Seminar on Internet Technologies (Summer 2026): Difference between revisions
| Line 92: | Line 92: | ||
| Vision-language models & Large language model | | Vision-language models & Large language model | ||
| [Haihan Zhang, haihan.zhang@cs.uni-goettingen.de] | | [Haihan Zhang, haihan.zhang@cs.uni-goettingen.de] | ||
| | |||
| Yes | |||
|- | |||
|- | |||
| CoT Compression for Efficient Reasoning | |||
| In this topic, you will study methods to improve the efficiency of Chain-of-Thought (CoT) reasoning in large language models by reducing redundant or verbose reasoning steps. This includes techniques such as summarization-based compression, iterative reasoning, and latent reasoning representations. You will explore how compression affects reasoning accuracy, computational cost, and attention mechanisms, and implement approaches to balance efficiency and performance. | |||
| Large Language Models & NLP (familiarity with Transformer architecture is recommended) | |||
| [Hao Xu, hao.xu@cs.uni-goettingen.de] | |||
| | | | ||
| Yes | | Yes | ||