The course is administered centrally via ILIAS. All current announcements are also communicated here.
13.10.2021 to 02.02.2022
- Lecture: 2 SWS, Wednesdays 2:30 – 4:00 PM, digital
- Exercise: 2 SWS, Wednesdays 4:30 – 6:00 PM, digital
- Master course "Informatik" (PO 2005, PO 2015)
- Master course "Artificial Intelligence and Data Science" (PO 2019, PO 2021)
- 5 CP
Content of the lecture:
- Introduction: Architecture of a spoken dialogue system, dialogue acts, turn management issues
- Semantic decoding: Representing and decoding meaning from user inputs, semantic decoding as a classification task, semantic decoding as a sequence-to-sequence learning task
- Dialogue state tracking: Tracking beliefs over multiple turns, classical generative and discriminative approaches, recent deep learning approaches, integration of decoding and tracking
- Dialogue Management: Modelling via Markov Decision Processes, reinforcement learning, gradient methods, Gaussian Processes
- Response Generation: Template methods, generative models, recent neural network approaches
- Current research topics: Incremental dialogue, towards open-domain systems, end-to-end neural network architectures
Content of the exercise:
Students will be provided with a set of Python tools which will enable them to configure and test a simple spoken dialogue system. They will be asked to implement a simple dialogue state tracker and a reinforcementlearning algorithm and optimize the dialogue manager in interaction with a simulated user. This will give them an opportunity to explore a practical example of reinforcement learning.