IIT Madras student chosen for Google PhD Fellowship | The Hindu - Education

Preksha Nema shares what it took for her to become a Google PhD Fellow. 

Photo: The Hindu

IIT-Madras student Preksha Nema is one of four recipients in the country, of the Google PhD Fellowship 2017. This programme recognises graduate students doing “exceptional work in computer science and related research areas.” Besides connecting the chosen Fellows to a Google Research Mentor, the programme provides monetary support and a stipend.

Preksha is jointly guided by Professor Mitesh M. Khapra and Professor B. Ravindran at the Department of Computer Science and Engineering. Under their supervision, she does research in machine learning for natural language processing (NLP), specifically, a machine learning method known as deep learning. It is this exciting area that is responsible for a variety of applications such as Apple’s Siri, Amazon’s Alexa, chatbots, and personalised music or shopping recommendations. 

Excerpts from an interview with Preksha.

Beginnings
I really liked programming from the time it was introduced in school. It makes you think step by step on how to reach a solution; it helps you enumerate your thoughts. It is like solving puzzles, which I always liked.

Roadblocks
Quitting my job, and returning to higher education after working for three years was a bit worrisome. I was not sure if I would be able to perform well. But once the coursework started, I got accustomed to the routine. Luckily, I have a strong support system comprising my advisors, family, and friends. The dedication and patience with which my father does any work always amazes me. I keep trying to be like him.

Research
I work in deep learning for NLP. The broader goal here is to make machines understand and generate semantically and syntactically correct sentences in English (or any other natural language). Google Translate and Google Search are some common applications of NLP.

My initial PhD work introduced a novel method for computers to generate human-like summaries of an article. For example, a trained machine can read a document on biogas and then generate a summary based on a given query. For example, “What are the environmental benefits of using biogas?”

Now, I am focussing on Q&A systems; we would like the system to accurately answer questions based on some contextual information. We are also exploring ideas to get the system itself to generate meaningful questions.
Read more... 

Source: The Hindu