Theresa Eimer - Hyperparameters in RL
Автор: RL and Agents Reading Group
Загружено: 2024-11-05
Просмотров: 117
Описание:
UoE RL Reading Group | 31 October 2024
Speaker: Theresa Eimer (Leibniz University Hannover)
Title: Hyperparameters in RL
Abstract: Hyperparameters are a necessary evil for training RL agents - tuning them is crucial for training success. How to best do this, however, is an open question. RL papers often feature limited grid searches and research on hyperparameters in RL usually lack standardized comparisons. This talk will shed light on the attributes of the hyperparameter landscape in RL, make a case of automated tuning approaches over manual tuning and compare some popular hyperparameter optimization approaches. We will also discuss the current limitations of tuning approaches for RL and where improved RL-specific methods could shine.
Link(s): https://arxiv.org/abs/2306.01324
Bio: Theresa Eimer is a PhD student at the Leibniz University Hannover broadly interested in making Reinforcement Learning more efficient and easier to apply via AutoRL. Her focus is on how to make low-level design decisions for RL algorithms.
Повторяем попытку...
Доступные форматы для скачивания:
Скачать видео
-
Информация по загрузке:
