03:08
B17 CheXpert++
-
Video
-
Table of contents
-
Video
- Description
- Transcript
- Discussion
About the talk
CheXpert++: Approximating the CheXpert labeler for Speed, Differentiability, and Probabilistic Output
00:15 Purpose
01:00 CheXpert++ Pipeline
02:10 Exploiting Inductive Bias via Active Learning
02:50 Active learning performance
About speakers
Wei-Hung Weng is a PhD graduate at MIT EECS and Computer Science and Artificial Intelligence Laboratory. Prior to MIT, he received an MMSc degree in Biomedical Informatics from Harvard Medical School/Massachusetts General Hospital and MD from Chang Gung University (Taiwan). He also worked as a physician and pathologist for years. His research focuses on creating and applying machine learning algorithms for multimodal representation learning and reinforcement learning toward better clinical decision making.
View the profilePlease note that I am not a regular user of LinkedIn (or any other social networking site). Please do not be offended if I fail to respond to posts or connection requests. By far the best way to contact me if you have something specific to communicate about is by email.
View the profileDrive to McDermott differentiability and diagnostic labels for cretex radiology report. Limited ability in certain use cases class, that is probably not from our study. One of those application, there is an active learning of the label, the most benefit for me, lately, and use that to improve overall lately. Checkered past. I'm always trained and very simple way we take unlabeled Radiology, reports protects real reports. And then we training Arbor Baychester plus size model simply to match
those rule-based labels. There's no additional disagreement between A little farther than this and actually asked permission to examine cases, were check request class, and check, Purdue disagree provided, a better answer. And surprisingly we find that check for glasses. Actually preferred over the source, check for model, almost on average, across almost all of our diseases, true, despite the fact that a lot of inductive is going on in the training effect, request class, or the use of
refrigerant convert model and conducting bias that property to improve performance. Overall, how is going on in this work? If we trained different than the normal way, then we label a collection of sentences with them and we look at those sentences sentences is most uncertain. How many almonds are training set with those newly label sentences and we reach me the model again, with this new information by hand, gold standard, testing set. We can actually show that this
Buy this talk
Ticket
Interested in topic “Artificial Intelligence and Machine Learning”?
You might be interested in videos from this event
Similar talks
Buy this video
Conference Cast
With ConferenceCast.tv, you get access to our library of the world's best conference talks.
