There is a sought-after type of candidate that banks, buy-side firms and fintech companies are contending to attract. And yes, it does relate to a niche within data science and artificial intelligence (AI): natural-language processing (NLP).
AI consists of machines that show intelligence resembling that of humans applied to pattern-recognition and decision-making, of which machine learning (ML), deep learning and NLP are overlying subfields.
Tejas Shastry, is a vice president of data science at GreenKey Technologies, a fintech firm offering an NLP-powered speech recognition and voice-to-data system that records and interprets quotes and trades and feeds that data into trade-ticket, reporting and compliance systems.
Shastry says that AI researchers use ML algorithms that improve at generalizing patterns through experience, similar to how humans do. If ML helps a computer to hear, then NLP helps it to understand the meaning behind what it heard.
"NLP can sometimes involve additional machine-learning algorithms, or it could be rules-based," Shastry says. "Not all data scientists work on AI, and not all AI researchers use data science day-to-day, but they are overlapping and related in that both machines and humans need math and statistics to understand the world."
Candidates with expertise in both have no worry finding a lucrative job.
Here's his advice for students who are interested in a career focused on data science, AI, ML and NLP.
If you want to be an NLP practitioner, you'll need a STEM PhD. You'll also need to be a great programmer (Shastry advises spending 10 minutes a day on pet projects like classifiers on HackerRank), including knowing how to code in Python and Go, plus deployment tools like Docker. Shastry isn't only after polished NLP practitioners: He also wants data scientists and developers who can grow to become experts in the field over time.
He inspires people to learn NLP by reading the latest academic research papers about it, a skill that most scientists have cultivated in graduate school that is essential to performing at the rapid pace necessary to succeed in AI generally and NLP specifically. That will help you to gain familiarity with common ML techniques such as classification, clustering, regression and Bayes, as well as the most common NLP tasks, which are grouped under the categories of syntax, semantics, discourse and speech.
"I read about 50 NLP papers over three weeks before I felt comfortable enough to apply bleeding-edge NLP to our work at GreenKey," Shastry says. "Throughout the process, I was experimenting with applying what I read about.
"If you're given a new paper about NLP and can figure out how to implement the techniques described, you can do NLP," he says. "You don't have to know all the machinery behind the entire field, but you have to know how to navigate around work presented by other data scientists and developers."