KIUA2004 Deep Reinforcement Learning & Neural Networks
- Course codeKIUA2004
- Number of credits10
- Teaching semester2026 Autumn
- Language of instructionNorwegian/English
- CampusHamar
- Required prerequisite knowledge
Recommended: KIUA2003 High Performance Computing with Big Data
The course will provide students with advanced knowledge in deep reinforcement learning (RL) and covers topics such as deep Q-networks, policy gradients, stakeholder-critical models, ethical considerations and societal impacts. Students will develop practical skills in implementing deep reinforcement learning algorithms using neural networks, improving problem-solving capacity and mastering neural network optimisation. The course will also promote critical thinking, collaboration and ethical decision-making. Upon completion, students will be well-prepared for practical applications in artificial intelligence and able to make informed, ethical decisions and contribute to the advancement of the field.
- Deep reinforcement learning (RL)
- deep Q-networks, policy gradients, stakeholder-critical methods, trade-offs for exploration-exploitation, multi-agent RL, model-based RL, transfer learning and meta-reinforcement learning
- Neural network architectures for RL
- feedforward, convolutional and recurring networks with an understanding of network initialisation, regularisation techniques, batch normalisation and how to solve common problems such as overfitting, vanishing/exploding gradients and catastrophic forgetting
Learning Outcome
Upon successfully passing the course, students will have achieved the following learning outcomes:
The student will have
- knowledge of deep reinforcement learning (RL) concepts
- knowledge of neural network architectures used in deep RL
- knowledge of certainty, fairness and interpretations in deep RL and the broader societal impact of deep RL in various domains
The student will be able to
- implement advanced deep reinforcement learning algorithms using neural networks to learn to train, evaluate and fine-tune models, analyse performance and understand the impact of different hyperparameters and architectures
- use advanced deep RL algorithms to solve complex reinforcement learning tasks and to learn to design and adapt algorithms for various scenarios, analyse trade-offs between exploration and exploitation and optimise performance in different environments
- analyse and optimise neural network architectures for reinforcement learning using techniques for network initialisation, regularisation and management of common challenges, such as “overfitting” and “vanishing/exploding” gradients in order to ensure stable and effective learning
The student will be able to
- evaluate and compare different advanced deep RL algorithms and approaches in order to develop the ability to critically assess strengths, limitations and trade-offs associated with different techniques in different contexts
- effectively communicate an understanding of advanced deep RL concepts through presentations, share insights and collaborate on problem-solving and algorithm design in projects
- promote a research-oriented mindset and explore and contribute to progress in advanced deep reinforcement learning, propose innovative solutions and conduct experiments to push the boundaries of deep RL
The course comprises a combination of lectures, practical exercises, independent study and academic supervision.
- 2 individual assignments
Compulsory coursework requirements that have been passed are valid for 12 months only. Students wishing to take examinations after 12 months must pass the compulsory coursework requirements again in connection with the next scheduled delivery of the course.
| Form of assessment | Grading scale | Grouping | Duration of assessment | Support materials | Proportion | Comment |
|---|---|---|---|---|---|---|
Written assignment | ECTS - A-F | Individual | 100 |
- 1 individual project-based assignment
The assignment is assessed using a grading scale from A-F, where E is the lowest passing grade.
Students are able to choose which language to use for their examination. The available options are Norwegian Bokmål, Nynorsk and English.
Permitted aids:
- Literature
- All printed and written resources
- Any use of AI-generated text and content must be clarified with the lecturer, clearly labelled and academically justified in the submission