kth.sePublications
Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Developing a Risk-Averse Distributional Reinforcement Learning Algorithm
KTH, School of Engineering Sciences (SCI).
KTH, School of Engineering Sciences (SCI).
2024 (English)Independent thesis Basic level (degree of Bachelor), 10 credits / 15 HE creditsStudent thesis
Abstract [en]

This thesis explores the development of a risk-averse distributional reinforcement learning (DRL) algorithm designed to address the limitations of traditional reinforcement learning (RL) methods in handling uncertain and risk-sensitive environments. We present modifications to the standard C51 algorithm, and propose a risk-averse strategy by introducing the risk-adverse modifications for both loss function and policy selection. Through empirical evaluations conducted across four Atari games, we assess the performance of these modifications in comparison to the original C51 framework. Our results demonstrate that strategic adjustments to the policy can significantly enhance performance by reducing variability. Surprisingly, the modified algorithms achieve higher mean scores for some games. This result could be attributed to the chosen parameters and more data is needed in order to verify these results. Conversely, modifications to the loss function showed mixed results, often failing to improve and sometimes even degrading performance.

Place, publisher, year, edition, pages
2024.
Series
TRITA-SCI-GRU ; 2024:156
Keywords [en]
Reinforcement learning, distributional reinforcement learning, risk aversion, Conditional Value at Risk, C51 algorithm, Atari 2600 games
National Category
Mathematics
Identifiers
URN: urn:nbn:se:kth:diva-349249OAI: oai:DiVA.org:kth-349249DiVA, id: diva2:1880200
Educational program
Master of Science in Engineering -Engineering Physics
Supervisors
Examiners
Available from: 2024-07-01 Created: 2024-07-01 Last updated: 2024-07-01Bibliographically approved

Open Access in DiVA

No full text in DiVA

By organisation
School of Engineering Sciences (SCI)
Mathematics

Search outside of DiVA

GoogleGoogle Scholar

urn-nbn

Altmetric score

urn-nbn
Total: 135 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf