Institution profile

University of Leeds

Academic institutioneurope · gb
Official website
Research library319linked papers
Opportunities0open roles
Selected work

Representative Papers

Deep Reinforcement Learning for Bipedal Locomotion: A Brief Survey

Apr 25, 2024arXiv.org

Existing deep reinforcement learning (DRL) frameworks for bipedal robot multi-task motion control suffer from poor generalization, low sim-to-real transfer efficiency, and a lack of systematic benchmarking. Method: We propose the first unified DRL taxonomy tailored to bipedal control, rigorously delineating the trade-offs between end-to-end and hierarchical control across task coverage, policy interpretability, and sim-to-real transfer. Our integrated framework synergizes PPO/SAC, hierarchical RL, model predictive control (MPC), and neural policy representation, accompanied by deployment principles balancing robustness and scalability. Contribution/Results: Leveraging a structured evaluation matrix spanning 30+ state-of-the-art methods, we identify multi-task generalization and cross-domain transfer as the two fundamental bottlenecks. This work establishes both theoretical foundations and engineering guidelines for DRL-driven embodied intelligent control.

3 citationsRead paper

The New Compiler Stack: A Survey on the Synergy of LLMs and Compilers

Jan 05, 2026CCF Transactions on High Performance Computing

Traditional compilers face limitations in development accessibility, optimization capabilities, and application scope. This work proposes the first multidimensional classification framework for large language model (LLM)-driven compilation, offering a systematic survey of existing research through four analytical dimensions: design philosophy, methodology, level of code abstraction, and task type. The study identifies three core design paradigms—Selector, Translator, and Generator—and highlights three transformative directions: democratizing compiler development, discovering novel optimization strategies, and expanding functional boundaries. It further argues that hybrid systems represent a critical pathway forward and provides a technical roadmap for building correct, scalable, and intelligent compilation tools.

1 citations1 influentialRead paper

Generative AI Voting: Fair Collective Choice is Resilient to LLM Biases and Inconsistencies

May 31, 2024arXiv.org

This work investigates fairness and representativeness when large language models (LLMs) serve as proxies for human voters in collective decision-making. We construct over 50,000 LLM-based voter personas and simulate their behavior across 306 real-world elections. For the first time, we apply proportional representation aggregation methods—such as Equal Shares—to LLM-based deliberative settings. By comparing complex preference voting against plurality rule and integrating multiple model generations (e.g., GPT-3/3.5, Llama2), we identify significant preference inconsistency among LLMs; yet fair aggregation yields highly representative outcomes. Notably, it robustly recovers representation for abstainers, enhancing democratic resilience under low turnout. Our core contribution is the empirical validation of AI-mediated representation feasibility and the demonstration that fairness-aware aggregation algorithms simultaneously improve fairness for both human constituents and AI proxies—a synergistic fairness gain.

1 citations1 influentialRead paper

To Predict or Not to Predict? Towards reliable uncertainty estimation in the presence of noise

Mar 07, 2026

This study addresses the challenge of unreliable predictions in multilingual text classification caused by noise and off-topic content. Focusing on the complex–simple sentence classification task, it systematically evaluates various uncertainty estimation methods—including Monte Carlo Dropout and softmax confidence—under low-resource settings, domain shifts, and multilingual conditions. The evaluation employs multidimensional metrics such as calibration, discriminative capacity, and decision threshold stability. Results demonstrate that Monte Carlo Dropout significantly outperforms conventional softmax-based approaches in noisy environments. Building on this insight, the authors propose a novel strategy that proactively abstains from high-risk predictions based on uncertainty estimates. Empirical validation on the Readme task shows that discarding the 10% most uncertain samples improves the macro F1 score from 0.81 to 0.85, underscoring the practical utility of this approach in enhancing model robustness and reliability.

1 citationsRead paper

How Much Noise Can BERT Handle? Insights from Multilingual Sentence Difficulty Detection

Mar 07, 2026

This study addresses the detrimental impact of noisy labels—particularly sentence-level labels derived from crowdsourced document-level annotations—on language model performance in non-topical classification tasks such as sentence difficulty prediction. It presents the first systematic evaluation of multiple denoising strategies, including Gaussian Mixture Models (GMM), Co-Teaching, noise transition matrices, and label smoothing, within a multilingual sentence difficulty prediction framework using multilingual BERT for cross-lingual training. The work reveals a complementary relationship between the inherent robustness of pretrained models and explicit denoising techniques, and introduces the largest multilingual sentence difficulty corpus to date. Experiments demonstrate that on small datasets, GMM-based denoising improves AUC from 0.52 to 0.92, with combined methods reaching 0.93; on larger datasets, while performance gains are modest (0.92→0.94), approximately 20% of noisy samples are successfully filtered, substantially enhancing corpus quality.

1 citationsRead paper
Recent publications

Latest Papers