"LLMs for Python Engineers: Advanced Data Analysis and Semantic Kernel",Oleks...
Deep learning intro
1. Everything you’ve always wanted
to know about deep learning*
(*but were afraid to ask**)
(**in 15 minutes or less)
Andrew Beam, PhD
DBMI Postdoc happy hour
05/20/2016
e-mail: Andrew_Beam@hms.harvard.edu
twitter: @AndrewLBeam
4. Neural Net History
Slide credit: http://www.slideshare.net/deview/251-implementing-deep-learning-using-cu-dnn
5. Neural Networks
Neural Networks are a flexible family of machine learning
algorithms.
• Not just one algorithm/model (MLP, RNN, CNN, etc)
• Paired with (relatively) generic optimizers that “work” in a lot of
scenarios
• Trained using GPUs = big models on big data
• Make it sound like you’re doing cool, cutting-edge neuroscience
6. Neural Nets
Given data matrix (X), model response (y) compositionally:
y = f3(f2(f1(X)))
where fi() takes a weighted combination of inputs,
followed by a nonlinear transformation
For example:
f1(x) = max(0,x*w)