How algorithms could bring empathy back to medicine | Nature
Deep Medicine summarizes hype and threat, then takes us to a place where no one else has gone: a future in which AI helps to re-establish empathy and trust between doctors and patients. Dr. Eric Topol’s thesis, expressed in the subtitle, is that AI “can make healthcare human again”. Can AI reverse this trend and heal the doctor–patient relationship? Or will it exacerbate the problems of technology, but with devices that replace more humans and destroy privacy? Topol takes an optimistic view. (A book review on Deep Medicine: How Artificial Intelligence Can Make Healthcare Human Again)
Best practices for benchmarking germline small-variant calls in human genomes | Nature Biotechnology
Standardized benchmarking approaches are required to assess the accuracy of variants called from sequence data. Here, as part of the Global Alliance for Genomics and Health (GA4GH), researchers present a benchmarking framework for variant calling.
Fixup Initialization: Residual Learning Without Normalization | arXiv
Normalization layers are a staple in state-of-the-art deep neural network architectures. They are widely believed to stabilize training, enable higher learning rate, accelerate convergence and improve generalization, though the reason for their effectiveness is still an active research topic. In this work, researchers challenge the commonly-held beliefs by showing that none of the perceived benefits is unique to normalization. Specifically, they propose fixed-update initialization (Fixup), an initialization motivated by solving the exploding and vanishing gradient problem at the beginning of training via properly rescaling a standard initialization.