Practical Deep Learning
Submitted by saurabh agarwal (@saurabh-agl) (proposing) on Thursday, 22 June 2017
This is a proposal requesting for someone to speak on this topic. If you’d like to speak, leave a comment.
Understanding the nuts and bolts of a Deep Learning (DL) architecture has always been a tough ride for people with a not-so-mathematical background. The goal of the workshop is to get the participants to understand the practicalities to be considered while building deep networks such as Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) by using a hands-on approach, not necessarily mathematical. This workshop is meant for people with intermediate level knowledge in Deep Learning. Intermediate = an understanding of backpropagation, stochastic gradient descent and some linear algebra.
The workshop will comprise two phases and the participants will be provided with 100$ worth AWS credits for using GPU powered instances. We will provide AMIs to use and also provide notebooks with the initial setup to get started and reduce the learning curve, The goal is to speed up the experimentation and play around with larger size networks.
A brief outline of the things that will be covered in this workshop:
A general overview into deep neural networks and how they are trained.
Intro to CNNs and exploring their power with the help of pre-trained models.
A deep dive into CNN architecture and how transfer learning can be used to build your own CNN implementations by using state of the art pre trained models and comparatively little data.
Intro to RNNs and its types.
Deepdive into using LSTM cells and its usage in different type of tasks.
(IF possible) Demo models that use CNNs for image processing and LSTMs for generating captions.
1.The participants are required to have a basic knowledge in training deep neural networks.
2.Knowledge of Python and NumPy are essential
3.The participants need a laptop with GNU/Linux or Mac OS or any UNIX based OS, or should know how to connect to EC2 from a windows machine. No assistance will be provided to windows users to overcome OS limitations.
Dr. Anand Chandrasekaran is the founder and CTO of Mad Street Den, an AI company specializing in computer vision. In addition to the academic background in the fields of neuroscience and neuromorphic engineering, he has been a member of teams working on DARPA projects in cognition and vision.