Author: akshatamohan

Revisiting Fundamentals: Exploring Analytics with Professor Joydeep Ghosh

by akshatamohan

My classmates and I were fortunate to have the Advanced Predictive Modeling course taught by Dr. Joydeep Ghosh, where we learn more about the mathematics behind data mining algorithms to better understand intuition behind these algorithms. I was lucky to interview Dr. Ghosh for our blog and here’s a short post –

1. What are the research topics that interest you most now?

I am very interested in analyzing healthcare data in order to model the evolution of a person’s health trajectory, predict the costs
associated with individualized healthcare and for treating a person with personalized medicine. This involves building statistical models
at the individual level using large, heterogeneous data, which is very challenging and tricky.  This is why I find myself spending most of my time on addressing this problem. There is a great potential for analytics in healthcare which makes up almost 1/5th of the economy.

2. Could you talk a little about yourself and what got you into this field?

While doing my PhD in parallel computing, I designed  a “paper” computer with 64,000 processors, and was studying how such a super machine could be used to simulate the human brain. Fortuitously, the first  IEEE Conference on Neural Networks was held during that period (1987), so I decided to attend it and realized that brian-like models can be used for a variety of pattern recognition tasks.  This intrigued me even more and I knew my calling. After completing my PhD in 1988, I joined UT Austin as an Assistant Professor, and gradually switched my research area from parallel computing to pattern recognition and machine learning. Interestingly with the current excitement about deep learning and big data analytics, both strains from my background have become very relevant.

3. What should students in an analytics program focus on?

My advice in general to students everywhere – It’s important to  understand the concepts underlying any model or technique that you use. If you get the basics right, everything else will fall into place.
To analytics students in particular – when solving any modeling problem, think of it also in business terms — are you addressing a relevant problem and is your solution actionable? How does it affect the goals of the enterprise? Make sure you are rooted in the business issue that motivated the problem in the first place. It’s also important to be tools agnostic. Never make the mistake of learning only one set of tools and going after buzz-words without knowing what you are modeling. Tools keep evolving, and new ones are rapidly emerging in the analytics space. So generalizing from your familiar set of tools is a useful skill to learn. And, like I said, and I’ll always say – focus on the fundamentals and you’ll realize how easy it is to continually learn and keep abreast of the latest technology.

4. What are your long term wishes for the program?

I believe low-end analytic tasks will be automated in the near future. I hope for our students to be one step ahead in all aspects – business, technical and soft skills in this quickly evolving world. I hope they continue focusing on fundamentals and the mathematics behind what they are learning in the MSBA program. This is the one thing that will stay with you forever and will take you places.

With Prof. Ghosh’s wise words about focusing on fundamentals and Prof. Barua’s insightful advice on building a keen business acumen [ refer to the post from 19th Sept ’16 ], we sure know what to prioritize now! It was such an honor to be able to attend their courses this semester and learn from them. And I must mention that each one of the courses that we’ve taken have served to broaden our thinking and challenge us in ways we’ve never been challenged before. Thank you to all the Professors who’ve taught us until now! Have a great winter break y’all!

– Akshata Mohan

On “likes”, social media and beyond – exploring analytics with Professor Anitesh Barua.

by akshatamohan

This is the first part of a faculty interview series where I interview a faculty member of the MSBA program to learn more about their research, teaching and more importantly for all of us, steal some information about how to be a better data driven story teller.

 

anitesh2

1. Is it naive to say that analytics was only just discovered as quite a few believe it to be?

Discovering insights from data goes back a very long time, but it was primarily happening in academia. But there is a new realization in most businesses today that we can compete better with data and evidence based decision making. Today’s Information Technologies allow for massive quantities of diverse data as well as tools to analyze them. Text mining in the 1970s
was restricted to research and universities. But these days, due to social media and other user generated content, there is an explosion of unstructured data which are being exploited by many companies to gain deep insights into many aspects of their business.

