Saturday 11:00–11:45 in Audimax

Visual concept learning from few images

Vaibhav Singh

Audience level:


In this talk I will try to explain the concept of few shot learning and will also give a brief glimpse where current research trends are in this field. Finally I will present some practical approaches on how to train a classifier with help of GAN’s and siamese networks given few labeled examples. Slides:


Deep Learning is really good when dealing with images where conventional machine learning methodologies fell short. Still when training a deep neural network we need a lot of labeled examples unlike a human which can learn an object from even a single image. Collecting labeled images is not only cumbersome but also expensive. This can be solved either by augmenting existing images or generating new images from them. Some practical demos and code in the talk would explain these concepts.

On a different note a typical convolution neural network contains thousands of parameters. Training a class with few examples will simply overfit on the training dataset and will not generalize well. I would cover some approaches on how to regularize and learn these parameters in CNN's showcasing works from Ruslan Salakhutdinov, Hugo Larochelle.

All in all this talk will cover some of the approaches that can be used to train a Neural network based image classifier given few examples from one class. Audience will get to learn the concept of few shot learning, current research trends, common approaches to tackle the problem and finally some practical implementations.

Subscribe to Receive PyData Updates