Institution profile

Simon Fraser University

Academic institutionnorthamerica · ca
Official website
Research library498linked papers
Opportunities0open roles
Selected work

Representative Papers

Smart Split-Federated Learning over Noisy Channels for Embryo Image Segmentation

Jun 04, 2023IEEE International Conference on Acoustics, Speech, and Signal Processing

This work addresses the detrimental impact of communication noise on training stability and accuracy in SplitFed learning for medical image segmentation. To mitigate this issue, the authors propose an intelligent averaging strategy that enhances model robustness against highly noisy communication channels within the Split-Federated learning framework. The proposed method maintains high segmentation accuracy while tolerating communication noise levels up to two orders of magnitude greater than those manageable by conventional averaging mechanisms. Consequently, it significantly improves both the stability and performance of the system under adverse channel conditions, enabling more reliable deployment of SplitFed learning in real-world medical imaging applications where communication quality may be limited.

9 citationsRead paper

Evaluating Human-AI Interaction via Usability, User Experience and Acceptance Measures for MMM-C: A Creative AI System for Music Composition

Aug 01, 2023International Joint Conference on Artificial Intelligence

This study investigates the adoption mechanisms of the lightweight AI music co-creation tool MMM-C among amateur and professional composers, focusing on usability, user experience, and technology acceptance. Methodologically, it introduces a novel “single-parameter” minimalist AI co-creative interface and conducts the first dual-layer mixed-method empirical evaluation within a real-world DAW environment (Cubase), integrating standardized instruments—System Usability Scale (SUS), User Experience Questionnaire (UEQ), and Unified Theory of Acceptance and Use of Technology 2 (UTAUT2)—alongside surveys, semi-structured interviews, and behavioral observation. Results indicate high system usability (SUS >75), strong acceptance (UTAUT2 mean >4.2/5), and broad user endorsement of novelty and ease of use; however, controllability and predictability require further refinement. Crucially, no significant performance differences were observed between amateur and professional users, empirically validating the cross-experience applicability of this lightweight co-creation paradigm.

7 citations1 influentialRead paper

Climate Implications of Diffusion-based Generative Visual AI Systems and their Mass Adoption

May 24, 2025International Conference on Innovative Computing and Cloud Computing

The rapid proliferation of diffusion-model-based generative visual AI poses significant, yet poorly quantified, climate risks due to surging GPU compute demand, uncertain user behavior, and opaque embodied carbon emissions. Method: We develop a comprehensive analytical framework integrating energy consumption modeling, GPU compute profiling, user adoption forecasting, and full-lifecycle energy accounting. Contribution/Results: This study provides the first empirical quantification of per-request carbon footprints for mainstream consumer-grade text-to-image systems (e.g., DALL·E, Stable Diffusion), revealing that their compute-specific energy intensity substantially exceeds that of conventional cloud services. Owing to massive user bases and high interaction frequency, annual electricity demand may reach tens of TWh—potentially surpassing cryptocurrency mining in climate impact. We identify a “scale–energy non-linear amplification effect” inherent to generative AI, wherein marginal increases in usage trigger disproportionate energy growth. These findings deliver critical, empirically grounded metrics to inform green AI governance, sustainability standards, and climate-aware system design.

5 citations2 influentialRead paper

Apollo: An Interactive Environment for Generating Symbolic Musical Phrases using Corpus-based Style Imitation

Apr 18, 2025

This study addresses the lack of lightweight, corpus-driven style imitation tools for symbolic music composition. We propose a corpus-based symbolic music style imitation method and implement it in Apollo, an interactive desktop application. Apollo enables users to construct custom Western classical music corpora; it performs low-latency stylistic melody generation via MIDI parsing and local pattern matching. The system supports corpus management, real-time preview, controllable parameter adjustment, and MIDI export/streaming. Its primary contribution is the first corpus-driven, user-customizable, and real-time interactive symbolic music style generation system designed specifically for musicians and researchers—filling a critical gap in lightweight creative assistance tools. Experimental evaluation demonstrates that Apollo reliably generates stylistically coherent musical phrases and significantly enhances compositional exploration efficiency and stylistic analysis capability.

4 citationsRead paper

Glocal Smoothness: Line Search can really help!

Jun 14, 2025

Classical iteration complexity analyses for first-order optimization methods rely on global Lipschitz continuity of the gradient, failing to exploit beneficial local smoothness—where the Lipschitz constant varies across regions—and thus incur unnecessary conservatism. Method: We introduce “glocal smoothness,” a novel structural assumption that simultaneously captures both global and local smoothness properties of the objective function—without dependence on algorithmic trajectories—thereby enabling trajectory-agnostic complexity bounds governed solely by intrinsic function constants. Contribution/Results: Under glocal smoothness, we establish improved iteration complexity for gradient descent with backtracking line search—surpassing that of fixed-step accelerated methods. Moreover, we provide a unified, refined convergence analysis for diverse algorithms including Polyak’s step size, adaptive gradient descent (AdGD), coordinate descent, stochastic and deterministic gradient methods, and nonlinear conjugate gradient, yielding significantly tighter complexity bounds across all cases.

2 citationsRead paper
Recent publications

Latest Papers

Recurrent Dynamic Range Extension

Sep 11, 2026

该研究通过递归执行网络逐步扩展图像的动态范围,使用记忆回放减少重建误差,从而解决复杂场景中高动态范围图像的重建问题。

0 citationsRead paper