Closed gwaybio closed 7 years ago
Skimming the section headings, it looks like they divide papers by network type (e.g. MLP, CNN, RNN) and then give a few examples from pharma (virtual screening) and bioinformatics in each section. They don't say too much about any individual method. I didn't get to the part about neuromorphic chips.
The paper is tough to get through - the paragraphs are disjointed and there are several sloppy errors (e.g. listing "Deep Auto-Encoder Networks" twice in the reviewed architectures in section 2. Deep Artificial Neural Networks in Pharmacology and Bioinformatics
). The paper also talks about hardware more than other reviews - something that is definitely very important, but probably not worth as significant a portion of our review.
We don't need this for the virtual screening section. Reopen if we want it for its neuromorphic computing discussion. @evancofer is this relevant to our hardware section in the Discussion?
@agitter I am not entirely sure, so I will look into it.
The concepts seem interesting, though the errors are distracting. More importantly however, I think that neuromorphic computing for deep learning (at least right now) is not as proven, supported, or as relevant as some of the other specific hardware we've covered. The limited number of benchmarks also makes it hard to compare to other hardware. Though the authors cite https://arxiv.org/abs/1604.00697 as a use of neuromorphic chips for DNN's in bioinformatics, it looks like it was a fringe focus of the cited paper (as I could not find thorough discussion of it within said citation).
@evancofer Thanks for looking it over, I'll keep the issue closed.
https://doi.org/10.3390/ijms17081313