Institution profile

University of Calgary

Academic institutionnorthamerica · ca
Official website
Research library358linked papers
Opportunities0open roles
Selected work

Representative Papers

NeRF: Neural Radiance Field in 3D Vision, A Comprehensive Review

Oct 01, 2022

This survey addresses the lack of unified taxonomies and reproducible benchmarks in existing NeRF literature. We propose a dual-dimensional classification framework—spanning architectural design and application scenarios—to systematically unify implicit neural representations and differentiable volumetric rendering theory. Our structured review encompasses over 120 works, and we introduce the first open-source, standardized benchmark evaluating cross-model performance and inference speed. Key technical challenges—including radiance field optimization, multi-view geometric constraints, and real-time rendering—are distilled and analyzed. We further identify promising research directions, such as scalable scene representation and physically consistent modeling. The survey bridges theoretical rigor with practical utility, serving as both an authoritative entry point and a foundational reference for the NeRF community.

53 citations2 influentialRead paper

Reducing Hallucinations in LLMs via Factuality-Aware Preference Learning

Jan 06, 2026arXiv.org

Preference alignment methods such as Direct Preference Optimization (DPO) can inadvertently exacerbate hallucinations in large language models by favoring fluent and confident—but potentially inaccurate—responses. To address this, this work proposes Factuality-aware DPO (F-DPO), which enhances standard DPO by incorporating binary factuality labels to invert preference rankings when necessary and introduces a factuality-aware margin that amplifies learning from samples with pronounced factual discrepancies. Requiring only binary labels—and no auxiliary reward models, token-level annotations, or multi-stage training—F-DPO significantly simplifies the alignment pipeline while improving generalization. Evaluated across seven open-source models ranging from 1B to 14B parameters, F-DPO reduces hallucination rates by up to fivefold (e.g., on Qwen3-8B) and boosts factuality scores by 50%. On TruthfulQA, it improves MC1 and MC2 accuracy by 17% and 49%, respectively.

2 citations1 influentialRead paper

Double-Edge-Assisted Computation Offloading and Resource Allocation for Space-Air-Marine Integrated Networks

Sep 01, 2025IEEE Transactions on Vehicular Technology

To address high energy consumption and stringent end-to-end latency requirements in computational task offloading for Maritime Autonomous Surface Ships (MASSs) within Space-Air-Sea Integrated Networks (SAMINs), this paper proposes a dual-edge collaborative computation offloading and resource allocation framework. It enables MASSs to concurrently offload tasks to both Unmanned Aerial Vehicle (UAV)-based and Low Earth Orbit (LEO) satellite-based edge servers. We innovatively design an “air–space” dual-edge collaborative architecture and jointly optimize offloading decisions, task partitioning ratios, and wireless/computational resource allocations. An alternating optimization (AO) approach combined with a hierarchical solution strategy is adopted to minimize total system energy consumption under strict end-to-end latency constraints. Simulation results demonstrate that the proposed scheme reduces energy consumption significantly compared to baseline algorithms, achieving up to a 23.6% improvement in energy efficiency—thereby validating the effectiveness and superiority of air–space collaborative edge computing in maritime applications.

2 citations1 influentialRead paper

Physics-constrained DeepONet for Surrogate CFD models: a curved backward-facing step case

Mar 14, 2025

This work addresses the insufficient accuracy of surrogate modeling for backward-facing curved step flows under sparse-data conditions. We propose a physics-constrained DeepONet (PC-DeepONet), which, for the first time, enforces mass conservation—i.e., zero divergence of the velocity field—as a hard constraint within the DeepONet architecture. The method integrates parameterized geometric mapping with CFD data-driven training. Compared to purely data-driven baselines, PC-DeepONet achieves convergence using only 50 training samples and 50 optimization iterations, significantly improving prediction accuracy and physical consistency in low-data regimes—particularly enhancing generalization capability for velocity and pressure fields. Our key contribution is the pioneering design of a divergence-free neural operator that simultaneously ensures high fidelity and strong physical interpretability, establishing a new paradigm for CFD surrogate modeling in geometrically complex, data-scarce scenarios.

1 citationsRead paper

When Agents Look Like Beacons: NIDS Evasion by Model Context Protocol Traffic

Sep 16, 2026

"This study addresses the challenge that MCP traffic in enterprise networks exhibits behavior similar to malware C2 beacons, making it difficult for traditional NIDS to distinguish. The research establishes a Docker-based test environment to simulate MCP JSON-RPC traffic patterns under various TLS conditions, evaluating the performance of Suricata signature matching and RITA behavioral scoring. This work proposes a set of native network indicators for AI agents, including Agent-Native ALPN and standardized out-of-band headers, to enhance the detection capabilities of NIDS for generative AI inference loop traffic. Experimental results demonstrate that, across all tested conditions, MCP traffic consistently evades detection, with a uniform behavioral beacon score of 0.0 and minimal content alerts."

0 citationsRead paper
Recent publications

Latest Papers

When Agents Look Like Beacons: NIDS Evasion by Model Context Protocol Traffic

Sep 16, 2026

"This study addresses the challenge that MCP traffic in enterprise networks exhibits behavior similar to malware C2 beacons, making it difficult for traditional NIDS to distinguish. The research establishes a Docker-based test environment to simulate MCP JSON-RPC traffic patterns under various TLS conditions, evaluating the performance of Suricata signature matching and RITA behavioral scoring. This work proposes a set of native network indicators for AI agents, including Agent-Native ALPN and standardized out-of-band headers, to enhance the detection capabilities of NIDS for generative AI inference loop traffic. Experimental results demonstrate that, across all tested conditions, MCP traffic consistently evades detection, with a uniform behavioral beacon score of 0.0 and minimal content alerts."

0 citationsRead paper