Large language models (LLMs) have shown promise in providing scalable mental health support, while evaluating their counseling capability remains crucial to ensure both efficacy and safety. Existing evaluations are limited by the static assessment that focuses on knowledge tests, the single perspective that centers on user experience, and the open-loop framework that lacks actionable feedback. To address these issues, we propose Ψ-Arena, an interactive framework for comprehensive assessment and optimization of LLM-based counselors, featuring three key characteristics: (1) Realistic arena interactions that simulate real-world counseling through multi-stage dialogues with psychologically profiled NPC clients; (2) Tripartite evaluation that integrates assessments from the client, supervisor, and counselor perspectives; (3) Closed-loop optimization that iteratively improves LLM counselors using diagnostic feedback. Experiments across eight state-of-the-art LLMs show significant performance variations in different real-world scenarios and evaluation perspectives. Moreover, reflection-based optimization results in up to a 141% improvement in counseling performance. We hope Ψ-Arena provides a foundational resource for advancing reliable and human-aligned LLM applications in mental healthcare.

Character-based dialogue (aka role-playing) enables users to freely customize characters for interaction, which often relies on LLMs, raising the need to evaluate LLMs’ character customization capability. However, existing benchmarks fail to ensure a robust evaluation as they often only involve a single character category or evaluate limited dimensions. Moreover, the sparsity of character features in responses makes feature-focused generative evaluation both ineffective and inefficient. To address these issues, we propose CharacterBench, the largest bilingual generative benchmark, with 22,859 human-annotated samples covering 3,956 characters from 25 detailed character categories. We define 11 dimensions of 6 aspects, classified as sparse and dense dimensions based on whether character features evaluated by specific dimensions manifest in each response. We enable effective and efficient evaluation by crafting tailored queries for each dimension to induce characters’ responses related to specific dimensions. Further, we develop CharacterJudge model for cost-effective and stable evaluations. Experiments show its superiority over SOTA automatic judges (e.g., GPT-4) and our benchmark’s potential to optimize LLMs’ character customization.

Jinfeng Zhou, Yuxuan Chen, Jianing Yin, Yongkang Huang, Yihan Shi, Xikun Zhang, Libiao Peng, Rongsheng Zhang, Tangjie Lv, Zhipeng Hu, Hongning Wang, Minlie Huang. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Indoor positioning within the Internet of Things (IoT) has become critically important, with accuracy and robustness being essential for its practical applications. However, the complexity of real-world disturbances, including both Gaussian and non-Gaussian noise, poses significant challenges to existing methods. Kernel adaptive filtering (KAF) has proven effective in addressing these challenges, yet traditional KAF approaches face issues with growing structural complexity and increasing memory demands. To address these drawbacks, this study introduces an enhanced sparsification technique within the random Fourier features (RFFs) framework. The improvement lies in adopting the generalized Gaussian distribution (GGD) to be better suited for real-world noise, leading to the development of the ESCGKAF algorithm—a robust solution for high-precision indoor positioning in online scenarios. The algorithm leverages the half-quadratic (HQ) optimization on the kernel risk-sensitive loss (KRSL) cost function, which is further refined through the conjugate gradient (CG) method, to achieve a superior performance in noisy environments. The proposed approach permits a more generalized representation of random features, surpassing traditional Gaussian-based RFF methods in terms of adaptability and robustness. To further enhance performance in complex scenarios, MESCGKAF algorithm is proposed by introducing a multikernel strategy. The effectiveness of the proposed algorithms is validated in two real-world scenarios, showing marked performance improvement.

Chujie Zheng, Jinfeng Zhou, Yinhe Zheng, Libiao Peng, Zhen Guo, Wenquan Wu, Zheng-Yu Niu, Hua Wu, Minlie Huang. Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing. 2022.

In this article, the multisynchronization issue of coupled memristive neural networks (CMNNs) with unbounded time delays is investigated. To begin with, a class of generalized Gaussian-wavelet-type activation functions is adopted to extend the number of stable equilibrium states. On this basis, a distributed impulsive controller is constructed to realize the multiple synchronization of the delayed CMNNs. Under the concepts of$\mu $-stability, Filippov solution, and differential inclusion, some sufficient conditions are derived such that the addressed system can possess$(2r + 1)^{n}\,\,\mu $-stable synchronization manifolds. The convergence performance of solutions is determined by the time delays. As the time delays increase, the stability of synchronization manifolds will transform from exponential stability to power-stability, log-stability, or log–log-stability as special cases. Moreover, considering the modeling error and external disturbance, we further investigated the multisynchronization of delayed CMNNs with parametric uncertainties and stochastic perturbations, and some robust multisynchronization criteria are obtained. Finally, the effectiveness of the obtained results is verified by numerical simulations.

In this letter, a novel kernel function named$q$-Rényi kernel is proposed. Based on it, a new online adaptive learning algorithm is presented, which is derived based on the recursive adaptive filtering paradigm under the reproducing kernel Hilbert space. The proposed learning algorithm is different from the conventional kernel-based learning paradigm in two senses: first, the reproducing kernel so-called$\boldsymbol {q}$-Rényi kernel is firstly derived and employed; and second, a sparsity constraint is utilized to generate a small size of neural networks while maintaining a high learning performance. The effectiveness of the proposed algorithm is demonstrated via numerical simulations.

In this letter, a novel information theoretic measure, namely q-Gaussian kernel mean p-power error (QKMPE), is proposed by defining the mean p-power error in the q-Gaussian kernel space, which is a generalization of the kernel mean p-power error measure. Furthermore, a recursive kernel adaptive filter algorithm, named as recursive least q-Gaussian kernel mean p-power, is derived under the least QKMPE criterion for robust learning in noisy environment. This new proposed algorithm reveals superior performance against Gaussian-type noise as well as the nonGaussian perturbation, especially when the data contain large outliers. Experimental results in the context of Mackey-Glass time series prediction confirm the effectiveness of the proposed algorithm.