Shoaib Nazir

The adversarial attacks and defenses for LLMs encompass a wide range of techniques and strategies. Manually crafted and automated red teaming methods expose vulnerabilities, while white box access reveals potential for prefilling attacks. Defense approaches include RLHF, DPO, prompt optimization, and adversarial training. Inference-time defenses and representation engineering show promise but face limitations. The control vector baseline enhances LLM resistance by manipulating model representations.

The 3D occupancy prediction methods faced challenges in depth estimation, computational efficiency, and temporal information integration. Monocular vision struggled with depth ambiguities, while stereo vision required extensive calibration. Temporal fusion approaches, including attention-based, WrapConcat-based, and plane-sweep-based methods, attempted to address these issues but often lacked robust temporal geometry understanding.