Attendance for this event is restricted to members who identify as women or non-binary. Our Code of Conduct is available online & applies to all our spaces, both online & off.
https://github.com/WiMLDS/starter-kit/wiki/Code-of-conduct ---
This is an intermediate level deep learning talk. We will focus on Convolutional Neural networks (CNNs) using Python & pyTorch. First we will review the components of a CNN & discuss why these networks work so well for computer vision tasks. Then we will compare two styles of CNN implemented in pyTorch on CIFAR10; a VGG-like network [1] & a Residual Network. [2]
This event assumes a basic understanding of neural networks, backpropagation, & neural network training through stochastic gradient descent.
Topics covered:
- Introduction to convolutional neural networks
- Motivation & biological inspiration
- Kernels
- Padding, stride
- Walk through of a basic CNN in pytorch
- Comparison of a VGG-like & Residual Network architecture on CIFAR 10 in pytorch Reading Preparation (optional)
No preparation is required but if you do not have any background on neural networks then here are a few resources Slides:
WiMLDS Deep Learning Workshop
https://github.com/lgraesser/Neural-Networks-Workshop-Materials-WiMLDS
Reading:
Intro to Neural Nets (1 - 1.5 hour read) https://learningmachinelearning.org/2016/07/26/introduction-to-neural-networks/
Regularization (30 min read)
https://learningmachinelearning.org/2016/08/01/regularization-for-neural-networks/
Neural Networks & Deep Learning book by Michael Nielsen
http://neuralnetworksanddeeplearning.com/
Videos:
20 hours of videos from the Bay Area Deep Learning School
Day 1: https://www.youtube.com/watch?v=eyovmAtoUx0
55 minute deep learning intro (if you have access to safari) https://www.safaribooksonline.com/library/view/introduction-to-deep/9781491999608/
pyTorch:
Here is a great whistle stop tour.
http://pytorch.org/tutorials/beginner/deep_learning_60min_blitz.html
About the Speaker
Laura Graesser is studying for an MS in computer science at NYU, focusing on machine learning. Laura is particularly interested in neural networks & their application to computer vision, NLP & reinforcement learning. Most recently, Laura is interested in combining reinforcement learning with supervised learning, knowledge distillation, & in tackling multi-modal & multi-task learning.
In her spare time, Laura enjoys dancing, listening to jazz, going to art exhibitions, & writing about machine learning. Twitter: @lgraesser3
Github: lgraesser
LinkedIn: Laura Graesser
What to bring
Important to know