How A Discussion With Dad Ignited Fire In The Belly For Aswini Thota To Kickstart His ML Journey

“A couple of areas where we need to put our immediate attention to are – Explainable AI and AI governance”

Aswini Thota

Analytics India Magazine caught up with Aswini Thota, Principal Data Scientist at Bose Corporation, for the ML practitioner’s series. Aswini holds two Master’s degrees in Electrical Engineering and Data Science. He is an AI and Data Science technical leader who solves organisational and business problems leveraging data. In the past, he had served in several data leadership roles at Fidelity Investments. Aswini is also a research scholar who studies the impact of AI in improving business efficiencies. His work has been published in several scientific journals and trade magazines. 

AIM: How did your interest in Machine Learning begin? Can you share some of the unique challenges you have faced in your journey?

Aswini: I was first exposed to experimentation and research methods through my Dad. He started his MPhil when I was in 10th grade. He used to discuss his experiments, and I even remember visiting his lab a couple of times. I have always been fascinated by Math and how engineers and researchers use it to model real-world phenomena. Courses in my undergraduate and graduate programs were math, programming, and simulation heavy. But it wasn’t until 2012 that I developed a genuine interest in machine learning. Coursera was a new thing back then, and the idea of learning from top experts in the field excited me a lot. 

I enrolled in a Social Network Analysis (SNA) course that Dr Lada Adamic taught, which changed everything for me. The notion of social networks and measuring and quantifying information in networks ignited a deep curiosity in me. I connected with researchers and professors in my metroplex, and I sat in several graduate-level AI classes to understand different topics in AI. After gaining solid knowledge, I started developing AI use cases for the problems we have in our group. I was fortunate to have managers that supported me, and as a result, I was able to break into data science roles.

The journey was not easy. It takes a lot of discipline and hard work to self-learn data science. Kaggle was not a thing when I was learning data science; I approached several professors to assist them with their research and learn new skills along the way. What helped me the most was my approach towards learning; I treated machine learning as one more tool in the analytics toolbox. This enabled me to embed machine learning approaches to solve the problems we had in the organisation.

AIM: Can you tell us about your current role at Bose Corporation?

Aswini: I am a data scientist in AI for a business organisation. We develop AI/ML-enabled solutions to help business groups achieve scale and efficiency. Our solutions help marketing, sales, supply chain, and HR organisations make data-driven decisions using the predictive and prescriptive insights we generate. 

AIM: With such vast experience, what would be your suggestions for future ML aspirants?

Aswini: Many aspiring data scientists relate ML to Python or R. While programming in Python and R is critical for a data scientist, the critical skill every data scientist needs is problem-solving. A problem solver will first try to understand the gaps in the current solution and hypothesise how machine learning can help fill those gaps. The hypothesis you defined will then help you identify the required datasets, form a project team, and set proper success measures for the project. So always ask this question before you do anything Data Science – “What’s the problem I am trying to solve?”

AIM: Is it good to go for a PhD, or should one join the industry after doing Masters and gain some industry experience? What would you suggest and why?

Aswini: I think this largely depends on the individual’s goals. A research-based PhD degree will allow you to dig deep into a specific area of interest. Publishing papers and academic teaching will establish you as an authority in that particular subfield. On the other hand, if your goal is to transition careers and get into the workforce faster, a Master’s degree is a great choice. Starting with a Master’s degree can be a good option if you are unsure. Most of the Master’s programs allow you to convert to a PhD if you develop a research interest along the way.

AIM: What books, blogs or other resources would you recommend for the students in India?

Aswini: I found several resources helpful when I started my journey, and the good thing is they are still relevant. Some of my all-time favourite books to learn about machine learning include: 

• Machine Learning: A Probabilistic Perspective by Kevin P. Murphy
• Pattern Recognition and Machine Learning by Christopher Bishop 
• If you like watching videos, I highly recommend Dr Ali Ghodsi and Dr Nando de Freitas lectures.

AIM: You have worked in the industry, so what, according to you, are some of the prominent skills that the current ML aspirants should try to have in order to be industry-ready?

Aswini: An aspiring data scientist should be skilled in programming, statistics, machine learning algorithms, domain knowledge, and data storytelling. A great way to learn these skills is – to participate in competitions such as Kaggle if you are a complete fresher and to raise your hand and seek internal opportunities if you are an experienced professional.

Apart from the technical skills mentioned above, critical thinking and problem-solving skills will differentiate you from an average data scientist. For example, when looking at a Jupyter notebook or a YouTube video, try to ask yourself why a certain decision was made. How would you do it differently if you were given a chance?

AIM: After working for several years in this domain, what are some of the issues that need to be addressed when it comes to AI? Can you give some suggestions for the same?

Aswini: We have come a long way as an AI community. There is a significant intentional push by leaders in all domains to integrate AI into existing solutions. Despite all this buzz, there are several issues we need to fix. A couple of areas where we need to put our immediate attention to are – Explainable AI and AI governance. 

Building good models is not always about achieving state-of-the-art accuracy; it’s about building models that can provide good inferences. Data scientists, especially those assisting businesses to help make decisions, should pay close attention to developing models that are good at providing inferences and prescriptive solutions. Similarly, good AI governance practices will help organisations create effective solutions and prevent teams from deploying biased models.

AIM: Which domain of AI do you think will come up strongly in the next ten years? Any specific reasons to say the same?

Aswini: AI governance, AI bias, and AI product lifecycle management are the three areas I am excited about for the next decade. Industry leaders currently are investing heavily in all things AI; as we go further down the AI maturity curve, tasks such as AI governance and managing AI products through their entire lifecycle become critical.

Having good data is the cornerstone of any successful AI and machine learning project. Conversely, biased models, by-products of bad data, will have a significant economic impact and will create mistrust among consumers. To this end, I see organisations allocating time and resources in the near future to collect qualitative data to fill the current gaps.