̽��ֱ��

��

I was the first international doctoral student to be funded by the Turing Institute, the UK’s national institute for artificial intelligence and data science. I’m currently pursuing my PhD in probabilistic machine learning; my research interests include Gaussian processes and their applications to contemporary physical sciences like collider physics and astronomy. Before coming back to school in 2016, I was a quantitative high-frequency trader at Citadel LLC, a Chicago-based hedge fund.

I had a ‘tiger mother’ who believed mathematical proficiency was sine qua non for success in real life. Growing up in Madras, school was super competitive and you had to be really good to stand out. I remember being quite good at math and shying away from the humanities. It is interesting how some things don’t change!

I worked in the City for five years before coming back to school at Cambridge. While my job provided scope for mathematical work, it was an intellectual straitjacket. I missed the undercurrents and freedom of academia. Upon receiving my offer letter and to the astonishment of friends, family & my boss at the time, I reluctantly left the bourgeois trappings of London finance. Looking back I can connect the dots but it seemed like an abrupt transition then.

Machine learning is frequently confused with automation: it can be used to achieve automation but is not the same thing. It can be used for making predictions or decisions but that is the end result of the learning process, and should not be confused with the conceptual meaning of a learning algorithm. For instance, if the task is to cluster particle decay signatures into groups which share similar properties, you could either set out to do it by specifying in a computer program how exactly to look for them and how many to look for in an explicit instruction set – this would be classical programming, or, you could use a learning algorithm that encodes a certain belief about what constitutes a cluster and when it encounters data in a process called ‘t�������Ծ��Բ�’, it develops the ability to identify them without an explicit instruction set. ̽��ֱ��former is deeply limited in its ability to discover complex structure encountered in real-world data and the latter is paradigm defying and powerful.

My research is specifically in Bayesian non-parametrics, a subfield of machine learning that allows a user to stipulate a prediction in terms of a probability distribution rather than a point estimate, providing a sense of confidence in the predictions. ̽��ֱ��model’s complexity is dynamically calibrated as it sees more data. For example, in the clustering task, new clusters would be created on the fly if more data comes in which does not fit any of the existing clusters.

Modern machine learning has the ability to transform the physical sciences. ̽��ֱ��intuitive and (often) deterministic models of systems are being replaced by abstract models of 'data'. In high energy physics, for instance, the discovery of new and exotic particles is largely a statistical problem. Machine learning is often the chosen framework to parse large volumes of high dimensional data with the aim of capturing a hidden or latent structure that would evade classical analysis. Machine learning is becoming the lynchpin rather than something ancillary to the scientific process. I think scientists everywhere are waking up to this.

I feel completely at home in Cambridge, both the city and the institution. I’ve had a great experience as a graduate student; what I like most is there are very few rules. For the most part, you can define your own pace and own work sometimes crisscrossing different departments. ̽��ֱ��opportunities for learning are limitless, I frequently attend undergraduate lectures, sometimes just to relearn things. One can embrace College life as little and or as much as one wishes to. I have two supervisors: Dr Chris Lester is a high energy physicist who introduced me to the world of collider physics; a field prime for machine learning. Professor Carl Rasmussen is a world-leading expert on Gaussian processes, I sometimes forget how much of a privilege it is to be working with him.

Science isn’t formulaic like other professions, there is something more to it than sheer hard work. You have to really love it and embrace it without fear. Setbacks and failure are par for the course, but nothing is permanent. ̽��ֱ��point is to keep moving forward even if we are far from where we want to be.

I enjoy science because it refines how I think about everything else. When you are a researcher, some of the traits that come with performing research tend to permeate many other aspects of your life. On the whole, that is a positive thing. You also tend to get comfortable with complexity and abstraction, where most people would run away from it.

It’s true that women face an uphill battle: there are entrenched social norms and they have to resist the urge to quit because they believe they can’t compete or will never acquire the skills fast enough. That is a fallacy. Many people ask me if I face invisible or unconscious prejudice because of my gender or race. I do not know the answer to this and I choose not to carry that burden. I have embraced life in Britain because I believe in its meritocracy, but I know that we all have our struggles, and they can’t be conquered overnight. I won’t compromise on what I love to do because fewer women choose to do it or out of the fear that it is not an even playing field. I like the words of Indra Nooyi - “when you love something - throw your head, heart and hands into it. Be so brilliant that you cannot be ignored. There is no other way.”

You can follow her on Twitter