東華大學圖書館 |

語系: 繁體中文

說明(常見問題)

回圖書館首頁

手機版館藏查詢

登入

回首頁 到查詢結果 [ null ]

切換: 標籤 | MARC模式 | ISBD

Algorithms for Deep Reinforcement Le...

Xie, Annie,

FindBook

Google Book

Amazon

博客來

Algorithms for Deep Reinforcement Learning in Non-Stationary Environments /

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	Algorithms for Deep Reinforcement Learning in Non-Stationary Environments // Annie Xie.
作者:	Xie, Annie,
面頁冊數:	1 electronic resource (123 pages)
附註:	Source: Dissertations Abstracts International, Volume: 86-05, Section: B.
Contained By:	Dissertations Abstracts International86-05B.
標題:	Robots. -
電子資源:	https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=31652063
ISBN:	9798346390053

Algorithms for Deep Reinforcement Learning in Non-Stationary Environments /
Xie, Annie,

Algorithms for Deep Reinforcement Learning in Non-Stationary Environments /Annie Xie. - 1 electronic resource (123 pages)

Source: Dissertations Abstracts International, Volume: 86-05, Section: B.

Reinforcement learning typically assumes that the agent's learning environment is static and unchanging. However, this assumption rarely holds in more realistic applications of reinforcement learning. In robotics applications, for example, non-stationarity can manifest in many ways, such as the need to navigate unpredictable terrains or assist human users whose preferences change over time. Therefore, the ability to quickly adapt to changing conditions of the environment is critical to a successful decision-making system. To model this non-stationarity more formally, we consider a setting where the learning environment evolves over decision time-steps. More specifically, there exist features of the learning environment that change with time, and potentially change as a function of the agent's actions. In response to this dynamism, the agent must learn to eventually cope with this non-stationarity on its own, and adapt its behavior to the changes, e.g., making different decisions to better assist users with new preferences.In this thesis, we focus on the design of algorithms for such dynamic environments. Concretely, we first present a unifying framework to capture the relevant features of the environment into a compact representation. We demonstrate that, under certain properties of the environment, this representation can be learned from interactions with the environment alone, without additional access to the underlying features during training. By treating the environment as a sequential latent variable model, with the time-varying features as the latent variables, we can leverage advances in unsupervised representation learning techniques to learn this representation from data. Furthermore, in situations where the dynamics of these features can be modeled, we learn a predictive model of them to forecast how they will change, especially in response to different actions taken by the agent.After acquiring such a representation, we can then design learning objectives to shape the agent's behavior. There are several types of non-stationarity that we characterize in this thesis, such as passive versus active and the frequency of changes, and leverage to design more effective policy learning algorithms. Specifically, using the learned representations, we present objectives that enable agents that (i) anticipate future environment features for faster and more robust adaptation and (ii) influence or shapehow these features change for higher long-term rewards. Finally, when the future latent features are unpredictable, we design agents that actively gather information about them through exploration of the environment.

English

ISBN: 9798346390053Subjects--Topical Terms:

529507
Robots.

Algorithms for Deep Reinforcement Learning in Non-Stationary Environments /
LDR:03927nmm a22003613i 4500 001 2430825
005 20250603111625.5
006 m o d
007 cr|nu||||||||
008 261011s2024 miu||||||m |||||||eng d
020 $a 9798346390053
035 $a (MiAaPQD)AAI31652063
035 $a (MiAaPQD)Stanfordty110wc6590
035 $a AAI31652063
040 $a MiAaPQD $b eng $c MiAaPQD $e rda
100 1 $a Xie, Annie, $e author. $3 3819338
245 1 0 $a Algorithms for Deep Reinforcement Learning in Non-Stationary Environments / $c Annie Xie.
264 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2024
300 $a 1 electronic resource (123 pages)
336 $a text $b txt $2 rdacontent
337 $a computer $b c $2 rdamedia
338 $a online resource $b cr $2 rdacarrier
500 $a Source: Dissertations Abstracts International, Volume: 86-05, Section: B.
500 $a Advisors: Finn, Chelsea Committee members: Bohg, Jeannette; Brunskill, Emma; Bent, Stacey F.
502 $b Ph.D. $c Stanford University $d 2024.
520 $a Reinforcement learning typically assumes that the agent's learning environment is static and unchanging. However, this assumption rarely holds in more realistic applications of reinforcement learning. In robotics applications, for example, non-stationarity can manifest in many ways, such as the need to navigate unpredictable terrains or assist human users whose preferences change over time. Therefore, the ability to quickly adapt to changing conditions of the environment is critical to a successful decision-making system. To model this non-stationarity more formally, we consider a setting where the learning environment evolves over decision time-steps. More specifically, there exist features of the learning environment that change with time, and potentially change as a function of the agent's actions. In response to this dynamism, the agent must learn to eventually cope with this non-stationarity on its own, and adapt its behavior to the changes, e.g., making different decisions to better assist users with new preferences.In this thesis, we focus on the design of algorithms for such dynamic environments. Concretely, we first present a unifying framework to capture the relevant features of the environment into a compact representation. We demonstrate that, under certain properties of the environment, this representation can be learned from interactions with the environment alone, without additional access to the underlying features during training. By treating the environment as a sequential latent variable model, with the time-varying features as the latent variables, we can leverage advances in unsupervised representation learning techniques to learn this representation from data. Furthermore, in situations where the dynamics of these features can be modeled, we learn a predictive model of them to forecast how they will change, especially in response to different actions taken by the agent.After acquiring such a representation, we can then design learning objectives to shape the agent's behavior. There are several types of non-stationarity that we characterize in this thesis, such as passive versus active and the frequency of changes, and leverage to design more effective policy learning algorithms. Specifically, using the learned representations, we present objectives that enable agents that (i) anticipate future environment features for faster and more robust adaptation and (ii) influence or shapehow these features change for higher long-term rewards. Finally, when the future latent features are unpredictable, we design agents that actively gather information about them through exploration of the environment.
546 $a English
590 $a School code: 0212
650 4 $a Robots. $3 529507
650 4 $a Adaptation. $3 3562958
650 4 $a Sensitivity analysis. $3 3560752
650 4 $a Hockey. $3 3689723
650 4 $a Robotics. $3 519753
690 $a 0771
710 2 $a Stanford University. $e degree granting institution. $3 3765820
720 1 $a Finn, Chelsea $e degree supervisor.
773 0 $t Dissertations Abstracts International $g 86-05B.
790 $a 0212
791 $a Ph.D.
792 $a 2024
856 4 0 $u https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=31652063