Training a model using a support vector machine (SVM) [ talk ]
Given a binary classification problem and a dataset known to have label noise what is the best way to use a limited number of human resources to manually validate and correct the data? This is a very common problem setting in a variety of problem domains and in this talk we introduce a technique that is both theoretically well founded and produces good results in many of our datasets. The essence of the technique is to train a model using a support vector machine (SVM) and then rank the data points based on their distance to the separating hyperplane. Experimental results on a number of real world datasets will be presented and the focus will be on applying this technique to real world scenarios.Nikhil Ketkar is a Director of Engineering at Indix where he leads the machine learning team. Ex-Team lead of the data mining team in Guavus, a startup doing big data analytics in the telecom domain.
Machine Learning [ talk ]
'Machine Learning' (ML) is relevant across the vertical industry segments and ML's significance is deep rooted. Armed with delivery channels such as Cloud, Mobile etc. ML is making a massive sense towards every form of data i.e., Structured/Unstructured Text, Video, Audio, Image etc. The current session intends to share couple of practical Enterprise use-cases while helping towards a right approach to identify a right technique/algorithm (towards addressing a specific business problem).Dr G Subrahmanya VRK Rao (Dr Rao) strives to address business/technology problems by help building enterprise-class adaptive systems.
|09:45 AM - 10:00 AM||Setting up|
|10:00 AM - 10:45 AM||
talk: Training a model using a support vector machine (SVM)
|10:45 AM - 11:30 AM||
talk: Machine Learning
Dr G Subrahmanya VRK Rao
|11:30 AM - 12:00 AM||Open Discussions|
Interested in attending? Please login. You can use your existing Twitter or Google account, and if you have previously voted on a session proposal or attended a hacknight, you already have a HasGeek account.