A no nonsense approach to building an Active learning pipeline - A road to self supervised learning

Vishal Srinivas (~Vi-Sri)


Description:

The key idea behind this talk on active learning is that a machine learning algorithm can achieve greater accuracy with fewer training labels if it is allowed to choose the data from which it learns. An active learner may pose queries, usually in the form of unlabeled data instances to be labeled by a human annotator. Active learning is well-motivated in many modern machine learning and deep learning problems, where unlabeled data may be abundant or easily obtained, but labels are difficult, time-consuming, or expensive to obtain.

This talk provides a general introduction to active learning and a survey of active learning and slowly progresses to rigorous implementation of the algorithms into building your own active learning framework. This also paves way into exploration of Self supervised problems in deep learning. This talks includes discussion of the scenarios in which queries can be formulated, and an overview of the query strategy frameworks proposed in the literature to date. An analysis of the empirical and theoretical evidence for successful active learning, a summary of problem setting variants and practical issues, and a discussion of related topics in machine learning research are also presented.

Timeline of the talk :

  • Active learning introduction with information theory - 3 Minutes
  • Scenarios - 3 minutes
  • Querying strategies and Design problems - 15 minutes
  • Demo of our Active learning framework - 3 minutes
  • Related ML problems ( special focus on self supervised learning ) - 4 minutes
  • Q&A - 2 minutes

Prerequisites:

  • The user is expected to have intermediate level understanding on college level statistics and calculus.
  • knowledge on Data-science and machine learning libraries like Pandas, Scikit-learn would be beneficial
  • Knowledge on any one of the deep learning libraries ( although implementations will be focussed on Tensorflow )
  • Exposure to production problems in ML would help the user connect to the motivation of the topic better

Video URL:

Youtube Video link

Content URLs:

Speaker Info:

I am Srinivas, Working as a machine learning engineer in Toyota Connected India. Focussed on Computer vision, I have started my career as a software engineer writing scalable multi agent systems and shortly after which i switched to writing vision firmwares for Camera ISPs, from then on i had fast lane of 4+ years in the field of computer vision and here i am in one of the largest automotive conglomerates in the world and the journey has been nothing but full of amazement and learning. I love math and specifically the math of uncertainity. My research interests are in transformative applications involving capabilities of computer vision to operate and learn from an environment through perception interventions, incorporating the formalism of causality for out-of-distribution generalisation, data and sample-efficient learning, and counterfactual reasoning.

Section: Data Science, Machine Learning and AI
Type: Talks
Target Audience: Intermediate
Last Updated: