Bo-Hsu (Hentci) Ke | 柯柏旭

Hello! I'm Bo-Hsu Ke. Some people also call me Hentci—but please don't confuse it with Hentai.

I am pursuing my Master's degree in Computer Science and Engineering at National Yang Ming Chiao Tung University (NYCU), under the supervision of Prof. Wei-Chen Chiu at the Enriched Vision Applications Lab. I am also co-advised by Prof. Yu-Lun Liu at the Computational Photography Lab, specializing in 3D Computer Vision.

I was advised by Prof. Jian-Jhih Kuo during my undergraduate studies before joining NYCU. I earned my Bachelor’s degree in Computer Science from National Chung Cheng University in 2024.

Email  /  CV  /  Scholar  /  Linkedin  /  Github

Publications

I'm interested in Computer Vision, Federated Learning, and AI Security. Currently, most of my research is about Novel View Synthesis and AI Security.

StealthAttack proposes a novel density-guided poisoning method for 3D Gaussian Splatting (3DGS) that strategically injects illusory objects into low-density regions to create viewpoint-dependent visual illusions. The method combines point cloud poisoning with adaptive noise scheduling to disrupt multi-view consistency, enabling successful attacks where illusions are clearly visible from target views while maintaining high fidelity in innocent viewpoints.

Developed a reference-guided 3D inpainting approach utilizing SDEdit on aligned Gaussian initialization, and created a 360° inpainting dataset (360-USID) for comprehensive evaluation.

This paper proposes a novel defense framework called FDDF (Features Distraction Defense Framework) to mitigate trigger backdoor attacks in federated learning-based intrusion detection systems by identifying and eliminating the most significant features that may contain triggers, without interfering with the model training process.

We propose the Knowledge Distillation Defense Framework (KDDF) to detect and remove features of the potential triggers during the inference. KDDF utilizes Knowledge Distillation (KD) to train a validation model on each IoT device, which is used to identify suspicious data.

This paper proposes a Successive Interference Cancellation-based Defense Framework (SICDF) to detect and eliminate the trigger during model inference.

Projects

Evaluating the effectiveness of various methods (Restormer, fine-tuned diffusion model, and Vision Mamba) in mitigating simulated image artifacts to enhance NeRF and 3DGS performance in novel-view synthesis.

Honors

Stolen from Jon Barron's website.
Last updated Sep 2024.