Posted on

Alumni Research Talks

Speaker #1: Rishabh Mehrotra

Dr Rishabh Mehrotra is currently working as a Senior Research Scientist at Spotify. He graduated from BITS Pilani with a bachelor’s degree in Computer Science Engineering and a master’s degree in Mathematics in 2013. He completed his PhD in Machine Learning from University College London, supported by a Google research grant. The talk was centred around the research involved in the development of Recommender Systems (RecSys)—algorithms that produce personalised and dynamic content for every user. 

After a brief introduction and an account of his time at BITS Pilani, Dr Mehrotra discussed the traditional approaches used in building RecSys models—Collaborative Filtering, Latent Variable Models, Neural Embeddings, and Neural Collaborative Ranking.

The models had a common user-centric nature, with recommendations focused on improving user interaction and satisfaction. However, these recommendation models were observed to be insufficient in scenarios involving multiple stakeholders. Instead, a contextual machine learning model was required that would take into account artist diversity, promotional objectives, platform economics, and consumer metrics.

While sharing advice for students hoping to get started with research work, Dr Mehrotra mentioned the following—

  • Know the most reputable conferences in your area of interest and look out for ‘tutorials’ that take place at the beginning of these conferences—these are a great way to brush up on key concepts.
  • Look out for workshops that typically happen near the end of these conferences for exposure to the most recent research directions in the industry.
  • Evaluate the relevance of your problem statement as that closely ties in with your chances of getting published.
  • Understand the nuanced expectations associated with publishing at particular venues and journals by reaching out to past contributors. 
  • Prepare your publication timeline keeping in mind the period for review and rebuttal, which is typically a couple of months, when applying to MS/PhD programs. 
  • Break away from the notion that only papers involving core algorithmic work and heavy experimentation are valued in the community—user studies, application-driven papers, insights, and analysis are a great starting point and key in inspiring future research directions. 
  • Get started with these publically available datasets—

The session ended with Dr Mehrotra answering some of the audience’s questions and a vote of thanks.

Speaker #2: Dr Anand Rao

The second speaker in the series was Dr Anand Rao, the Global Artificial Intelligence (AI) lead at PricewaterhouseCoopers (PwC). He recounted his journey through three different careers in Engineering, Research, and Management. 

After graduating from BITS Pilani in 1984 with an MSc (Tech) in Computer Science, Dr Rao pursued a PhD in Artificial Intelligence at the University of Sydney. During that time, he co-edited four books about Intelligent Agent Systems. He elaborated on his research, a major part of which was centred around space shuttle simulations. In 1996, he designed an abstract programming language called AgentSpeak.

He completed his MBA from Melbourne Business School in 1999. After that, he joined Michael Madison Group as a consultant and moved to London, distancing him from the AI sphere. He began working at Diamond Management and Technology Consultants in 2001 and continued there through a move to Boston in 2005. 

In 2007, he received the International Foundation for Autonomous Agents and Multiagent System (IFAAMAS) Influential Paper Award for his paper “Modelling Rational Agents within a BDI Architecture” published in 1991. Dr Rao said that this rekindled his passion for AI. When PwC acquired Diamond Management in 2010, he was the natural choice to lead research in Big Data, Analytical Techniques, and Artificial Intelligence. He mentioned that he has worked with 28 government agencies, including those from Luxembourg, New Zealand, Australia, and India.

In 2018, he received the BITS Pilani Distinguished Alumnus Award for corporate leadership, which brought him back to his alma mater.

Dr Rao has been making the most of his free time during lockdown by maintaining a blog on practical applications of Vedanta, combining the wisdom of ancient scriptures/text with modern management theories. Besides this, he also has a blog for articles related to technology and ethics. 

Speaker #3: Aruna Balasubramaniam

Dr Aruna Balasubramian, Associate Professor at Stony Brook University and a BITS alumna, talked about her journey through academia. She was a dual degree student, pursuing a bachelor’s in Information Systems and a master’s in Management Studies. She went on to complete an MS at the University at Buffalo, a PhD at the University of Massachusetts, Amherst (UMass) and a post-doctorate at the University of Washington (UW). After this, she accepted a position as an assistant professor at Stony Brook, where she is now an associate professor with tenure.

The core themes of the talk were the key lessons she learnt from her experiences and how she applied them in her research career. She felt that some of her advice would be contradictory in nature, but it was up to the individual to choose what to apply in their lives. 

The talk started with her reminiscing about her time at BITS and how she entered academia. A few projects in Networking, talking to other MS and PhD students at BITS, and work as a member of the Department of Paper Evaluation and Presentation (PEP) helped her get a better idea of what research was like. She applied for an MS following in the footsteps of her friends and seniors.

While at the University at Buffalo, she was drawn towards the world of Network Theory. She found herself to be quite talented at the subject, and learnt to play to her strengths by getting selected to work on a NASA-funded project on network security during her MS.  

