STEAM Academy Seminar: Towards Effective and Efficient Reasoning Language Models

by Prof. Maciej Besta (ETH Zurich)

Tuesday 28 Jul 2026, 16:00 → 17:30 Europe/Zurich

500/1-001 - Main Auditorium (CERN)

Abstract: The field of natural language processing has witnessed significant progress in recent years, with a notable focus on improving language models’ performance through innovative reasoning techniques. Among these, structure-enhanced reasoning has emerged as a very promising paradigm, with designs such as Chain-of-Thought (CoT), Tree of Thoughts (ToT), and Graph of Thoughts (GoT), in which the LLM reasoning is guided by a structure such as that of a graph. The key idea and primary advantage of such a graph-based reasoning paradigm is the ability to model the information generated/used by an LLM as an arbitrary graph, where units of information ("LLM thoughts") are vertices, and edges correspond to dependencies between these vertices. This approach enables combining arbitrary LLM thoughts into synergistic outcomes, distilling the essence of whole networks of thoughts, or enhancing thoughts using feedback loops. In this talk, we overview this very recent field, focusing on fundamental classes of harnessed structures, the representations of these structures, algorithms executed with these structures, and relationships to other parts of the generative AI pipeline such as knowledge bases, databases, verification pipelines, and others. The overview will dive deeper into several representative developments related to prompting (Graph of Thoughts), AI assistants (Knowledge Graph of Thoughts), training of Reasoning Language Models, and the underlying abstractions (Topologies of Reasoning). 

Bio: Maciej Besta leads research on large language models and broad graph computations at the Scalable Parallel Computing Lab at ETH Zurich and at the ETH Future Computing Lab; he also works on interconnects, general sparse computing, and others. Maciej published, as the main leading author, around 40 papers at top conferences and journals. He won, among others, the IEEE TCHPC Award for Excellence in High-Performance Computing Early Career (2024), the HiPEAC Tech Transfer Award (2024), the OlympusMons Award for contributions to scalable storage systems (2024), the IEEE TCSC Award for Excellence in Scalable Computing Early Career (2023), ACM Research Highlights (2018), the ACM/IEEE-CS High-Performance Computing Fellowship (2015), the first Google Fellowship in Parallel Computing (2013), and the competition for the Best Student of Poland (2012). His doctoral dissertation on irregular computations received awards from ETH (2021), IEEE (2021), SPEC (2022), and ACM (2022). Maciej also won Best Paper awards and nominations at ACM/IEEE Supercomputing 2013, 2014, 2019 (for 2 different papers), 2022, and 2023 (for 2 different papers); at ACM HPDC 2015 and 2016, and others. Finally, as an explorer working at the overlap of exploration and science, Maciej is Fellow in The Explorers Club (2022). More detailed information on: https://people.inf.ethz.ch/bestam/ .

This seminar is part of the CERN STEAM Academy Seminar Series. 

Networking cocktail will follow the seminar. 

With the support of CERN's Next Generation Triggers Project.