Genetic Programming/Auto-ML for One-Shot Learning

ABSTRACT - Are computers able to learn and classify from only one sample just like humans do? Will they able to categorise in ways that are mostly indistinguishable from people? A model example of such capabilities is the one-shot learning method, which correctly makes predictions by using only a single example of each new class. In this paper, I explore a learning method based on Genetic Programming which makes use of TPOT, a Python Automated Machine Learning tool. The One-Shot learning approach in this work will concern characters recognition. By having a small set of character samples, it will be able to recognise whether pairs of random characters written by different people are actually the same character, or not. Similar efforts have been done in the past, which only few of them are comparable with humans. The aim of this project, therefore, will be trying to simulate as much as possible humans' capabilities.

Date: March 2018 -> April 2018

Technologies Used: Python3 | Pandas | Numpy | Skimage | Scikit-Learn | TPOT | Omniglot

Report: download

GitHub: One-Shot Learning