Description:

Given enough data, lots of compute power and a clear objective, deep learning models are powerful predictive tools that can be applied to a wide range of tasks. Getting the best possible performance out of deep learning models however, can take some work. While having more data to train the network on is always beneficial, carefully chosen model architectures and hyperparameters can yield faster and better convergence.

This workshop will be a tour of the important techniques that can be used to get the best results from your deep learning models. The models will be constructed using Tensorflow and make use of Python 3.6. (Keras may also be used to speed things up if suitable)

We will review a broad list of topics in the workshop, including:

• Weight initialization strategies
• Batch Normalization
• Dropout
• Choice of optimizers
• A discussion of activation functions
• Optimizer hyperparameters
• Model hyperparameters
• Model architectures for specific tasks (CNNs, RNNs, GANs etc.)

Prerequisites:

Basic familiarity with neural networks will be a necessary prerequisite for the workshop. We will also be visualising model learning curves so previous data analysis/visualisation experience in python would be beneficial.

Ideally, someone who is attending the workshop will have worked with neural networks before and understands backpropagation and gradient descent. If you're not familiar with these topics but have a strong interest in deep learning, the following resources are great starting points:

• Michael Nielsen's Book (Great for understanding backpropagation)
• Andrew Trask's blog (neural network implementations in pure python)
• Tensorflow tutorials

Software Requirements:

• Python 3.6
• Numpy 1.12+
• Tensorflow 1.1+
• Matplotlib 2.0.2
• Seaborn 0.7+

Content URLs:

http://slides.com/sngsahil/deck#/

Speaker Info:

I'm an economics graduate who found his passion for programming and experience in economics merging together in machine learning. I've been working with Python for the past 3 years and till recently was working as a Data Scientist at Adpushup (a Microsoft Accelerator backed startup based in New Delhi).

Along with researching and implementing machine learning models, I've worked on big data applications developed using Spark and Spark Streaming.

Speaker Links:

• Blog
• LinkedIn