Workshop: Make your own DL framework

Learn to write Deep Learning Framework in Pure Python and numpy

09:00 AM to 5:00 PM, 29 July 2018, Bangalore

Scope of the workshop

This will be a theory and hands-on workshop. The theory part will be focused towards explaining the fundamentals and giving reasons for various design decisions during the implementation.

Sections of the workshop

  1. The basics.
  2. Implementing a neural Network.
  3. Computational Graph and implementing auto grad.
  4. Implementing your very own nabla framework.

What do you need to know/have a priori!

  • A laptop with good battery backup.
  • A python environment installed and ready to go (python3 and numpy latest versions).
  • Some basic knowledge in how neural networks work.

Detailed overview

The Basics

  • Parameterized models
  • Gradient Descent
  • Computational view of Neural Network
  • Arriving at backprop algorithm

Implementing a Neural network

  • Deriving gradients of a network
  • details of implementation
  • Some pitfalls and pointers
  • Why this approach cannot be used to build a framework

Computational graph and implementing auto grad

  • What is Computational Graph?
  • Automatic differentiation
  • Forward and backward accumulation

The nabla framework[coding session]

  • Implementing forward computation
  • Implementing tensors, variables and parameters
  • Implementing operations
  • Implementing forward computation
  • Implementing local gradients
  • Implementing backward computation
  • Putting it all together


Nithish Divakar

Computer Vision Research Engineer, Cogknit Semantis





L77, 15th Cross Rd, Sector 6, HSR Layout, Bengaluru, Karnataka 560102


Venue Partner