Institution profile

University of Western Ontario

Academic institutionnorthamerica · ca
Official website
Research library252linked papers
Opportunities0open roles
Selected work

Representative Papers

Enabling AutoML for Zero-Touch Network Security: Use-Case Driven Analysis

Jun 01, 2024IEEE Transactions on Network and Service Management

In 6G zero-touch networks (ZTNs), AI/ML-based security mechanisms suffer from labor-intensive hyperparameter tuning and vulnerability to adversarial attacks, posing critical safety bottlenecks. Method: This work proposes the first AutoML-empowered fully automated security framework for ZTNs, integrating automated machine learning (AutoML), adversarial machine learning (AML), multi-source anomaly detection, and explainable AI (XAI) to jointly model network traffic and model behavior—enabling autonomous intrusion detection and integrated adversarial defense. Contribution/Results: Experimental evaluation across multiple representative scenarios demonstrates >98.5% intrusion detection accuracy and robust resistance against mainstream adversarial attacks, including FGSM and PGD. The framework delivers the first deployable end-to-end AutoML solution for ZTN security, significantly reducing human intervention while enhancing model robustness and operational autonomy.

3 citationsRead paper

Boosting Methods for Interval-censored Data with Regression and Classification

Jan 25, 2026International Conference on Learning Representations

Interval-censored data are prevalent in survival analysis, yet conventional boosting methods struggle to handle them effectively. This work proposes a nonparametric boosting approach tailored for such data, uniquely integrating an unbiased transformation with functional gradient descent. By designing a customized loss function and employing imputation of response variables, the method enables scalable regression and classification modeling. Theoretical analysis elucidates the trade-off between mean squared error and optimality, while empirical experiments demonstrate that the approach is robust under finite-sample settings and substantially improves predictive accuracy. These advantages underscore its practical utility in domains such as medicine and engineering.

1 citationsRead paper

Full-Stack Alignment: Co-Aligning AI and Institutions with Thick Models of Value

Dec 03, 2025

This paper addresses governance risks arising from misalignment between AI systems and societal institutions’ or individuals’ values. To tackle this, it proposes a “full-stack alignment” framework centered on a novel “thick value model” that distinguishes enduring values from context-sensitive preferences, thereby enabling normative reasoning, modeling of collective goods, and cross-level value embedding. Methodologically, the framework integrates value-sensitive decision architectures, socially embedded agent design, value-aware institutional and economic mechanisms, and is empirically validated across five domains: AI governance, normative agent construction, win-win negotiation, meaning-preserving incentive design, and democratic oversight institutions. Results demonstrate that the framework systematically enhances value consistency between AI development and societal well-being. It offers a theoretically grounded yet practically viable alignment paradigm for trustworthy AI—bridging normative theory, institutional design, and technical implementation.

1 citationsRead paper

Event-Driven Online Vertical Federated Learning

Jun 17, 2025International Conference on Learning Representations

This paper addresses practical challenges in vertical federated learning (VFL), where clients possess non-overlapping features, data streams arrive asynchronously, and updates are triggered by local events—scenarios poorly supported by existing synchronous or time-driven online VFL frameworks. Method: We propose the first event-driven online VFL framework. It formally models event-asynchrony in VFL, introduces Dynamic Local Regret (DLR) as a novel performance metric, and establishes the first rigorous convergence theory for non-convex, non-stationary settings. We further design an event-triggered activation mechanism and an asynchronous collaborative optimization algorithm. Contribution/Results: We theoretically prove an upper bound on DLR that converges over time. Experiments demonstrate that our method achieves superior model stability under non-stationary data compared to state-of-the-art online VFL approaches, while significantly reducing communication overhead and computational cost.

1 citationsRead paper

Leveraging Hypernetworks and Learnable Kernels for Consumer Energy Forecasting Across Diverse Consumer Types

Feb 01, 2025IEEE Transactions on Power Delivery

Existing LSTM and Transformer models struggle to capture abrupt load changes and exhibit limited generalization across heterogeneous residential energy consumers—specifically students in dormitories, single-family homes, EV-integrated households, and townhouses. Method: We propose HyperEnergy, the first framework to employ a hypernetwork for dynamically generating LSTM parameters, coupled with a learnable hybrid kernel combining polynomial and radial basis function (RBF) components to enable adaptive modeling across diverse, non-stationary scenarios. The architecture integrates the hypernetwork, hybrid kernel, and LSTM backbone, trained jointly on real-world data from all four user types. Results: HyperEnergy consistently outperforms ten state-of-the-art baselines—including AttentionLSTM and Transformer—achieving an average 18.7% reduction in mean absolute error (MAE). It demonstrates superior prediction accuracy, cross-scenario generalization, and robustness to load volatility.

1 citationsRead paper
Recent publications

Latest Papers