Beta Decay of $^{136}$Xe with PandaX-4T

We report the search for neutrinoless double-beta decay of $^{136}$Xe from the PandaX-4T experiment with a 3.7-tonne natural xenon target. The data reconstruction and the background modeling are optimized in the MeV energy region. A blind analysis is performed with data from the commissioning run and the first science run. No significant excess of signal over the background is observed. A lower limit on the half-life of $^{136}$Xe neutrinoless double-beta decay is established to be $2.1 \times 10^{24}$~yr at the 90\% confidence level, with a $^{136}$Xe exposure of 44.6~kg$\cdot$year. Our result represents the most stringent constraint from a natural xenon detector to date.

nucl-ex↗