Institution profile

Florida International University

Academic institutionnorthamerica · us
Official website
Research library250linked papers
Opportunities0open roles
Selected work

Representative Papers

Fairness Definitions in Language Models Explained

Jul 26, 2024arXiv.org

Large language models (LLMs) often inherit and amplify societal biases—such as gender and racial biases—while existing fairness definitions suffer from conceptual ambiguity, ill-defined boundaries, and unclear applicability, hindering rigorous fairness evaluation and governance. Method: We propose the first taxonomy of fairness concepts specifically designed for LLMs, systematically distinguishing over 12 mainstream fairness definitions based on their theoretical foundations and operational mechanisms. Through empirical experiments across model scales—including bias measurement and cross-definition comparative analysis—we evaluate context-dependent applicability. Contribution/Results: Our work clarifies logical boundaries and practical efficacy of fairness definitions, establishes a unified terminology framework, and releases open-source, reproducible code and pedagogical resources. This advances standardization, comparability, and methodological rigor in LLM fairness research.

11 citationsRead paper

Preserving AUC Fairness in Learning with Noisy Protected Groups

May 24, 2025

When protected attributes (e.g., gender, race) are corrupted by label noise, AUC-based fairness metrics—particularly the AUC difference (ΔAUC)—are highly susceptible to degradation. Existing methods assume clean sensitive attributes, limiting their practical robustness. Method: This paper proposes the first theoretically grounded robust AUC fairness optimization framework. Departing from prior assumptions, it introduces distributionally robust optimization (DRO) into AUC fairness learning, deriving a provable upper bound on ΔAUC under attribute noise. The method integrates AUC gradient approximation, noise-robust loss design, and multi-task fairness constraints to ensure robust modeling despite noisy protected attributes. Results: Extensive experiments on tabular and image datasets demonstrate that our approach significantly outperforms state-of-the-art methods, reducing average ΔAUC by 37% while preserving baseline AUC performance—confirming both fairness improvement and predictive utility.

2 citationsRead paper

Grammar-Forced Translation of Natural Language to Temporal Logic using LLMs

Dec 18, 2025International Conference on Machine Learning

To address three key bottlenecks in natural language (NL) to temporal logic (TL) translation—imprecise atomic proposition (AP) extraction, challenging coreference resolution, and poor few-shot generalization—this paper proposes the first syntax-constrained, two-stage collaborative optimization framework. In the *lifting* stage, a theoretically grounded constrained decoding mechanism compresses the solution space to enhance learning efficiency. In the *translation* stage, TL syntactic structure guidance, AP extraction constraints, and domain-adaptive prompting are jointly integrated. Crucially, our method enables end-to-end joint optimization of lifting and translation—previously unachieved. Evaluated on CW, GLTL, and Navi benchmarks, it achieves an average 5.49% improvement in end-to-end accuracy and a 14.06% gain in cross-domain generalization accuracy.

1 citationsRead paper

Enhancing Robot Navigation Policies with Task-Specific Uncertainty Management

May 20, 2025arXiv.org

Robots navigating complex environments face uncertainty arising from sensor noise, environmental dynamics, and incomplete information; moreover, localization accuracy requirements vary significantly across task regions—e.g., high precision near obstacles versus relaxed tolerance in open areas. To address this, we propose the Task-Specific Uncertainty Map (TSUM) and the Generalized Uncertainty Integration for Decision-making and Estimation (GUIDE) framework. GUIDE is the first to explicitly model and embed spatially and task-coupled uncertainty tolerances into navigation policies, eliminating reliance on handcrafted reward functions. Our approach jointly models multi-sensor noise, performs online uncertainty propagation estimation, and integrates with model-free reinforcement learning (PPO/SAC). Real-world experiments demonstrate a 37% increase in path success rate, a 52% reduction in collision rate, and a 2.1× improvement in localization accuracy within uncertainty-sensitive regions.

1 citationsRead paper

Prediction of Received Power in Low-Power Networks Deployed on the Surface of Rough Waters

Feb 19, 2025

To address link instability and unpredictable received power caused by wave-induced motion in low-power IoT nodes operating over rough water surfaces, this paper proposes a lightweight motion-aware received power prediction model tailored for embedded devices. Methodologically, it replaces the high-complexity matrix inversion in conventional MMSE estimation with online gradient descent for channel parameter estimation; furthermore, it introduces the first joint modeling of surface-node motion statistics—such as pitch and roll distributions—with multipath fading, enabling a time-aware received power prediction framework. Experimental results demonstrate 91% prediction accuracy with minimal iterations, while computational overhead is reduced by two orders of magnitude—enabling real-time, adaptive communication on resource-constrained IoT nodes. Key contributions include: (i) a novel motion-channel joint modeling paradigm; (ii) a low-complexity online parameter estimation mechanism; and (iii) a lightweight prediction architecture specifically designed for dynamic aquatic environments.

1 citationsRead paper
Recent publications

Latest Papers