These days, if you want an undergraduate degree in Deep Learning, you’ll need to take courses in applied math, computer science, and possibly much more. In other words, to be seriously legit, a degree in Deep Learning is a highly interdisciplinary one. And that’s without any domain-specific knowledge of data – e.g., economic, retail, sales/marketing, science, … With all the hype around Deep Learning, you might have the impression that you must head back to school to acquire the credentials to re-invent yourself for this significant opportunity. And while it is not my intention to disparage education in any way, it may be that your creds are a gap analysis and only a few ‘courses’ away.
I’ll illustrate using myself as an example. In my case, I acquired an undergraduate degree in Solid Earth Geophysics from Montreal’s McGill University in … well, let’s just say it was the last century. Geophysics is highly interdisciplinary; and from my undergraduate degree, I can leverage my background to address Goodfellow et al.’s applied math basics for Deep Learning as follows:
My purpose in suggesting how I’m finding myself able to leverage my background in applied maths is not to dismiss this as a requirement easily addressed, or even to make myself feel good; rather, my purpose is to stress the importance of such a background – as do others (e.g., director of AI Research at Facebook and NYU professor Yann LeCun). Why? Owing to the meteoric hype around Deep Learning, the field is attracting a tremendous amount of attention – especially from those with a seemingly brilliant idea for a startup. Whereas I have no intention to detract from that enthusiasm, I do think it’s important to leverage the message of increasingly proven specialists focused on the matter of learning about Deep Learning and that is, simply, at some point your knowledge of applied maths will be important.
Rather than wait for these math gaps to catch you at an inopportune time, I suggest a more proactive approach – even if it only means you have a raised level of awareness of your strengths and weaknesses in this foundational area of Machine Learning and then, Deep Learning. Stated differently, without the applied-maths background, you’ll only have a superficial understanding of ML and DL; with this background, you’ll have already mathematically internalized Bayesian stats, SGD, and other Machine Learning basics, so you’ll be far better placed to handle the existing practices and frenetically developing frontiers of Deep Learning.
In my case, I’m feeling increasingly confident that my applied maths background has well prepared me for various initiatives regarding Deep Learning. In part, I have validated this by working through Goodfellow et al.’s textbook – a comprehensive introduction to the field that appears to well target someone with my interdisciplinary background in the physical sciences. With these prerequisites addressed, I can pragmatically focus my attention on the Deep Learning practices and research that are salient to my own projects. Currently, for me, that is the application of Natural Language Processing to data I’m extracting from Twitter for a science-driven interest – something you can find out more about once the presentations from the recent GTC 2017 San Jose event become available.
As you ponder your own challenges and opportunities in Deep Learning, I hope you’ll similarly be able to tap into your background, and maybe even be reminded that attending university isn’t about job training – a point too often lost on today’s undergraduates. An aspect of education at the university level is learning how to learn; and if we’re to be successful in our endeavours with Deep Learning, it’s clear that we all have a lot of learning to do – ourselves, as well as our algorithmic implementations. Perhaps NVIDIA does have it right: I am AI.