About the speaker

Erik Arakelyan is a Machine Learning researcher with a Ph.D. in Machine Learning at The University of Copenhagen (UCPH), specializing in topics of Neuro-symbolic AI, query answering over Knowledge Graphs, Large Language Models, and Explainability in NLP. Erik was a visiting researcher at Amazon working on grounded LLM reasoning and knowledge-based completion at Alexa AI and Amazon AGI. Creating scalable pipelines for experimentation and inference over KGs and researching multi-hop reasoning with retrieval-based augmentation over LLMs.

About the talk

During the talk, Erik will explain ways to improve large language models (LLMs)' reasoning, or "thinking," through problems to give accurate answers.

We’ll focus on key areas:

- ensuring models are faithful (sticking closely to facts),

- verifiable (answers can be checked), and

- logical (making sense).

Starting with what we mean by "reasoning" for LLMs, we’ll look at how to train and guide models to make smart, context-aware responses.

Agenda

6:30 PM Registration

7:00 PM Talk and Q&A

8:15 PM - 8:45 PM Networking