arXiv subjects

Heyi Zhang

Publications and source records attributed to Heyi Zhang.

6 recordsLinked to original sources

arXiv2026

ALMA High-resolution Observation of the HH46/47 Outflow/disk/envelope System

We present $0.1^{\prime\prime}$ ($\sim 50$ au) resolution Atacama Large Millimeter/submillimeter Array (ALMA) observations of the HH 46/47 molecular outflow and its envelope-disk system. The 1.3 mm continuum emission reveals a compact central source surrounded by a circumbinary disk with substructures. The companion, identified in optical and infrared observations, is not detected in the millimeter continuum but coincides with a local intensity minimum. Two spur-like features extending from the primary source toward the companion are identified and are likely induced by gravitational perturbations from the companion. The envelope-disk system is traced by C$^{18}$O, SO, H$_2$CO, and CH$_3$OH. C$^{18}$O primarily traces the extended envelope, while SO probes the inner envelope, and H$_2$CO and CH$_3$OH trace compact, faster-rotating structures near the centrifugal barrier. The observations are well reproduced by a rotating-infalling envelope transitioning to an inner disk at a radius of $\sim 30$ au around a $0.3~M_\odot$ protostar. The $^{12}$CO emission, together with JWST NIRCam images, reveals multiple shell structures in the outflow. Using C$^{18}$O and $^{13}$CO to correct for optical depth, we derive the spatial distributions of outflow mass, momentum, and kinetic energy, as well as their corresponding rates. A model-independent analysis of a well-defined redshifted shell yields its three-dimensional velocity field, showing that the shell expands radially rather than flowing along its surface. Although a transverse velocity gradient is detected, interpreting it as rotation implies an unphysically large magnetic lever arm, disfavoring a direct disk-wind origin. Instead, the shell kinematics support an entrainment scenario.

astro-ph.GA↗

arXiv2025

Auditing Data Membership in Reinforcement Learning With Verifiable Rewards

Reinforcement Learning with Verifiable Rewards (RLVR) has become a core training stage in recent large language models (LLMs). Its reliance on non-public, high-value prompt sets raises concerns about unauthorized data use, creating a need for exposure auditing. A natural tool is membership inference attacks (MIAs), but existing methods detect fitting to a fixed target string. This does not apply to RLVR, which generates responses from the model itself and reinforces successful ones, thus hindering the auditing of data exposure. We show that it remains detectable: RLVR reshapes the model's response distribution on training prompts, producing behavioral traces that can be surfaced through targeted auditing. We propose Divergence-in-Behavior Auditing (DIBA), a white-box query-level auditing framework for RLVR. DIBA compares a fine-tuned model against its pre-RLVR checkpoint along two axes: reward-side evidence capturing changes in verifiable task success, and policy-side evidence capturing prompt-conditioned behavioral drift. By aggregating over multiple stochastic rollouts, DIBA produces a stable query-level auditing signal. Under a white-box setting, DIBA consistently outperforms strong transferred likelihood-based baselines, including calibrated and self-generated variants, achieving around 0.8 AUC and an order-of-magnitude stronger TPR@0.1%FPR. We further show that RLVR auditing is stronger when training leaves non-trivial prompt-specific traces and weaker when the base model already performs well on the prompt. Under a practical grey-box setting, transfer is often robust across model sizes under the same RLVR algorithm, but more varied across algorithms, and can remain useful under distribution shift with carefully chosen shadow data.

cs.CR↗

arXiv2025

SoK: Benchmarking Poisoning Attacks and Defenses in Federated Learning

Federated learning (FL) enables collaborative model training while preserving data privacy, but its decentralized nature exposes it to client-side data poisoning attacks (DPAs) and model poisoning attacks (MPAs) that degrade global model performance. While numerous proposed defenses claim substantial effectiveness, their evaluation is typically done in isolation with limited attack strategies, raising concerns about their validity. Additionally, existing studies overlook the mutual effectiveness of defenses against both DPAs and MPAs, causing fragmentation in this field. This paper aims to provide a unified benchmark and analysis of defenses against DPAs and MPAs, clarifying the distinction between these two similar but slightly distinct domains. We present a systematic taxonomy of poisoning attacks and defense strategies, outlining their design, strengths, and limitations. Then, a unified comparative evaluation across FL algorithms and data heterogeneity is conducted to validate their individual and mutual effectiveness and derive key insights for design principles and future research. Along with the analysis, we frame our work to a unified benchmark, FLPoison, with high modularity and scalability to evaluate 15 representative poisoning attacks and 17 defense strategies, facilitating future research in this domain. Code is available at https://github.com/vio1etus/FLPoison.

cs.CR↗

arXiv2024

Qibo: A Large Language Model for Traditional Chinese Medicine

Large Language Models (LLMs) has made significant progress in a number of professional fields, including medicine, law, and finance. However, in traditional Chinese medicine (TCM), there are challenges such as the essential differences between theory and modern medicine, the lack of specialized corpus resources, and the fact that relying only on supervised fine-tuning may lead to overconfident predictions. To address these challenges, we propose a two-stage training approach that combines continuous pre-training and supervised fine-tuning. A notable contribution of our study is the processing of a 2GB corpus dedicated to TCM, constructing pre-training and instruction fine-tuning datasets for TCM, respectively. In addition, we have developed Qibo-Benchmark, a tool that evaluates the performance of LLM in the TCM on multiple dimensions, including subjective, objective, and three TCM NLP tasks. The medical LLM trained with our pipeline, named $\textbf{Qibo}$, exhibits significant performance boosts. Compared to the baselines, the average subjective win rate is 63%, the average objective accuracy improved by 23% to 58%, and the Rouge-L scores for the three TCM NLP tasks are 0.72, 0.61, and 0.55. Finally, we propose a pipline to apply Qibo to TCM consultation and demonstrate the model performance through the case study.

cs.CL↗

arXiv2015

New factorization algorithm based on a continuous representation of truncated Gauss sums

In this paper, we will describe a new factorization algorithm based on the continuous representation of Gauss sums, generalizable to orders j>2. Such an algorithm allows one, for the first time, to find all the factors of a number N in a single run without precalculating the ratio N/l, where l are all the possible trial factors. Continuous truncated exponential sums turn out to be a powerful tool for distinguishing factors from non-factors (we also suggest, with regard to this topic, to read an interesting paper by S. Woelk et al. also published in this issue [Woelk, Feiler, Schleich, J. Mod. Opt. in press]) and factorizing different numbers at the same time. We will also describe two possible M-path optical interferometers, which can be used to experimentally realize this algorithm: a liquid crystal grating and a generalized symmetric Michelson interferometer.

quant-ph↗

arXiv2015

Factoring numbers with a single interferogram

We construct an analog computer based on light interference to encode the hyperbolic function f(ζ) = 1/ζ into a sequence of skewed curlicue functions. The resulting interferogram when scaled appropriately allows us to find the prime number decompositions of integers. We implement this idea exploiting polychromatic optical interference in a multipath interferometer and factor seven-digit numbers. We give an estimate for the largest number that can be factored by this scheme.

quant-ph↗