Speaker: Veronika Cheplygina, IT University of Copenhagen

Title: Incentives for machine learning in healthcare

Machine learning has many promises for healthcare, however real-world progress” can be limited due to incentives in the field. First, there is a drive for larger and larger datasets, which comes at a trade-off with dataset quality, and ultimately leads to inaccurate or biased outcomes for patients. Second, algorithm development is regarded as a more prestigious activity, which leads to many more algorithms being introduced than datasets, and poor validation practices. This hurts our ability to generalize about which algorithms are good when, since the space of results (dataset x method combinations) becomes emptier over time. Additionally, introduced algorithms might be too similar, leading to diminishing returns where each new algorithm has less added value), as well as opportunity cost since other important problems are being overlooked. Finally, these issues lead to research waste, both in terms of environmental impact, and in terms of the research community.

Website: https://purrlab.github.io/

Host: Surabhi Verma

Organisers: Surabhi Verma and Hartanto Wong