I heart Deep Learning

You know how you want to start deep learning but you don't know how? We solve that, MineTorch. We teach deep learning.

Comment

Inspiration

  • The passion to make something as complex as an open-source machine learning framework inclusive for everyone

  • To build a creative and well-implemented solution using PyTorch that could create a positive impact on people and their aspirations

  • Our goal is to make young people not only consumers but tech producers

  • Our main objectives are to teach the basics of deep learning concepts and expand programming capability via interesting projects

  • Our original project, MineTorch won third place in the 2019 PyTorch summer hackathon. We saw a need to develop a tutorial and this challenge inspired us to reunite and fulfill the need

  • Education is the future, technology is a tool to help humanity to live better, and it shouldn't be a social economics privilege

What it does

  • Our project shows how to build MineTorch, and how to develop something similar so others can teach deep learning using PyTorch

  • MineTorch is a Blockly programming platform for novice and children, with the help of a teacher to learn and integrate deep learning models into their projects

  • MineTorch is the most expressive way for children and adults to turn a graphic template (visual programming blocks) into a real Pytorch model architecture

  • In a matter of one minute students can create a PyTorch model by drag-n-drop custom Blockly blocks (a visual programming language created by Google and MIT educators)

How we built it

We used:

  • JavaScript
  • HTML
  • Python
  • PHP
  • Dart
  • Lua

and the Pytorch framework. For making the tutorial, we used Markdown.

Our process:

  1. Clone Blockly Github Repository
  2. Create javascript files for backend code
  3. Write Backend function for Block
  4. Create HTML file for UI
  5. Write HTML Code
  6. Test HTML code in Browser

Challenges we ran into

  • Understand the rules

  • Get a holistic view, looking both at rules and various recommendations

  • How to translate and build an inclusive tutorial from intermediate deep learning concepts, vocabularies to simple and easy to understand to learners and move intuitively

Accomplishments that we're proud of

  • Good flow, communication

  • Hanging in there – not giving up

  • Collaborating remotely

  • Moving fast

  • Be able to witness humanity among team members when things became difficult and challenging. Understanding and support for each other.

  • Be able to make a tutorial in spite of limited, pressed time. Being flexible and willing to change

What we learned

  • Trust the process and our work

  • The importance of good collaboration, communication, alignment, and relationship

  • Taking risks

What's next for tutorial

  • We will put effort into distributing the tutorial so people gain the knowledge of how to build MineTorch or alike

  • Accessibility. We want to see this MineTorch evolve -- never boring, but exciting!

  • Soon we aim for MineTorch to grow into something inclusive and user-friendly

Share this project:

Updates