arXiv subjects

Songlin Zhou

Publications and source records attributed to Songlin Zhou.

5 recordsLinked to original sources

arXiv2026

Well-posedness of Filtering Equations in Weighted Sobolev Spaces with Unbounded System Coefficients

Nonlinear filtering problem is one of the core subjects in modern control theory. In this paper, we will study the well-posedness of the three fundamental evolution equations arising in continuous-time nonlinear filtering--the robust Duncan-Mortensen-Zakai (DMZ) equation, the stochastic DMZ equation, and the Kushner-Stratonovich equation--within a unified buffered weighted formulation. An exponential-type weight function and the corresponding weighted Sobolev spaces are introduced to enable a variational treatment of the filtering equations in a more general setting, in which the coefficients of the filtering system may be unbounded with polynomial growth. Under mild and easily verifiable assumptions, we first establish the well-posedness of the weak solution to the robust DMZ equation in these weighted spaces. Using the gauge (exponential) transformation and its inverse, these results are then transferred to the stochastic DMZ equation and the Kushner-Stratonovich equation, whose solutions are shown to exist and be unique in buffered weighted Sobolev spaces, yielding a unified treatment of all three filtering equations. Sufficient conditions for the well-posedness are also summarized, which illustrate the wide applicability of the proposed framework to general nonlinear filtering systems.

math.OC↗

arXiv2026

AEM: Adaptive Entropy Modulation for Multi-Turn Agentic Reinforcement Learning

Reinforcement learning (RL) has substantially improved the ability of large language model (LLM) agents to interact with environments and solve multi-turn tasks. However, effective agentic RL remains challenging: sparse outcome-only rewards provide limited guidance for assigning credit to individual steps within long interaction trajectories. Existing approaches often introduce dense intermediate supervision, such as process reward models or auxiliary self-supervised signals, which increases supervision and tuning complexity and may limit generalization across tasks and domains. We present AEM, a supervision-free credit assignment method that adaptively modulates entropy dynamics during RL training to improve the exploration-exploitation trade-off. Since in agentic RL the environment is typically affected by a complete response, rather than an individual token, our analysis lifts entropy dynamics from the token level to the response level, aligning uncertainty estimation with the effective action granularity of LLM agents and reducing sensitivity to token-level sampling noise. We further show that entropy drift under natural-gradient updates is governed by the interaction between the sampled-response advantage and its relative surprisal. Motivated by this result, AEM derives a practical response-level uncertainty proxy and uses it to rescale advantages, leveraging the evolving balance between positive and negative samples to naturally transition from exploration to exploitation. Extensive experiments on ALFWorld, WebShop, and SWE-bench-Verified with models ranging from 1.5B to 32B demonstrate that AEM consistently improves strong RL baselines, including a +1.4\% gain when integrated into a state-of-the-art software-engineering RL training framework.

cs.AI↗

arXiv2025

Interpretable Data Mining of Follicular Thyroid Cancer Ultrasound Features Using Enhanced Association Rules

Purpose: Thyroid cancer has been a common cancer. Papillary thyroid cancer and follicular thyroid cancer are the two most common types of thyroid cancer. Follicular thyroid cancer lacks distinctive ultrasound signs and is more difficult to diagnose preoperatively than the more prevalent papillary thyroid cancer, and the clinical studies associated with it are less well established. We aimed to analyze the clinical data of follicular thyroid cancer based on a novel data mining tool to identify some clinical indications that may help in preoperative diagnosis. Methods: We performed a retrospective analysis based on case data collected by the Department of General Surgery of Peking University Third Hospital between 2010 and 2023. Unlike traditional statistical methods, we improved the association rule mining, a classical data mining method, and proposed new analytical metrics reflecting the malignant association between clinical indications and cancer with the help of the idea of SHAP method in interpretable machine learning. Results: The dataset was preprocessed to contain 1673 cases (in terms of nodes rather than patients), of which 1414 were benign and 259 were malignant nodes. Our analysis pointed out that in addition to some common indicators (e.g., irregular or lobulated nodal margins, uneven thickness halo, hypoechogenicity), there were also some indicators with strong malignant associations, such as nodule-in-nodule pattern, trabecular pattern, and low TSH scores. In addition, our results suggest that the combination of Hashimoto's thyroiditis may also have a strong malignant association. Conclusion: In the preoperative diagnosis of nodules suspected of follicular thyroid cancer, multiple clinical indications should be considered for a more accurate diagnosis. The diverse malignant associations identified in our study may serve as a reference for clinicians in related fields.

cs.LG↗

arXiv2023

A Beam-Steering Reflectarray Antenna with Arbitrary Linear-Polarization Reconfiguration

This work presents a beam-steering reflectarray antenna that achieves arbitrary linear polarization (LP) reconfiguration. This antenna employs a dual-circular polarization (CP) reconfigurable reflectarray and an LP feed horn to generate an LP beam. The incident LP wave is decomposed into two CP components, whose reflection phases are independently adjusted. By re-combining these reflected CP beams with different phase constants, arbitrary LP states can be synthesized. Experimental validation is performed using a 16$\times$16 1-bit reconfigurable reflectarray operating at Ku band, demonstrating reconfigurable LP states of LP(0$^\circ$), LP(45$^\circ$), LP(90$^\circ$) and LP(135$^\circ$), as well as dynamic beam scanning functionality from 0$^\circ$ to 60$^\circ$. This polarization-reconfigurable, beam-steering reflectarray has potential for applications in satellite and mobile communications, where both beam patterns and polarization alignment are crucial.

physics.app-ph↗

arXiv2022

A Radiation Viewpoint of Reconfigurable Reflectarray Elements: Performance Limit, Evaluation Criterion and Design Process

Reconfigurable reflectarray antennas (RRAs) have rapidly developed with various prototypes proposed in recent literatures. However, designing wideband, multiband, or high-frequency RRAs faces great challenges, especially the lengthy simulation time due to the lack of systematic design guidance. The current scattering viewpoint of the RRA element, which couples antenna structures and switches during the design process, fails to address these issues. Here, we propose a novel radiation viewpoint to model, evaluate, and design RRA elements. Using this viewpoint, the design goal is to match the element impedance to a characteristic impedance pre-calculated by switch parameters, allowing various impedance matching techniques developed in classical antennas to be applied in RRA element design. Furthermore, the theoretical performance limit can be pre-determined at given switch parameters before designing specific structures, and the constant loss curve is suggested as an intuitive tool to evaluate element performance in the Smith chart. The proposed method is validated by a practical 1-bit RRA element with degraded switch parameters. Then, a 1-bit RRA element with wideband performance is successfully designed using the proposed design process. The proposed method provides a novel perspective of RRA elements, and offers a systematic and effective guidance for designing wideband, multiband, and high-frequency RRAs.

physics.app-ph↗