Jump to content Jump to search

Master's Seminar Machine Learning

summer semester 2025

Organizational:

  • Talk: 30 Min + 10 min discussion
  • Handout: 2 pages
  • Hand in slides 2 weeks before presentation so that we can go through

 

Papers for the Machine Learning Seminar

Large Language Models:

  1. Eureka: Human-Level Reward Design via Coding Large Language Models  https://arxiv.org/abs/2310.12931
  2. Auto MC-Reward: Automated Dense Reward Design with Large Language Models for Minecraft https://arxiv.org/abs/2312.09238
  3. DeepSeek v2 https://arxiv.org/pdf/2405.04434
  4. DeepSeek v3 https://arxiv.org/abs/2412.19437
  5. DeepSeek R1 https://arxiv.org/abs/2501.12948
  6. Chain-of-Thought Prompting Elicits Reasoning in Large Language Models  https://arxiv.org/abs/2201.11903
  7. Connecting Large Language Models with Evolutionary Algorithms Yields Powerful Prompt Optimizershttps://arxiv.org/abs/2309.08532
  8. Language Models Don't Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Prompting https://arxiv.org/abs/2305.04388
  9. Large Language Models Are Human-Level Prompt Engineers  https://arxiv.org/abs/2211.01910
  10. Automatic Prompt Optimization with "Gradient Descent" and Beam Search  https://arxiv.org/abs/2305.03495
  11. Rho-1: Not All Tokens Are What You Need https://arxiv.org/abs/2404.07965
  12. s1: Simple test-time scaling h ttps://arxiv.org/abs/2501.19393
  13. Finetuned Language Models Are Zero-Shot Learners  https://arxiv.org/abs/2109.01652

 Fine-Tuning:

  1. LoRA: Low-Rank Adaptation of Large Language Models https://arxiv.org/abs/2106.09685
  2. QLoRA: Efficient Finetuning of Quantized LLMs https://arxiv.org/abs/2305.14314

 3D Deep Learning:

  1. Neural Radiance Fieldshttps://arxiv.org/abs/2003.08934
  2. pixelNeRF: Neural Radiance Fields from One or Few Images  https://arxiv.org/abs/2012.02190
  3. Putting NeRF on a Diet: Semantically Consistent Few-Shot View Synthesis  https://arxiv.org/abs/2104.00677
  4. Learned Initializations for Optimizing Coordinate-Based Neural Representations  https://arxiv.org/abs/2012.02189
  5. FreeNeRF: Improving Few-shot Neural Rendering with Free Frequency Regularization https://arxiv.org/abs/2303.07418
  6. SinNeRF: Training Neural Radiance Fields on Complex Scenes from a Single Image https://arxiv.org/abs/2204.00928
  7. 3D Gaussian Splatting for Real-Time Radiance Field Rendering https://repo-sam.inria.fr/fungraph/3d-gaussian-splatting/

 ARC Challenge:

  1. Graphs, Constraints, and Search for the Abstraction and Reasoning Corpus https://arxiv.org/abs/2210.09880
  2. LLMs and the Abstraction and Reasoning Corpus: Successes, Failures, and the Importance of Object-based Representations https://arxiv.org/abs/2305.18354
  3. Tackling the Abstraction and Reasoning Corpus (ARC) with Object-centric Models and the MDL Principle https://arxiv.org/abs/2311.00545
  4. CodeIt: Self-Improving Language Models with Prioritized Hindsight Replay https://arxiv.org/abs/2402.04858
  5. Getting 50% (SoTA) on ARC-AGI with GPT-4o https://redwoodresearch.substack.com/p/getting-50-sota-on-arc-agi-with-gpt
  6. Combining Induction and Transduction for Abstract Reasoning https://arxiv.org/abs/2411.02272
  7. The LLM ARChitect: Solving ARC-AGI Is A Matter of Perspective https://da-fr.github.io/arc-prize-2024/the_architects.pdf
  8. The Surprising Effectiveness of Test-Time Training for Abstract Reasoning https://arxiv.org/abs/2411.07279
  9. CompressARC: https://iliao2345.github.io/blog_posts/arc_agi_without_pretraining/arc_agi_without_pretraining.html
  10. Searching Latent Program Spaces https://arxiv.org/abs/2411.08706

 Interesting facts about Neural Networks:

  1. Intriguing Properties of Neural Networks: h ttps://arxiv.org/abs/1312.6199
  2. The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks: https://arxiv.org/abs/1803.03635
  3. Neural Tangent Kernel: https://arxiv.org/abs/1806.07572

 Other:

  1. Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction: https://arxiv.org/abs/2404.02905
  2. Group Equivariant Convolutional Networks https://arxiv.org/abs/1602.07576
  3. Deep Sets https://arxiv.org/abs/1703.06114
  4. Video PreTraining (VPT): Learning to Act by Watching Unlabeled Online Videos https://arxiv.org/abs/2206.11795
  5. Ghost in the Minecraft: Generally Capable Agents for Open-World Environments via Large Language Models with Text-based Knowledge and Memory https://arxiv.org/abs/2305.17144
  6. Voyager: An Open-Ended Embodied Agent with Large Language Models  https://arxiv.org/abs/2305.16291
  7. First return, then explore https://arxiv.org/abs/2004.12919
  8. Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning Algorithm https://arxiv.org/abs/1712.01815
  9. Discovering faster matrix multiplication algorithms with reinforcement learning  https://www.nature.com/articles/s41586-022-05172-4