PODCAST DESCRIPTION
Baselines, simple models offering reasonable results, are essential for contextualising more complex models and identifying areas for improvement. However, the sources highlight the problem of weak baselines in research, where comparisons against insufficiently tuned or considered baselines lead to misleadingly optimistic results. The authors advocate for stronger, well-tuned baselines and suggest community-level changes to improve research rigor and avoid inflated claims of progress. This includes procedural changes in conferences and a shift in research culture to prioritise quality over quantity.

DISCLAIMER
This NotebookLM-generated podcast is based on a manuscript by Dennis Li, Euxhen Hasanaj and Shuo Li, entitled "3 - Baselines". All rights are reserved to them, and uses of this reproduction should only be made for educational purposes.

Baselines In Machine Learning - A Critical Evaluation

Adelson de Araujo is a PhD candidate from the University of Twente. He received his MSc degree in computer science from the Federal University of Rio Grande do Norte, Brazil, where he specialized in developing AI applications for smart cities. Adelson's research projects included developing smart city applications and conversational agents for the learning sciences. In his PhD, he developed and evaluated Clair, a collaborative conversational agent to foster productive student dialogue. Adelson's recent research interests involve developing a new generation of conversational agents in education that can effectively deliver adaptive guidance to students in collaborative learning activities.