Saravanan Baskaran, from Full-Stack Development to Implementing AI Research Paper
Saravanan implemented NIPS 2017 paper “Detecting Hate Speech in Social Media”, and is a winner of the Global NIPS Paper Implementation Challenge. See his code implementation here.
Tell us a little about yourself?
I am an engineer with 10+ years of experience developing OSS/BSS systems in the Telco domain. I picked up Machine Learning out of personal interest — mostly learning from MOOCS and developing pet projects. I recently enrolled in Udacity ML Nanodegree and have been learning a lot from lectures and feedback from staff on projects.
How did you get started in AI?
I started with enrolling in Andrew Ng’s Machine Learning Course. I also volunteered to work with Grad students from the National Institute of Advanced Studies (NIAS) who were studying the migration of people across India, by scraping data from online sources and doing simple analysis and graphs.
What are you most passionate about in AI?
I am a tinkerer and like to explore and learn new things. Machine learning and AI techniques are the perfect tools to make sense of the world around me. I would like to build intelligent systems that complement and enhance human intelligence.
I am very interested to make sense of a large amount of both the visual and textual content — I started with text analysis of tweets. I soon found that the majority of tweets have images and started to use Convolutional Neural Networks (CNNs). My interests in AI are mostly CNNs and NLP, both of which are still new to me and I definitely have a lot to work on.
Can you give us an overview of your implementation in the Challenge?
The implementation on paper “Detecting Hate Speech in Social Media” consists of various feature engineering techniques like bigrams, skipgrams, char n-grams to extract information and evaluating performance to find the one which works best. An Oracle classifier is constructed to establish the max bound.
Were there any challenges while implementing your selected paper?
I faced a few challenges in understanding the data and in preprocessing and sampling it. I reached out to the author, Shervin Malmasi, and he’d very generously helped me with the details. For example, I got a very different result to the paper on cross validation. I was advised to use StratifiedKFold since the data was not balanced and that fixed it.
What’s next for you in your work?
I am taking the lessons learned here to build a requirement search and classification system at work. It is a system that attempts to find similar requirements to give more context and bigger picture on how a particular component or enhancement is impacting the bigger system. It does this by finding related requirements (requirements for the same feature from previous iterations and related requirements). That will keep me busy!
Saravanan is a Senior System Engineer at IBM India. To keep up to date with Saravanan, check out his Github.
This is a feature of the winner of the Global NIPS Paper Implementation Challenge. You can read other winners’ feature here. Let us know if you enjoyed this series and would like to see more of content like this, drop us a comment or an email at info@nurture.ai