As a grad student at UMass, her initial dissertation topic was rendered obsolete due to the boom of cellular technologies and networks. She explained how she adapted to fit with the industry trend by changing her area of research. Her dissertation ended up winning the best dissertation award at UMass. This taught her that knowing when to pivot and abandon projects is crucial for success in academia. 

Dr Balasubramnian entered academia during the global recession; since faculty jobs were scarce, she did a post-doc at the University of Washington (UW). During her time there, she leveraged UW’s close ties with Google to communicate with them frequently about creating a new architecture named WProf. Multiple protocols and projects have since been developed on her work. Her takeaway was to leverage one’s environment in the best way possible.

After her postdoctoral, she was offered a faculty position at the University of Stony Brook. She currently works at the NetSys lab of the university, which focuses on culminating interdisciplinary fields with Network Systems and creating technologies that have a societal impact. She is also a champion for greater representation of female and minority students, working on two funded projects aimed at improving their conditions such that they finish their education. 

She ended the talk with the advice of enjoying one’s time at BITS and shared a photo of a Zoom call she had with all her old wing mates 20 years after her graduation. She spoke to the support her friends had provided her after she graduated.

Speaker #4: Dr Siddhartha Nath

Dr Siddhartha Nath works as a senior R&D software engineer at Synopsys Inc. He is currently working in the development of AI-accelerated chip-design algorithms. Dr Nath graduated from BITS Pilani in 2003 with a bachelor’s degree in Electrical and Electronics Engineering and received his master’s degree and PhD from the University of California, San Diego. His areas of research are Integrated Circuit Design Management using combinatorial optimisation, mathematical programming, and machine learning. The main focus of his talk was next-generation Electronic Design Automation (EDA) physical design and its empowerment using machine learning.

The session started with an overview of the Very Large-Scale Integration(VLSI) design flow, focusing on the various steps involved in its physical design aspect. He proceeded to discuss the domain knowledge and skill set required in EDA related fields, citing varied examples from our daily life to depict the importance of the topic of discussion. He went on to try and bridge the gap between physical design and machine learning by discussing various aspects of optimization of IC designs and describing the tremendous opportunities in the field of machine learning.

The next part of the talk focused on the investments Synopsys makes to encourage machine learning and artificial intelligence. Further, Dr Nath explained the various research opportunities in the field of EDA and physical design. He concluded the talk by inspiring second-year students from circuital branches to consider the Synopsys academic program, internships, and full-time opportunities.

Speaker #5: Mukesh Jain 

Mr Mukesh Jain is the Chief Technology and Innovation Officer (CTIO), VP & Head of 890 by Capgemini. He has previously held senior positions at Microsoft and Jio. He received his bachelor’s degree in Computer Science and Engineering from MGM College of Engineering in 1995 and a master’s degree in Software Systems and Specialisation in Analytics from BITS Pilani in 2018.

Mr Jain’s talk revolved around data, analytics, Machine Learning (ML), and Artificial Intelligence (AI). In the beginning, he talked about users and the decisions they make while choosing any particular company for their service. These decisions are then analysed in the form of data by companies such as YouTube, Amazon, and Uber. He highlighted the importance of data-driven information in a company’s strategy to capture users, improve their services, and derive profits. Furthermore, he shared the story of a simple two-week-long experiment he had conducted during his time at Microsoft to determine the shade of blue used by Microsoft in their webpages and how it affected users’ choices and preferences when using their services, thereby explaining the intricacies of data collection and analysis.

Mr Jain mentioned that the “Send Error Report” button on Windows crash reports and the Junk filter for Outlook were his ideas . According to him, he wanted to know the reason behind the cause of errors and glitches, in an attempt to improve Microsoft Office. Faced by a lack of data, he proposed a “Send Error Report” button to get that data directly from the users. This was initially dismissed, but after looking at the improvement in the functioning of MS Office, it was implemented at a larger scale in the organisation—thus giving the students a lesson on curiosity, analytical reasoning, and implementing the  “Jack of all trades, master of some” mindset.

He concluded the talk by sharing personal learnings from his experience working with some big names in the industry, laying key emphasis on his ideology of ‘Poochne mein kya jaata hai? (What’s the harm in asking?)’. 

The session ended with Mr Jain answering the attendees’ queries followed by a vote of thanks by the CSA.

Speaker #6: Ms Vidhi Jain

Ms Vidhi Jain, who graduated from BITS Pilani in 2018, is currently a research assistant at Carnegie Mellon University (CMU). Using machine learning techniques such as natural language processing (NLP), robotics, and computer vision, she works on navigation by reinforcement learning and planning. During her time at BITS, she was involved in Association for Computing Machinery (ACM), Department of Publications and Correspondence (PCr) APOGEE, and National Social Service (NSS) alongside her computer science bachelor’s degree. She interned at FZI, Germany, in her second year and at Mitacs, Canada, in her third year.

In her advice for research aspirants, Ms Vidhi highlighted three points: do background work on your project, make a working prototype of your idea, and formulate a general theory based on the problem you solved.  For students who are uncertain about whether they should obtain a master’s degree or gain work experience, she listed the differences between both the avenues⁠—graduate schools give you the freedom to work closely with an advisor on novel projects, whereas industry jobs involve large teams with little scope for individual growth. In terms of the budget, industry projects operate at a much larger scale than graduate school programmes, where there are significant budget constraints. 

