ISP Forum Spring 2021

2 minute read

Published:

Date: 2021-01-22

Title: Neurobiology of Language

Speaker 1: Steven Small

Summary and thoughts:

Dr. Small presented his talk on the neurobiology of language and the different ways in which language operates in the real world and in our brain. He has dedicated his recent career to establishing the growing field of neurobiology of language through papers, books, and conferences. I found both the topic and content of his talk to be fascinating as he explained how the perception of language in our brain is not isolated in a single region or part of the brain - each region connects to the other to combine and give us the language system we know and use today. His theories, definitions and fallacies provide context to this complicated field with a philosophical bent. Moreover, the guiding principles focused on human language and its perception made the talk very interesting from both a computation and neurological perspective (through various diseases like ASL, speech disorders). I can the the potential of the field to extend to modeling neural networks (as the initial networks were based on neurons in the human brain).

Title: Clinician-Focused machine learning

Speaker 2: Harry Hochheiser

Summary and thoughts:

Dr. Harry Hochheiser from DBMI presented on clinician-focused machine learning (ML) which was a pleasant variation from listening to talks about patient-focused machine learning only. While we often talk about the users of decision support systems, it is good to keep in mind the very different and extremely skilled users (physicians and nurses) that these systems are designed for. This follows the understanding, then, that in any study to evaluate the system or ML model, the perspective of the clinician is critical. Through examples of racial bias in utilization of healthcare and recommendations of AI of standard and non-standard care, Dr. Hochheiser brough to light the inherent challenges in good evaluation of AI models from the physician’s and patient’s perspectives. His criteria for evaluation seemed quite sound - for a model to be accurate, adopted and appropriately impactful. I believe there is also an element of AI and ethics that must be considered in the development and evaluation of these models to better understand and mitigate the biases in the models. He further presented projects that attempt to follow good practices for evaluation by involving clinicians till the finish line (from conception to deployment) through dedicated focus groups, user studies in real-time and presenting incorrect information directly to quantify the bias produced by the ML models.