Institution profile

University of Sydney

Academic institutionaustralasia · au
Official website
Research library1,294linked papers
Opportunities0open roles
Selected work

Representative Papers

Early Stopping Against Label Noise Without Validation Data

Feb 11, 2025International Conference on Learning Representations

To address the performance degradation and model selection bias caused by validation-set-dependent early stopping under label noise, this paper proposes Label Wave—a validation-free, noise-robust early stopping method. Label Wave monitors the dynamic fluctuations of per-sample prediction confidence on the training set, revealing a strong correlation between minima in these fluctuations and the optimal generalization point. It automatically terminates training via sliding-window quantification and adaptive peak detection. As the first fully validation-free, noise-aware early stopping framework, it establishes a novel link between training dynamics (specifically, confidence fluctuation patterns) and generalization capability. Extensive experiments on benchmarks including CIFAR-10/100 and WebVision demonstrate that Label Wave consistently improves the accuracy of state-of-the-art noisy-label learning methods by 3.2–5.7% on average, while exhibiting superior stability compared to conventional validation-based early stopping.

9 citationsRead paper

Impact of Spatial Proximity on Drone Services

Oct 05, 2024UbiComp Companion

This study systematically investigates the pairwise impact of spatial proximity among drones during close-proximity flight on their energy consumption, with the aim of optimizing low-altitude logistics route planning. Through three-dimensional real-flight experiments conducted under varying wind conditions, high-fidelity energy consumption data were precisely collected across multiple drone segments. These data were integrated with high-precision spatial positioning and a customized graphical user interface to enable visualization of flight trajectories and inter-drone interactions. The work presents the first quantitative analysis of the coupled effects of inter-drone spacing, relative positioning, and wind conditions on energy use, uncovering key influence patterns that provide empirical foundations and optimization insights for designing efficient, energy-saving drone delivery systems.

4 citationsRead paper

SRL Proxemics: Spatial Guidelines for Supernumerary Robotic Limbs in Near-Body Interactions

Jan 31, 2026

This study addresses user concerns regarding safety, control, and trust when wearable supernumerary robotic limbs (SRLs) operate within peripersonal space. Through a Wizard-of-Oz experiment (n=18) integrating think-aloud protocols, semi-structured interviews, physiological signals, and post-task ratings, the research systematically investigates how varying levels of autonomy influence perceived safety and trust. The work proposes the SRL Proxemics framework, revealing that users conceptualize peripersonal space as distinctly partitioned and expect different limb components to adhere to specific coordination rules. Notably, higher autonomy does not necessarily enhance perceived safety; instead, spatially congruent and predictable behaviors are more effective in fostering trust. The framework underscores the necessity of fine-grained spatial calibration and enhanced behavioral legibility in SRL design, tailored to both proxemic zones and individual limb segments.

2 citationsRead paper

Action-Sketcher: From Reasoning to Action via Visual Sketches for Long-Horizon Robotic Manipulation

Jan 04, 2026arXiv.org

This work addresses the challenges of spatial reference ambiguity, task decomposition difficulty, and opaque decision-making that arise when using natural language instructions for long-horizon robotic manipulation. To overcome these issues, the authors propose employing editable visual sketches as an explicit intermediate representation, establishing a closed-loop “perceive–reason–sketch–act” workflow that precisely maps linguistic intent onto scene geometry. Through a multi-stage curriculum training framework, the approach integrates modality alignment, language-to-sketch consistency constraints, and sketch-guided reinforcement imitation learning to enable adaptive policy learning and real-time action prediction. Evaluated in both simulated and real-world complex environments, the method significantly improves task success rates and dynamic robustness while supporting human-in-the-loop correction and yielding interpretable, intervenable decision processes.

2 citationsRead paper

Inference on common trends in functional time series

Dec 01, 2023

This paper addresses unit root and cointegration inference for functional time series in Hilbert spaces, focusing on identifying the number of common stochastic trends—i.e., the dimension of the nonstationary subspace—and conducting hypothesis tests for the nonstationary and stationary subspaces. We systematically extend classical unit root and cointegration theory to arbitrary-dimensional Hilbert spaces for the first time, proposing a projection-operator spectral analysis method that is asymptotically efficient, fully data-driven (requiring no prior dimension specification), and uniformly applicable to high-dimensional vector time series, dynamic functional factor models, and curve-valued time series. We further develop theoretically grounded dimension-selection criteria and valid test statistics with rigorous asymptotic properties. Empirical applications to the U.S. yield curve and labor market indices robustly identify key common nonstationary trends, demonstrating the method’s interpretability and practical utility for real-world high-dimensional functional time series.

2 citationsRead paper
Recent publications

Latest Papers

Private Graph Property Testing

Sep 13, 2026

研究通过开发新的隐私放大技术,设计了具有正式隐私保证的高效图属性测试器,解决了大规模图属性测试中的隐私保护问题。

0 citationsRead paper