Delta på prøveforelesning - 5. mai kl. 10:15 (rom 402, Institutt for teknologisystemer)
"Reinforcement Learning in the Real World: From Transportation, Robotics
to Multi-Agent Systems".
Kreeringssammendrag
Forsterkelseslæring er en kraftig teknikk som hjelper programmer med å ta smarte avgjørelser, som for eksempel i videospill eller selvkjørende biler. Jeg har forsket på hvordan vi kan gjøre disse programmene mer effektive ved å forbedre måten de lærer på, slik at de trenger mindre data. Dette kan føre til mer pålitelige systemer i både virtuelle og virkelige situasjoner, og åpne for nye muligheter i teknologiutvikling.
Hovedfunn
This research aims to improve how computer programs learn to make decisions, like those used in self-driving cars or video games. A key focus has been enhancing how these programs understand and process the information they receive from their environment. By doing this, they can make quicker and more accurate decisions.
We explored several strategies to achieve this goal. One method involved using varied samples during training to diversify the learning process, helping the program to adapt to different situations. We also incorporated advanced models to interpret data more effectively, boosting the program's decision-making abilities without having to change the underlying method. Furthermore, combining human expert insights with the program's learning process helped improve its overall performance, especially in complex tasks where human-like judgment is beneficial.
These approaches suggest that better learning models can be applied beyond theoretical settings and could be useful for practical, real-world applications. The study's findings highlight the value of diverse and informed learning processes, paving the way for more reliable and efficient computer-based decision systems that can handle different environments and tasks with increasing sophistication and efficiency.
Leder av disputas
Prof., PhD Cecilie Rolstad Denby, Universitetet i Oslo
Bedømmelseskomité
1. Opponent: Førsteamanuensis, PhD, Yali Du, King's College London
2. Opponent: Førsteamanuensis, PhD, Djordje Grbic, IT-Universtitetet i København
Komitéleder: Prof., PhD, Josef Noll, Universitetet i Oslo
Veiledere
Hovedveileder: Prof., PhD, Paal Engelstad, Universitetet i Oslo
Medveileder: Prof., PhD, Morten Goodwin, Universitetet i Agder
Medveileder: Prof., PhD, Anis Yazidi, OsloMet
Kontaktinformasjon til kandidat
Epost: mengli11235@gmail.com
Mobil: 48 63 01 27
For mer informasjon
PhD-koordinator: Yvonne Baade