arXiv subjects

Zeqi Zhang

Publications and source records attributed to Zeqi Zhang.

8 recordsLinked to original sources

arXiv2025

PersonaX: A Recommendation Agent Oriented User Modeling Framework for Long Behavior Sequence

User profile embedded in the prompt template of personalized recommendation agents play a crucial role in shaping their decision-making process. High-quality user profiles are essential for aligning agent behavior with real user interests. Typically, these profiles are constructed by leveraging LLMs for user profile modeling (LLM-UM). However, this process faces several challenges: (1) LLMs struggle with long user behaviors due to context length limitations and performance degradation. (2) Existing methods often extract only partial segments from full historical behavior sequence, inevitably discarding diverse user interests embedded in the omitted content, leading to incomplete modeling and suboptimal profiling. (3) User profiling is often tightly coupled with the inference context, requiring online processing, which introduces significant latency overhead. In this paper, we propose PersonaX, an agent-agnostic LLM-UM framework to address these challenges. It augments downstream recommendation agents to achieve better recommendation performance and inference efficiency. PersonaX (a) segments complete historical behaviors into clustered groups, (b) selects multiple sub behavior sequences (SBS) with a balance of prototypicality and diversity to form a high quality core set, (c) performs offline multi-persona profiling to capture diverse user interests and generate fine grained, cached textual personas, and (d) decouples user profiling from online inference, enabling profile retrieval instead of real time generation. Extensive experiments demonstrate its effectiveness: using only 30 to 50% of behavioral data (sequence length 480), PersonaX enhances AgentCF by 3 to 11% and Agent4Rec by 10 to 50%. As a scalable and model-agnostic LLM-UM solution, PersonaX sets a new benchmark in scalable user modeling.

cs.IR↗

arXiv2023

Spacecraft Charging of the Morazán MRZ-SAT Satellite in Low Earth Orbit: Initial Results on the Influence of Energetic Electron Anisotropy on Differential Charging

The advent of the modular CubeSat satellite architecture has heralded a revolution in satellite missions, drastically lowering the technical and financial barriers to space. Surface charging resulting from energetic electron poses a direct risk to satellites in space, causing electric arcing and breakdowns. This risk is exacerbated for small technology demonstration CubeSats that are less resilient than larger satellites. An upcoming CubeSat launch is the first CubeSat project originating from Honduras, the Morazán satellite (MRZ-SAT), due to launch in 2024. This will carry earth observational payloads to detect natural disasters. This study conducts simulations using the Electro-Magnetic Spacecraft Environment Simulator code to study absolute and differential charging of the MRZ-SAT cube-sat in Low Earth Orbit (LEO). The MRZ-SAT hosts four antennas, an architecture which lends itself well to studying and understanding differential charging in LEO. The MRZ-SAT was first simulated in a typical benign ionospheric plasma environment. Here the antenna located in the ambient plasma wake displayed the maximum charging up to --0.9 V, 0.24 V biased to the main cube. An energetic electron population was then included and the wake antenna subsequently charged to greater values of --2.73 V, now 1.56 V biased to the main cube. The anisotropy of the energetic electrons was then varied, and this differential charging trend appeared exacerbated with anisotropies of 0.5 to 0.05 inducing absolute wake antenna voltages up to --4.5 V and differential voltage biases 50 and 100 \% greater than when an isotropic population was considered. This study highlights the importance of electron anisotropy in LEO to surface charging and identifies this property in the energetic electron distribution functions as inducing potentially greater risks to satellites of electrical arcing and breakdown.

physics.space-ph↗

arXiv2023

Cassini's floating potential in Titan's ionosphere: 3-D Particle-In-Cell Simulations

Accurate determination of Cassini's spacecraft potential in Titan's ionosphere is important for interpreting measurements by its low energy plasma instruments. Estimates of the floating potential varied significantly, however, between the various different plasma instruments. In this study we utilize 3-D particle-in-cell simulations to understand the key features of Cassini's plasma interaction in Titan's ionosphere. The spacecraft is observed to charge to negative potentials for all scenarios considered, and close agreement is found between the current onto the simulated Langmuir Probe and that observed in Titan's ionosphere. These simulations are therefore shown to provide a viable technique for modeling spacecraft interacting with Titan's dusty ionosphere.

physics.space-ph↗

arXiv2023

Simulating secondary electron and ion emission from the Cassini spacecraft in Saturn's ionosphere

The Cassini spacecraft's Grand Finale flybys through Saturn's ionosphere provided unprecedented insight into the composition and dynamics of the gas giant's upper atmosphere and a novel and complex spacecraft-plasma interaction. In this article, we further study Cassini's interaction with Saturn's ionosphere using three dimensional Particle-in-Cell simulations. We focus on understanding how electrons and ions, emitted from spacecraft surfaces due to the high-velocity impact of atmospheric water molecules, could have affected the spacecraft potential and low-energy plasma measurements. The simulations show emitted electrons extend upstream along the magnetic field and, for sufficiently high emission rates, charge the spacecraft to positive potentials. The lack of accurate emission rates and characteristics, however, makes differentiation between the prominence of secondary electron emission and ionospheric charged dust populations, which induce similar charging effects, difficult for Cassini. These results provide further context for Cassini's final measurements and highlight the need for future laboratory studies to support high-velocity flyby missions through planetary and cometary ionospheres.

