MIDL Vision 2021

be a lesson for younger researchers out there. “ Two of the datasets that we used in our work are from the M&Ms Challenge , and I didn’t realize until quite late in the day that we could ask the authors for additional details of this data, ” she reveals. “ In the end, we did this as part of the rebuttal process at the review stage. I contacted the authors, and Victor Campello from the challenge responded quickly and gave me lots of information. The actual pieces of equipment that had been used for extracting the data didn’t appear on the website of the challenge. It made the paper so much better! I felt a little awkward at first, but I’m more confident now and I’ve encountered other situations since where I have needed to contact people responsible for datasets, so I feel more comfortable doing it. Everyone is friendly in the community! ” Looking ahead, Camila wants to implement different self-supervised tasks . She would like to know how good the detection signal from the self-supervised loss is by a wide array of proxy tasks. Another desire is to have different uncertainty estimation methods. This work uses Deep Ensembles and Monte Carlo Dropout , and Camila tells us they are extending that for TreeNets – architectures where you have one main body and many different heads. She would also like models that include some variational inference components . We ask Camila, if she had a magic wand, which one of the above would she add to her model right now? 12 Presentation VISION MIDL

RkJQdWJsaXNoZXIy NTc3NzU=