Salvato in:
| Autori principali: | , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2024
|
| Soggetti: | |
| Accesso online: | https://arxiv.org/abs/2403.10946 |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866910665911828480 |
|---|---|
| author | Xu, Ziping Zhang, Kelly W. Murphy, Susan A. |
| author_facet | Xu, Ziping Zhang, Kelly W. Murphy, Susan A. |
| contents | Online Reinforcement Learning (RL) is typically framed as the process of minimizing cumulative regret (CR) through interactions with an unknown environment. However, real-world RL applications usually involve a sequence of tasks, and the data collected in the first task is used to warm-start the second task. The performance of the warm-start policy is measured by simple regret (SR). While minimizing both CR and SR is generally a conflicting objective, previous research has shown that in stationary environments, both can be optimized in terms of the duration of the task, $T$.
In practice, however, in real-world applications, human-in-the-loop decisions between tasks often results in non-stationarity. For instance, in clinical trials, scientists may adjust target health outcomes between implementations. Our results show that task non-stationarity leads to a more restrictive trade-off between CR and SR. To balance these competing goals, the algorithm must explore excessively, leading to a CR bound worse than the typical optimal rate of $T^{1/2}$. These findings are practically significant, indicating that increased exploration is necessary in non-stationary environments to accommodate task changes, impacting the design of RL algorithms in fields such as healthcare and beyond. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2403_10946 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | The Fallacy of Minimizing Cumulative Regret in the Sequential Task Setting Xu, Ziping Zhang, Kelly W. Murphy, Susan A. Machine Learning Online Reinforcement Learning (RL) is typically framed as the process of minimizing cumulative regret (CR) through interactions with an unknown environment. However, real-world RL applications usually involve a sequence of tasks, and the data collected in the first task is used to warm-start the second task. The performance of the warm-start policy is measured by simple regret (SR). While minimizing both CR and SR is generally a conflicting objective, previous research has shown that in stationary environments, both can be optimized in terms of the duration of the task, $T$. In practice, however, in real-world applications, human-in-the-loop decisions between tasks often results in non-stationarity. For instance, in clinical trials, scientists may adjust target health outcomes between implementations. Our results show that task non-stationarity leads to a more restrictive trade-off between CR and SR. To balance these competing goals, the algorithm must explore excessively, leading to a CR bound worse than the typical optimal rate of $T^{1/2}$. These findings are practically significant, indicating that increased exploration is necessary in non-stationary environments to accommodate task changes, impacting the design of RL algorithms in fields such as healthcare and beyond. |
| title | The Fallacy of Minimizing Cumulative Regret in the Sequential Task Setting |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2403.10946 |