2. How did you start out doing what you are doing now?

When I was a Ph.D student at CMU, there was a software ( quite a sophisticated one at the time ) called TETRAD, along with its theoretical underpinnings. It actually went beyond correlations, and focused on discovering causal relationships. I was fortunate to work with it in my research to demonstrate that you can discover new relationships in data. Not unexpectedly, there wer
e a lot of academics who were highly critical of the approach of data driven discovery, and dismissed it as “blind empiricism.”

3. Could you talk a little more about your research?

One that I am excited about is about the “financial value of a like”. One huge advantage of social media is that we can run real world experiments. We convinced a retailer to add like buttons on its page to increase sales. We designed an experiment where a large number of students in the treatment group were randomly instructed to like a product, and where we observed how many of their Facebook friends actually bought the product. We quantified effects of the like in two ways – those in the close contact circle of the person who liked the product (effect was stronger ) and the count of likes in popularity (this had a weaker effect ). There was a dramatic increase in sales in the treatment group, stemming primarily from “in network” effects.

Another topic of my current research involves a digital advertising supply chain. Aided by a massive data set with billions of digital impressions, we could study the decisions made by various players in this supply chain (e.g., ad agencies, publishers, brokers, etc.), and show that by accounting for cross-channel synergies, the supply chain can increase its profit by 356%. This wouldn’t be possible if we didn’t have access to such a massive quantity of data – you can theorize all you want to, but without data there is no way to validate your ideas!

4. What advice do you have for students who are starting out in this field?

Well, you guys (our MSBA students) have an incredible technical and quantitative foundation, and are being exposed to the most advanced theories and practices in these areas. I would advise you to pay close attention to the business details. All that you are learning now would be worth even more if you develop the business acumen to solve the problem in a real-world context. This is a skill that takes time to develop, but is a critical one. Read a lot of business issues in online magazines, blogs and social media in general. Be comfortable in understanding business strategy and processes. The combination of technical, quant and business skills is quite rare, and hence highly valued in the industry.

 

 

– Akshata Mohan

Data Science meetup at Indeed.com

by akshatamohan

11th August ’16 started off with some mock interviewing and career sessions at McCombs. At 5pm, four of us girls, exhausted and yet enthusiastic, shared a cab and reached the imposing Indeed.com building at 6pm. We were attending a data science meet up to meet with 30 other brilliant people in data science. As we reached the reception area, we were greeted by the Indeed folks and were ushered into the meet up room.

indeedDespite the exhausting day, we were surprised to find that we could hold intelligent “data science-y” conversations with other women in the room about their projects, work, etc.. We met an interesting medley of brilliant women working in varied backgrounds and projects some of which were assessing the structural damage progression in patients with rhematoid arthritis. It was fascinating to us because it just proved to us that analytics and statistics is indeed ubiquitous. Grabbing some food that was arranged for us, we attended interesting sessions about p values, confidence intervals and hypothesis testing. It served as a good revision for us since we had just completed our summer semester the previous week and we’d covered a lot of ground in statistics in our predictive analytics class. At the end of it, we had a couple of lighting talks by Julie Di Carlo [ http://stanford.academia.edu/JulieDicarlo ] on being cautious with interpreting the p values and by Donal Mc Mahon [ https://www.linkedin.com/in/donal-mcmahon-6107846 ] on A/B testing. They were really insightful talks on the current trends in data science. We also talked to Nikolaos Vergos [https://www.linkedin.com/in/nvergos ] who has a PhD in Physics and had just graduated from UT. He works as a data scientist at Accordion Health Company. Data scientists in Accordion Health help healthcare organizations lower their costs and improve quality outcomes with their custom analytics and precise predictions.

At the end of the three hours, we were still full of energy to go home and Google all the new stuff we learnt about. Exhaustion? I don’t think we even realized we were so exhausted. What a day to cherish.

 

– Akshata and Mengnan