physics.space-ph↗

arXiv2023

Robust Deep Learning Models Against Semantic-Preserving Adversarial Attack

Deep learning models can be fooled by small $l_p$-norm adversarial perturbations and natural perturbations in terms of attributes. Although the robustness against each perturbation has been explored, it remains a challenge to address the robustness against joint perturbations effectively. In this paper, we study the robustness of deep learning models against joint perturbations by proposing a novel attack mechanism named Semantic-Preserving Adversarial (SPA) attack, which can then be used to enhance adversarial training. Specifically, we introduce an attribute manipulator to generate natural and human-comprehensible perturbations and a noise generator to generate diverse adversarial noises. Based on such combined noises, we optimize both the attribute value and the diversity variable to generate jointly-perturbed samples. For robust training, we adversarially train the deep learning model against the generated joint perturbations. Empirical results on four benchmarks show that the SPA attack causes a larger performance decline with small $l_{\infty}$ norm-ball constraints compared to existing approaches. Furthermore, our SPA-enhanced training outperforms existing defense methods against such joint perturbations.

cs.LG↗

arXiv2022

Mitigating Unfairness via Evolutionary Multi-objective Ensemble Learning

In the literature of mitigating unfairness in machine learning, many fairness measures are designed to evaluate predictions of learning models and also utilised to guide the training of fair models. It has been theoretically and empirically shown that there exist conflicts and inconsistencies among accuracy and multiple fairness measures. Optimising one or several fairness measures may sacrifice or deteriorate other measures. Two key questions should be considered, how to simultaneously optimise accuracy and multiple fairness measures, and how to optimise all the considered fairness measures more effectively. In this paper, we view the mitigating unfairness problem as a multi-objective learning problem considering the conflicts among fairness measures. A multi-objective evolutionary learning framework is used to simultaneously optimise several metrics (including accuracy and multiple fairness measures) of machine learning models. Then, ensembles are constructed based on the learning models in order to automatically balance different metrics. Empirical results on eight well-known datasets demonstrate that compared with the state-of-the-art approaches for mitigating unfairness, our proposed algorithm can provide decision-makers with better tradeoffs among accuracy and multiple fairness metrics. Furthermore, the high-quality models generated by the framework can be used to construct an ensemble to automatically achieve a better tradeoff among all the considered fairness metrics than other ensemble methods. Our code is publicly available at https://github.com/qingquan63/FairEMOL

cs.LG↗

arXiv2021

Photodetachment and Test-Particle Simulation Constraints on Negative Ions in Solar System Plasmas

Negative ions have been detected in abundance in recent years by spacecraft across the solar system. These detections were, however, made by instruments not designed for this purpose and, as such, significant uncertainties remain regarding the prevalence of these unexpected plasma components. In this article, the phenomenon of photodetachment is examined and experimentally and theoretically derived cross-sections are used to calculate photodetachment rates for a range of atomic and molecular negative ions subjected to the solar photon spectrum. These rates are applied to negative ions outflowing from Europa, Enceladus, Titan, Dione and Rhea and their trajectories are traced to constrain source production rates and the extent to which negative ions are able to pervade the surrounding space environments. Predictions are also made for further negative ion populations in the outer solar system with Triton used as an illustrative example. This study demonstrates how, at increased heliocentric distances, negative ions can form stable ambient plasma populations and can be exploited by future missions to the outer solar system.

physics.space-ph↗

arXiv2021

Particle-In-Cell Simulations of the Cassini Spacecraft's Interaction with Saturn's Ionosphere during the Grand Finale

A surprising and unexpected phenomenon observed during Cassini's Grand Finale was the spacecraft charging to positive potentials in Saturn's ionosphere. Here, the ionospheric plasma was depleted of free electrons with negatively charged ions and dust accumulating up to over 95 % of the negative charge density. To further understand the spacecraft-plasma interaction, we perform a three dimensional Particle-In-Cell study of a model Cassini spacecraft immersed in plasma representative of Saturn's ionosphere. The simulations reveal complex interaction features such as electron wings and a highly structured wake containing spacecraft-scale vortices. The results show how a large negative ion concentration combined with a large negative to positive ion mass ratio is able to drive the spacecraft to the observed positive potentials. Despite the high electron depletions, the electron properties are found as a significant controlling factor for the spacecraft potential together with the magnetic field orientation which induces a potential gradient directed across Cassini's asymmetric body. This study reveals the global spacecraft interaction experienced by Cassini during the Grand Finale and how this is influenced by the unexpected negative ion and dust populations.

physics.space-ph↗