https://github.com/somtambe/omniglot-bcs

Teaching machines in comparably few shots.
https://github.com/somtambe/omniglot-bcs

cognitive-science deep-learning omniglot one-shot-learning

Last synced: 3 months ago
JSON representation

Teaching machines in comparably few shots.

• Host: GitHub
• URL: https://github.com/somtambe/omniglot-bcs
• Owner: SomTambe
• Created: 2020-06-08T10:54:45.000Z (over 5 years ago)
• Default Branch: master
• Last Pushed: 2020-07-15T06:00:08.000Z (about 5 years ago)
• Last Synced: 2025-01-10T03:11:18.803Z (9 months ago)
• Topics: cognitive-science, deep-learning, omniglot, one-shot-learning
• Language: Jupyter Notebook
• Homepage:
• Size: 88.5 MB
• Stars: 2
• Watchers: 3
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# The Omniglot Challenge [(paper)](https://arxiv.org/abs/1902.03477)

Own Model for the Omniglot Challenge.

**Brain & Cognitive Society, IIT Kanpur**

## Proposed methodology

1.  Convert all stroke data to 25-point splines. :white_check_mark:

2.  Generate a **b**-vector using a variational autoencoder for each and every such stroke. :white_check_mark:

3.  Use clustering to get number of primitives, and then turn them into vectors for each and every image in the background set using one-hot encoding. :white_check_mark: 

4.  Perform supervised learning using a Convolutional Neural Network on the images and respective vectors to get a network which maps your character images to stroke data latent vector space.**(In Progress)**

### How this last network can be further used

1.  Use the trained model for classification or one-shot learning tasks by introducing another model which maps the latent space of the image vector to desired output.

2.  Report acquired results.