Description:

Machine Intelligence algorithms, in their application to real world problems, are largely models trained in a supervised manner. Hence, they are hindered by the reality that in most practical situations unlabelled data is easier to come across and obtaining appropriately annotated and labelled data may be prohibitively expensive. Herein lies the appeal of semi-supervised learning algorithms that allow us to draw inferences with only a few labelled data samples existing among a vast amount of unlabelled data. In this talk. through the application of a variation of the tried and tested SVM, called the S3VM(Semi Supervised SVM) on standard dense and sparse data sets, we will explore the merits and demerits of semi-supervised learning.

We will also take a cursory look at a few approaches used to solve the modified optimisation problem that arises when we adapt the SVM for use in a semi-supervised setting.

The outline of the talk will broadly be the following:

1. Why Semi-Supervised Learning
2. Advantages of using Semi-Supervised algorithms rather than Supervised algorithms on limited data
3. Approaches to Semi-Supervised Learning: Transduction vs Induction+Deduction
4. Modifying the SVM for Semi-Supervised Learning
5. Approaches for solving the modified SVM: Label-switching vs deterministic annealing
6. Semi-Supervised Learning is not a silver bullet: Discussion of disadvantages

Prerequisites:

• Familiarity with Python Programming
• Minimal proficiency in Optimisation Methods
• Intermediate proficiency in Support Vector Machines

Content URLs:

Slides

Talk Specific Slides On Their Way

References

• QN-S3VM Python Package: http://www.fabiangieseke.de/index.php/code/qns3vm
• Semisupervised Learn Python Package: https://github.com/tmadl/semisup-learn
• S3VM Seminal Work: https://papers.nips.cc/paper/1582-semi-supervised-support-vector-machines.pdf

Speaker Info:

I'm Indraneil Paul, a final year Computer Science student at IIIT Hyderabad. I have been involved in machine learning, computer vision and mathematical optimisation for the best part of the past three years due to my research work. I was previously working in the Computer Vision lab on an autonomous driving project and am currently working on applying graph based machine learning models to social networks. I was also a Google Summer of Code '17 student under electric vehicle startup Green Navigation (now nav-e).

I occasionally foray into experimentation with Blockchain technology with Hyperledger.

Speaker Links:

Github: https://github.com/iNeil77