SIT796 - Reinforcement Learning
2023 unit information
|Enrolment modes:||Trimester 1: Waurn Ponds (Geelong), Cloud (online)|
SIT720, SIT771 and SIT787.
Students will on average spend 150 hours over the teaching period undertaking the teaching, learning and assessment activities for this unit.
|Scheduled learning activities - campus||
1 x 2 hour online class per week, 1 x 2 hour workshop per week.
|Scheduled learning activities - cloud (online)||
Online independent and collaborative learning including optional scheduled activities as detailed in the unit site.
Reinforcement Learning (RL) is one of the three fundamental paradigms of Machine Learning that is inspired by psychology and neuroscience and is concerned with the development of software agents capable of taking actions in an environment to achieve one or more goals. RL differs from supervised learning in that it does not require correct input/output pairs and incorrect actions do not need to be directly corrected, instead agents balance exploration and exploitation to search for optimal policies. In this unit students will explore, research and implement solutions to a range of RL problems or Markov Decision Process (MDP), including variants such as: discrete-time MDP; Semi-MDPs (SMDP); continuous-time MDPs; Partially Observable-MDP (POMDP) and Multi-objective MDPs (MOMDP). In solving these problems, students will apply their knowledge and skills with a range of techniques such as: multi-armed bandits; reward design; value iteration; policy gradient; temporal difference learning; on-policy; off-policy; eligibility traces; feature construction; and, continuous action. Additionally, students will research and discuss topics such as; curriculum; interactive; inverse; transfer; ensemble; hierarchical; curiosity, multi-goal; multi-agent; multi-objective and deep RL.
To be eligible to obtain a pass in this unit, students must meet certain milestones as part of the portfolio.
Unit Fee Information
Click on the fee link below which describes you: