Master Seminar: Machine Learning — SS 2025
Master Seminar: Machine Learning, Summer Semester 2025
Talk duration: 30 minutes plus 10 minutes for discussion
Handout: 2 pages required
Slides submission: Must be submitted 2 weeks before presentation for review
Paper Topics
Large Language Models
- Reward design
- Model architectures (DeepSeek v2/v3/R1)
- Reasoning techniques
- Prompt optimization
- Training approaches
- Chain-of-thought prompting
- Prompt engineering
- Token efficiency
Fine-Tuning
- LoRA (Low-Rank Adaptation)
- QLoRA (Quantized LLM fine-tuning)
3D Deep Learning
- Neural Radiance Fields variants
- Few-shot view synthesis
- Gaussian splatting
- Scene reconstruction from limited data
ARC Challenge
- Abstract reasoning
- Combining induction/transduction
- Test-time training
- Solution approaches using various methodologies
Neural Network Properties
- Adversarial properties
- Lottery ticket hypothesis
- Neural tangent kernels
Other Topics
- Visual generation
- Group equivariance
- Set operations
- Video pretraining
- Embodied agents
- Reinforcement learning
- Algorithmic discovery