Breadcrumb

Latest Colloquia

Robust Interactive Decision Making for Autonomous Navigation

Abstract: Modern intelligent systems, such as social robots and autonomous vehicles, interact frequently with humans. Their behaviors are highly complex and dynamic, which results from unobservable social interactions. Thus, building reliable autonomy that safely navigates multi-agent scenarios requires scalable and generalizable relational reasoning and interaction modeling between interactive agents. Meanwhile, robots should be able to...
By Jiachen Li |

Usefulness and safety Trade-offs in Language Models

Abstract: Recent progress in large language models (LLMs) calls for a thorough safety inspection of these models. In this talk, I will discuss three of our recent works on adversarial attacks related to natural languages. We first review common concepts of jailbreaking LLMs and discuss the trade-offs between their usefulness and safety. Then, we move...
By Yue Dong |

MixTraining:

Abstract: Pretrain-finetune has emerged as a powerful learning paradigm, achieving remarkable accuracy gains in various domains. However, its substantial computational requirements limit its application to broader areas. To address this challenge, we develop MixTraining, a novel training framework that---for the first time---incorporates asynchronous computation into the standard pretrain-finetune paradigm. At a high level, our MixTraining...
By Yinglun Zhu |

Ensemble and Context-Based Methods for Efficient Blackbox Attacks

Abstract: Artificial intelligence and machine learning models have experienced a transformative evolution in capability and deployment. Recently, large-scale models trained on expansive datasets have revolutionized numerous domains, from online search to media creation. However, with the growing power of these AI systems, the urgency to address their security concerns has also escalated. Understanding and fortifying...
By Zikui Cai |

Shrek MCMC:

Abstract: Markov chain Monte Carlo (MCMC) requires only the ability to evaluate the likelihood, making it a common technique for inference in complex models. However, it can have a slow mixing rate, requiring the generation of many samples to obtain good estimates and an overall high computational cost. In this talk, I will present a...
By Harini Venkatasan |

R^3: On-device Real-Time Deep Reinforcement Learning for Autonomous Robotics

Abstract: In this talk, I will introduce R^3, a new system that helps robots learn and make decisions faster and more efficiently right where they operate. R^3 smartly manages the robot's learning process by adjusting how much data it processes and remembers, ensuring it doesn't run out of memory. This is crucial for robots that...
By Zexin Li |

Towards robust edge intelligence in autonomous systems

Abstract: Using advanced 3D sensors and sophisticated deep learning models, autonomous systems such as self-driving cars, delivery drones are already transforming our daily life. However, a significant remaining challenge for further advancement is the reliability, robustness, and the ability to anticipate and handle long-tail events and corner-cases. Humans, on the other hand, are extremely good...
By Hang Qiu |

Source-Free Adaptation of Uni-Modal Models to Multi-Modal Targets

Abstract: Scene understanding using multi-modal data is necessary in many applications, e.g., autonomous navigation. To achieve this in a variety of situations, existing models must be able to adapt to shifting data distributions without arduous data annotation. Current approaches assume that the source data is available during adaptation and that the source consists of paired...
By Cody Simmons |
Let us help you with your search