[Paper Summary] Objective Mismatch in Model-based Reinforcement Learning

Автор: Nathan Lambert

Загружено: 4 февр. 2021 г.

Просмотров: 236 просмотров

Описание:

Two optimization problems leave model-based RL in a tricky point: you cannot optimize both the model and the controller simultaneously. This video points a direction for a new class of model-based RL algorithms.

[Paper Summary] Objective Mismatch in Model-based Reinforcement Learning

Доступные форматы для скачивания:

Скачать видео mp4

Информация по загрузке:

Скачать аудио mp3

Похожие видео

Traits of next generation reasoning models

Traits of next generation reasoning models

15min History of Reinforcement Learning and Human Feedback

15min History of Reinforcement Learning and Human Feedback

Early stages of the reinforcement learning era of language models

Early stages of the reinforcement learning era of language models

The Most Important Algorithm in Machine Learning

The Most Important Algorithm in Machine Learning

[Talk] Dissertation Talk: Synergy of Prediction and Control in Model-based Reinforcement Learning

[Talk] Dissertation Talk: Synergy of Prediction and Control in Model-based Reinforcement Learning

The Incredible Properties of Composite Materials

The Incredible Properties of Composite Materials

Blender Tutorial for Complete Beginners - Part 1

Blender Tutorial for Complete Beginners - Part 1

GRPO's new variants and implementation secrets

GRPO's new variants and implementation secrets

An update on DPO vs PPO for LLM alignment

An update on DPO vs PPO for LLM alignment

Hough Transform | Boundary Detection

Hough Transform | Boundary Detection