Yuhan Yang

Yuhan Yang, Ruipu Li, Alexander Rodríguez

Scientific simulators are increasingly being integrated into LLM-driven systems for high-stakes simulation-driven decision-making. However, existing frameworks primarily use LLMs to generate, calibrate, or execute simulators, treating them as black-box interfaces rather than as structured mechanistic systems that can be reasoned about. As a result, current approaches lack the ability to identify, represent, and reason about the assumptions and mechanisms underlying simulator behavior, limiting transparency, auditability, and decision justification. We introduce MechSim, a mechanism-grounded neuro-symbolic reasoning framework for executable scientific simulators. Unlike prior neuro-symbolic approaches that primarily reason over static symbolic structures, MechSim enables LLM agents to reason about the mechanisms, assumptions, and execution behavior of scientific simulators. Our framework represents simulators through a shared structured schema capturing assumptions, variables, mechanism dependencies, and execution traces. On top of this representation, LLM agents operate as constrained reasoning engines that generate structured, evidence-grounded explanations linking simulator outcomes to their underlying mechanisms. We evaluate our approach across multiple high-stakes domains and show that it improves mechanism-level explanation quality, simulator analysis, and downstream decision-making reliability.

Yuhan Yang, Yong Zhou, Youlong Wu et al.

Federated learning (FL), as a disruptive machine learning paradigm, enables the collaborative training of a global model over decentralized local datasets without sharing them. It spans a wide scope of applications from Internet-of-Things (IoT) to biomedical engineering and drug discovery. To support low-latency and high-privacy FL over wireless networks, in this paper, we propose a reconfigurable intelligent surface (RIS) empowered over-the-air FL system to alleviate the dilemma between learning accuracy and privacy. This is achieved by simultaneously exploiting the channel propagation reconfigurability with RIS for boosting the receive signal power, as well as waveform superposition property with over-the-air computation (AirComp) for fast model aggregation. By considering a practical scenario where high-dimensional local model updates are transmitted across multiple communication blocks, we characterize the convergence behaviors of the differentially private federated optimization algorithm. We further formulate a system optimization problem to optimize the learning accuracy while satisfying privacy and power constraints via the joint design of transmit power, artificial noise, and phase shifts at RIS, for which a two-step alternating minimization framework is developed. Simulation results validate our systematic, theoretical, and algorithmic achievements and demonstrate that RIS can achieve a better trade-off between privacy and accuracy for over-the-air FL systems.