I've been looking back at this career which is now called Data Science. When I first walked into a course on artificial neural networks, that was in 1995 where it was not at all a good career move. It was just weird topic that a very small group in computer science was really excited about. What I really liked about it from the very beginning is that through data, I felt I could relate what I was doing in computer science back to the real world to observe what happened. No networks were built back then so I couldn't do nearly as much as what deep learning capabilities are today. It did help me fall in love with data and I had avoided statistics before, but all of a sudden I really appreciated it. From there on my initial exposure, I then went and graduated with a Ph.D. in information systems from the Stern School of Business. That was in 2004 and still very much pre-big data. Doing machine learning in a business school at the time was very much the exception. It was still very much limited to computer science departments but the focus in a business school has always been more around the application and how it can be used in a business setting. As compared to the algorithmic focus that you would find in computer science and this is really the intersection that I always found most motivating is what kind of stuff actually do. I was more interested in how does it impact the world and how is it going to make it into something useful. So after my graduation, I decided to move into industrial research. I joined the IBM Research Lab and I worked there at the predictive modeling group for a good six years. In the day to day life there was a mixture of research where ultimately you were asked to publish and find patterns. I was participating in a lot of conferences and organizing them. I was also participating in early versions of cargo competitions. More than 20 years ago when people competed in their ability to build models that were very predictive. The majority of my work was applied to either IBM's internal projects. There we would support our sales teams who are the server group in terms of finding falls that compromise the trip production on to working with the global consulting arm. Where we would come in and help them to work with external clients on whatever data science or machine learning problem they had. So that was six years where I learned a lot about many different domains. Many different data sets I can do a lot of exposure on what makes data science project succeed from a political standpoint and how do you get it adopted by ultimately the stakeholders. At that time in 2010, I received an opportunity to be the chief scientist in a New-York based advertising startup and ultimately to be able to implement some of the work I did in my Ph.D. in the context of programmatic advertising. Which is the modern way where we're buying ads in auctions and real-time at the incredible skill with incredible amounts of data. I did that for good eight years before I finally found my way into the financial industry. Where again the diversity of data that you ultimately end up and engaging with on a day to day basis is incredibly broad. I have found a very nice home here at Two sigma in a science group.

I am co-heading the strategic data science school here at Two Sigma. If you think about quantitative investment pretty much everybody here is data scientist. I can't tell you what's different between me and the other 1500 people. We all have the need of enabling our own work and finding insights through the data. The only difference might be what type of data you end up touching and so on. My responsibilities here are in some ways just a little bit more long term than the immediate goal of finding good investments decision. I'm working on a little bit further looking slightly less well-defined investment strategies with a number of different groups. I'm also more collaborative in terms of working with people from very different parts of the organizations who typically in a financial institution tend to be a little bit more silent. That's just the nature of the very IP sensitive work that they do. A larger part of my work is identifying data sets. Looking at them, trying them, and to see whether they are suitable for various different uses. Also to think about what components of the data processing could benefit for maybe some machine learning for entity recognition for instance. Just the pre-processing of data takes a lot of thoughtful manipulation. I like to refer to it as the detective work you're trying to understand like where the skeletons are buried in the data set and what are the pieces that are missing that should be there. Or what are the quirks of how it was created that are as important as you're trying to generalize. In finance what I've learned is data quality is notably more important than advertising. Where the worst that you can do is showing the ad to the wrong person. A good amount of my time is working with my team on data quality. We are also interested to see what are some of the relevant investment trends and what kind of models can we build for them. Then there is a good part of my work that is really in the space of internal and external thought leadership. It includes being involved in initiatives around women in data science. There are a lot of events happening around this time of year, there are also mentorship opportunities where I or my team would participate. Internally, I am working with some of the other groups. Two Sigma also has a private investment arm. It has an eventual capital arm and so being involved there either directly doing some legion for them or maybe doing some advisory capacity for some of the companies that we have a stake in,

I'm more of an old-style machine learning person if that makes sense. I have historically been working in the exact opposite direction as the industry has gone. I started with artificial neural networks in 95 then I did my dissertation on Decision Trees in 2004. I won a large number of machine learning competitions in the years after. At IBM with logistic aggression some clearly going against the grain here to have deeply appreciated the incredible reliability and power that you can get from not necessarily the most modern and sophisticated machine learning algorithms. At Distillery in the advertising space we build very large scale logistic aggressions with two million features and sparse representation. I found that space of very high dimensional modeling with simple models to be no more interesting to me. I understood the models a lot better than some of the more modern on black boxes. I have a bit of old-style programming actually known as Pearl. I have adopted and moved on to python I think at this point Python is clearly the primary workforce that a lot of the work ends up being done both on the initial development of a prototype. Even for some of the production code we are using python for that. Algorithms broadly are mostly some form of supervised machine learning. Very rarely some form of unsupervised clustering in the pre-processing. I tend to start with simple algorithms and more thoughtful hypothesis-driven features over just throwing everything into deep learning our resentment and hoping for the best. The other piece of course is some basic skills around sequel being able to get to your data and just data manipulation on. I still do a lot of data inspecting literally on the UNIX. I do appreciate some of the original UNIX based methods to look at data because I feel I tend to get misled by graphs very easily. We often look at graphs and we see stuff that may or may not be there. There are certain things in terms of detail resolution that you lose by looking at more graphical representations. For instance seeing that there is a structure in the ID field where there shouldn't be. If the ID should be a random field that's typically not something that you find as easily in graphical, so I prefer to have initially a part with pretty basic tools.