arXiv subjects

Jinzhi Shen

Publications and source records attributed to Jinzhi Shen.

4 recordsLinked to original sources

arXiv2026

Forward-Invariant Policy Classes for Safe Reinforcement Learning in Multicopter Control

This paper proposes a reinforcement learning (RL) framework for safe gain scheduling based on forwardinvariance-induced action-space design. Under stated nominalmodel and inversion-domain assumptions, rather than enforcing safety through runtime shielding or penalty-based constraints, safety is embedded directly into the policy class. Specifically, we construct a finite library of feedback controllers sharing a common Lyapunov certificate that establishes forward invariance of a prescribed admissible set under arbitrary switching. Consequently, any policy whose actions are restricted to this library, including policies encountered during RL exploration, inherits the same certificate. Policy optimization can therefore focus on closed-loop performance without runtime safety filtering or action projection. The framework is instantiated for quadcopter hover regulation, where a DQN schedules among certified feedback controllers. Nonlinear MuJoCo simulations demonstrate state-dependent gain scheduling and empirically evaluate robustness to wind, model mismatch, sensor noise, and sensing delay. The results illustrate how safety certification can be separated from policy optimization by learning over a forward-invariant policy class.

eess.SY↗

arXiv2024

Review of Interpretable Machine Learning Models for Disease Prognosis

In response to the COVID-19 pandemic, the integration of interpretable machine learning techniques has garnered significant attention, offering transparent and understandable insights crucial for informed clinical decision making. This literature review delves into the applications of interpretable machine learning in predicting the prognosis of respiratory diseases, particularly focusing on COVID-19 and its implications for future research and clinical practice. We reviewed various machine learning models that are not only capable of incorporating existing clinical domain knowledge but also have the learning capability to explore new information from the data. These models and experiences not only aid in managing the current crisis but also hold promise for addressing future disease outbreaks. By harnessing interpretable machine learning, healthcare systems can enhance their preparedness and response capabilities, thereby improving patient outcomes and mitigating the impact of respiratory diseases in the years to come.

cs.LG↗

arXiv2024

Galaxy Triplets Alignment in Large-scale Filaments

Leveraging the datasets of galaxy triplets and large-scale filaments obtained from the Sloan Digital Sky Survey, we scrutinize the alignment of the three sides of the triangles formed by galaxy triplets and the normal vectors of the triplet planes within observed large-scale filaments. Our statistical investigation reveals that the longest and median sides of the galaxy triplets exhibit a robust alignment with the spines of their host large-scale filaments, while the shortest sides show no or only weak alignment with the filaments. Additionally, the normal vectors of triplets tend to be perpendicular to the filaments. The alignment signal diminishes rapidly with the increasing distance from the triplet to the filament spine, and is primarily significant for triplets located within distances shorter than $0.2$~Mpc$/h$, with a confidence level exceeding $20σ$. Moreover, in comparison to compact galaxy triplets, the alignment signal is more conspicuous among the loose triplets. This alignment analysis contributes to the formulation of a framework depicting the clustering and relaxation of galaxies within cosmological large-scale filament regimes, providing deeper insights into the intricate interactions between galaxies and their pivotal role in shaping galaxy groups.

astro-ph.GA↗

arXiv2023

Shape asymmetries and the relation between lopsidedness and radial alignment in simulated galaxies

Galaxies are observed to be lopsided, meaning that they are more massive and more extended along one direction than the opposite. In this work, we provide a statistical analysis of the lopsided morphology of 1780 isolated satellite galaxies generated by TNG50-1 simulation, incorporating the effect of tidal fields from halo centres. The isolated satellites are galaxies without nearby substructures whose mass is over $1\%$ of the satellites within their virial radii. We study the radial alignment (RA) between the major axes of satellites and the radial direction of their halo centres in radial ranges of $0$-$2R_{\rm h}$, $2$-$5R_{\rm h}$ and $5$-$10R_{\rm h}$ with $R_{\rm h}$ being the stellar half mass radius. According to our results, the RA is virtually undetectable in inner and intermediate regions, yet it is significantly evident in outer regions. We also calculate the far-to-near-side semi-axial ratios of the major axes, denoted by $a_-/a_+$, which measures the semi-axial ratios of the major axes in the hemispheres between backwards (far side) and facing (near side) the halo centres. In all the radial bins of the satellites, the numbers of satellites with longer semi-axes on the far side are found to be almost equal to those with longer semi-axes on the near side. Therefore, the tidal fields from halo centres play a minor role in the generation of lopsided satellites. The long semi-major-axes radial alignment (LRA), i.e., an alignment between the long semi-major-axes of satellite galaxies and the radial directions to their halo centres, is further studied. No clear evidence of LRA is found in our sample within the framework of $Λ$CDM Newtonian dynamics. Finally, we briefly discuss the possible origins of the asymmetry of galaxies in TNG50-1.

astro-ph.GA↗