구글 딥마인드의 LLM활용 로보틱스 논문이 한번에 여러개 나와,

네이버 클라우드 하정우 박사님께서 운영하시는 Weekly AI ArXiv에서 정말 가볍게 소개.

https://github.com/jungwoo-ha/WeeklyArxivTalk/issues/87#issuecomment-1606018653

 

[20230625] Weekly AI ArXiv 만담 시즌2 - 21회차 · Issue #87 · jungwoo-ha/WeeklyArxivTalk

Zoom webinar: https://navercorp.zoom.us/j/92208940283 News 6.25 73주년 순국선열들의 희생에 감사드립니다. Conferences EMNLP 2023 deadline: 모두 수고하셨습니다. CVPR 2023: 모두들 의미있는 학회 보내셨나요~ Best Paper: V

github.com

 

 

 

 

https://saytap.github.io/

 

SayTap: Language to Quadrupedal Locomotion

Use foot contact pattern to bridge LLM and low-level controller.

saytap.github.io

 

https://www.deepmind.com/blog/robocat-a-self-improving-robotic-agent?utm_source=twitter&utm_medium=social&utm_campaign=robocat 

 

RoboCat: A self-improving robotic agent

Robots are quickly becoming part of our everyday lives, but they’re often only programmed to perform specific tasks well. While harnessing recent advances in AI could lead to robots that could help in many more ways, progress in building general-purpose

www.deepmind.com

https://language-to-reward.github.io/

 

Language to Reward for Robotic Skill Synthesis

Project page for Language to Reward for Robotic Skill Synthesis.

language-to-reward.github.io

 

https://sites.google.com/view/agile-catching

'AI & RL > Foundation Model' 카테고리의 다른 글

[북마크] Formal Algorithms for Transformers (DeepMind, 2022)  (0) 2022.07.25
[요약] Do Prompt-Based Models Really Understand the Meaning of their Prompts? (Albert Webson, NAACL 2022)  (0) 2022.04.26
[북마크] Visual Prompting: Modifying Pixel Space to Adapt Pre-trained Models (방효진 님, ArXiv 2022)  (0) 2022.04.03

Author: Mary Phuong, Marcus Hutter
Paper Link: https://arxiv.org/abs/2207.09238
- 요약예정

'AI & RL > Foundation Model' 카테고리의 다른 글

Physical Grounding of LLM (SayTap, RoboCat, Language to Rewards for Robotic Skill Synthesis)  (0) 2023.06.22
[요약] Do Prompt-Based Models Really Understand the Meaning of their Prompts? (Albert Webson, NAACL 2022)  (0) 2022.04.26
[북마크] Visual Prompting: Modifying Pixel Space to Adapt Pre-trained Models (방효진 님, ArXiv 2022)  (0) 2022.04.03

Author: Albert Webson, Ellie Pavlick
Paper Link(arXiv): https://arxiv.org/abs/2109.01247

Paper Link(NAACL): https://openreview.net/forum?id=BhGMkxhZrW9 

Code: https://github.com/awebson/prompt_semantics

 

[NAVER AI Lab하정우 박사님 Weekly arXiv 소개내용 참고]

  • Large LM에서 prompt-based learning이 잘되는것이 prompt에 포함된 task instruction의 효과라고 생각해왔는데 이게 정말 그러한지를 실험적으로 분석함.

  • Prompt는 위와 같은 task와 연관되거나 관련없는 등 다양한 카테고리의 템플릿을 사용.

  • 위 표는 실험 결과로서, 체크표시는 instructive한 prompt가 그렇지 못한 prompt대비 통계적으로 유의하게 차이가 나는 경우를 의미.
  • 결과적으로 intrunction과는 성능이 크게 차이 없다는 놀라운 사실과 함께, 마지막 컬럼을 통해 prompt가 있기만 하면 few-shot 성능은 좋아진다는 것을 확인. 즉, LLM은 prompt의 instruction을 이해한것이 아니라는 기존의 생각과 반하는 결과. 특히 GPT-3는 체크표시가 없음.

'AI & RL > Foundation Model' 카테고리의 다른 글

Physical Grounding of LLM (SayTap, RoboCat, Language to Rewards for Robotic Skill Synthesis)  (0) 2023.06.22
[북마크] Formal Algorithms for Transformers (DeepMind, 2022)  (0) 2022.07.25
[북마크] Visual Prompting: Modifying Pixel Space to Adapt Pre-trained Models (방효진 님, ArXiv 2022)  (0) 2022.04.03

Author: Hyojin Bahng, Ali Jahanian, Swami Sankaranarayanan, Phillip Isola
Paper Link: https://arxiv.org/abs/2203.17274

Page: https://hjbahng.github.io/visual_prompting/

Code: yet.

 

'AI & RL > Foundation Model' 카테고리의 다른 글

Physical Grounding of LLM (SayTap, RoboCat, Language to Rewards for Robotic Skill Synthesis)  (0) 2023.06.22
[북마크] Formal Algorithms for Transformers (DeepMind, 2022)  (0) 2022.07.25
[요약] Do Prompt-Based Models Really Understand the Meaning of their Prompts? (Albert Webson, NAACL 2022)  (0) 2022.04.26

+ Recent posts

티스토리툴바