Through this talk, she explained in detail two of her recent research projects. The first one uses machine learning to predict human strategies in a simulated search and rescue operation. The AI uses two strategies: victim saving strategy prediction and next location prediction. She has created neural networks that achieve human-level accuracy in simulations of this project. Her second project is about learning to capture spatial semantic priors for indoor navigation. Its primary focus is to navigate an unseen indoor environment—using machine learning—by optimising paths for locating an object. In this project, she has achieved an accuracy greater than 90%. This is particularly promising as the model gives accurate results even as path lengths increase.

Ms Vidhi ended the talk with motivation for aspiring researchers, urging them to go after the field they want to see a change in. She encouraged them to work hard on their project, as it is the only way of achieving desirable results. 

Speaker #7: Dr Divyakant Agrawal

Dr Divyakant Agrawal is a Computer Science professor at the University of California, Santa Barbara. He is an alumnus of BITS Pilani who graduated in 1980. His present research areas are related to databases, distribution systems, cloud computing, and big data analysis. He is affiliated with various reputed organizations and has served on the editorial board of several publications.

Dr Agrawal began the session by giving a brief overview of his formative years. He pursued a bachelor’s degree in Electrical Engineering from BITS Pilani, and subsequently reminisced about his life as an undergraduate. After working for two years, Dr Agrawal pursued an MS and a PhD in Computer Science from Stony Brook University.

His lecture on “Demystifying Blockchains for Big Data” began by acknowledging his colleagues and students. The initial frame revolved around the advantages of blockchain, which include transparency, provenance, fault tolerance, immutability, and authenticity. Each of these was elaborated upon by giving relevant examples. He then drew parallels between the traditional banking system and blockchain. Bitcoin replaces traditional identifiers like usernames, passwords, and signatures with a “key pair” and provides for ledgers to be saved in a mutual and consistent log rather than a database. This also allows for hard-to-forge private keys—to easily verify the authenticity of a transaction—while a continuous ledger eases tracking the transaction history.

Dr Agrawal stated a few cons of cryptocurrencies, like double spending, low rate of transactions per second (TPS), probabilistic consistency guarantees, and high transaction confirmation latency. He briefly introduced the concept of sharding—partitioning a blockchain into various networks. However, this introduces complications in inter-network transactions, which can only be solved using “atomic” transactions and fixed protocols. He stated that the latest solution proposed in 2018 is incoherent. The solutions posed by Dr Agrawal and his colleagues and students involve using a “coordinator node” to solve the problems with “cross-sharding”.

Dr Agrawal then explained the applications of blockchain. He took the example of Dolby and how it earns royalties from the sale of its intellectual property to various parties. Blockchains help keep track of the payments received from a large number of clients. Walmart has also implemented a similar system to track down contaminated deliverables. Certain issues with permissioned blockchains still pertain, however. 

Towards the end, Dr Agrawal attended to the questions of a few students. He then posed certain thought exercises that stimulated the audience to think about the social impacts of blockchains and their viability in the current scenario. 

Speaker #8: Dr Ganesh Ananthanarayanan

The eighth talk of the series featured Dr Ganesh Ananthanarayanan, who graduated from BITS Pilani in 2005 with dual bachelor’s degrees in Civil Engineering and Computer Science. After a brief stint as a research fellow at Microsoft India, he went on to pursue a PhD from the University of California, Berkeley. He is currently working as a researcher at Microsoft, where he studies systems for large-scale data analytics, video processing, and internet performance.

Dr Ganesh primarily focussed on video analytics and detailed his quest to achieve low latency, high accuracy, and low cost in the field. Talking about the motivation behind pursuing research on this subject, Dr Ganesh stressed the role of video analytics in the evolution of technology. According to him, cameras are integral in the promise of a 5G future. ‘For every seven people in the US, there exists one camera,’ he added. Dr Ganesh demonstrated the applications of live video streams, referencing projects he had undertaken in the city of Bellevue and parts of Australia. His team had developed a system to display car, bike, and pedestrian counts by analysing footage from traffic cameras. Another application that the team had worked on was locating parking spots in densely populated cities. 

Dr Ganesh laid out the structure through which a video analytics pipeline is developed. A profiler and a scheduler are used to obtain a sample video of appropriate configurations and place them across a hierarchy of clusters. The key challenge, according to him, is reducing the cost of profiling. As the video content varies over time, the video pipeline has to be adapted accordingly. This further exacerbates the necessity of a profiler. In the latter part of the talk, Dr Ganesh elaborated on the utility of stored videos. Citing his work in Bellevue, he discussed interactive querying of stored datasets, the goal being to process video frames using a Convoluted Neural Network (CNN) in a way that is efficient and cost-effective. 

The talk ended with a question-answer session. Dr Ganesh shared his experience working at Microsoft’s Indian and American offices and offered insights into what the future looks like for video analytics.