BYR Achieve · 镜像论坛

来帮本科的毕设导师Changjae Oh发个招生帖，老师人很棒，21年跟着做毕设之后投了一篇文章，然后读MSc的时候老师又带着打了一个竞赛。全程基本上是手把手教我做科研，从看cs231n到怎么找创新点再到文章该怎么写，然后一遍遍和我迭代文章。我从学术小白到现在大概能独立完成一篇文章的初稿（虽然有时候还是会被现在的导师吐槽hhh），都要归功于Changjae老师的教导。当时学校申请的时候申了20多个项目，老师不同项目都给写了不同的推荐信（当时我被吐槽究竟还要再申多少所学校hhh），今年和MSc的导师读博，我本科老师的推荐信到的比现在的导师还要快。总之，老师的人真的很棒！科研方面，我当时是每周固定开会，聊聊天再聊聊进度，组里面GPU资源丰富，如果我没记错的话是有QM cluster加上英国的Tier 2 HPC可以用，完全不用担心没卡。论文方面，老师可以说是十分负责了，会一遍一遍的帮忙review，然后提建议，先交会我们怎么写，然后到ddl两周左右的时候基本就会上手帮忙改了（我当时论文被改完之后我都不认识了hhh）。果园有在Changjae Oh老师那里做毕设的同学应该比较清楚体验如何哈哈哈哈，具体的项目简介放在了下面，如果有兴趣的同学可以用学校邮箱直接给老师发邮件，可以提一嘴在北邮论坛上看到的帖子，估计应该会很快收到回复。 **CSC scholarship position at QMUL (September 2023 entrance)** **Title:** Adversarially robust AI agents **Abstract:** Artificial intelligence (AI) agents, e.g. robots, have successfully integrated deep neural networks that enable various data-driven AI tasks such as navigation and manipulation. However, training AI agents is difficult due to the requirement of a large amount of data in order to learn control policies for the end tasks, and it becomes even more challenging if the data is collected from the physical environment. AI agents are sensitive to input modification, such as illumination changes, noises in visual inputs or physical force applied to the AI agents, which can be frequently happened in a real-world scenario and affect their perception and policy to perform AI tasks. This project will investigate adversarial examples in an embodied AI where the perturbations can mislead vision-based perception and learned policy. The project aims to understand various types of adversarial examples and to improve the robustness of the perception and control policies of the AI agent under various types of unexpected events in a real-world scenario. The main challenge is to understand and design the adversarial perturbations that mislead embodied AI agents, model the adversarial perturbations, and devise the defence algorithms to achieve the robustness of the embodied AI agents. The project will use vision-based robot manipulators as embodied AI agents to investigate the problems and address them in a real-world setting. **Supervisor:** Changjae Oh ([http://eecs.qmul.ac.uk/~coh/](https://eur01.safelinks.protection.outlook.com/?url=http%3A%2F%2Feecs.qmul.ac.uk%2F~coh%2F&data=05|01|hengyi.wang.21%40ucl.ac.uk|7dafa4dcc34643c03a1408dac64f3c0b|1faf88fea9984c5b93c9210a11d9a5c2|0|0|638040340855602354|Unknown|TWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D|3000|||&sdata=mBeaaS5IA%2Bsa1A2wzyuN1SQHHQaocML9V2fce0td%2B2U%3D&reserved=0))

[招生] Queen Mary University of London PhD CSC scholarship