The Intel® Student Ambassador Program was created to work collaboratively with students at innovative schools and universities working in the machine learning (ML) and artificial intelligence (AI) space. AI Student Ambassador Amlaan Bhoi took time out of his busy schedule to talk about his work in the field of computer vision and his thesis on action localization applications.
Tell us about your background.
I was born in India, then moved to Thailand after 8 years. Nearly a decade later I moved back to India for 4 more years and now reside in Chicago, Illinois. I am a second-year Master of Science in Computer Science student at the University of Illinois at Chicago. I work primarily with computer vision applications including image classification, object detection, and semantic segmentation. I’m also starting my thesis on spatio-temporal action recognition. The primary task is to recognize actions and answer the questions “when” and “where” those actions occur in a given video.
What got you started in technology?
My dad inspired me to follow this passion as I would watch him work on enterprise applications. I wrote my first application, a basic calculator, in C programming language when I was 10. Around 2013, I started reading about the advancements in convolutional neural networks (CNN) and optimization algorithms and really got into machine learning and deep learning. I loved the ability and possibility of such techniques to change how we interact with information drastically. I’m also inspired and motivated by the extremely fast-paced research and development around the world on extremely important applications. I would definitely recommend Arxiv Sanity as a starting point for anyone interested in following the latest research.
What projects are you working on now?
I’m currently interning at a company called CCC Information Services in their headquarters here in Chicago. I work in the Computer Vision/Photo Analytics team, which works on developing state-of-the-art deep learning models aimed towards solving vision problems in the car insurance domain. More specifically, I’m working on a project to determine whether a car is damaged or total loss.
For my thesis, I’m working on action localization. I believe this field is an open question due to the nature of the problem and resource constraints attached with it. There are many ways to approach the problem, I think the most important issue is an efficient feature representation which utilizes minimum resources to extract and use.
As a side project, I’m also developing a technique to measure similarity between image datasets. This can be used to measure how similar two datasets are and can aid in transfer learning because it helps a network learn faster.
Tell us about a technology challenge you’ve had to overcome in a project.
One of the biggest constraints when working on computer vision problems are compute resources. This is where Intel® AI DevCloud came to the rescue. While working on “Tiramisu DenseNet Architecture for Precise Segmentation,” I utilized Intel’s AI DevCloud for training, testing, and evaluating the model. I will definitely be taking advantage of all these great compute resources for future experiments.
Another major issue is dataset curation, distribution, and representation of real-world conditions. This is an issue that is becoming more prominent as we start applying deep learning models to real-life scenarios rather than toy datasets or academia-based datasets.
Tell us about any Ambassador events you’ve attended.
Recently, I attended CVPR 2018 in Salt Lake City, Utah. I presented a poster on “Tiramisu DenseNet Architecture for Precise Segmentation” at the Intel® AI Booth. The experience was incredible as I got to meet amazing researchers, professors, and students from all over. I can’t wait to share my work at the next event!
How are you planning to leverage artificial intelligence or deep learning technologies in your work?
Deep learning technologies are crucial in some of my projects. They help enable some critical points otherwise not feasible with traditional techniques. The biggest applications I see are end-to-end training for autonomous driving, human pose estimation for long-stream videos, and resource constrained object detection.
How can Intel help students like you succeed?
There are many ways Intel helps students like us. Firstly, they provide compute resources through Intel AI DevCloud. This is an excellent resource to utilize. Secondly, Intel provides access to experts in AI and deep learning. If you are stuck, there is always someone who can help you with your problem. Thirdly, Intel provides guidance and a starting point for anyone who wants to enter this research field.
What impact on the world do you see AI having? And do you see yourself as part of it?
Andrew Ng famously said, “Artificial intelligence is the new electricity”. I could not agree more. The models and algorithms being developed today are going to affect industries all over the world. Their effect can already be seen in the software industry. This will expand to everything from energy, government, agriculture, and more. I definitely see myself as a contributing member to the impact of AI that has started already.
Outside of technology, what type of hobbies do you enjoy?
Outside technology, I love to read. I’m currently reading Sapiens: A Brief History of Humankind by Yuval Noah Harari (It’s an interesting proposition). Besides that, I follow soccer, play table tennis and snooker, and love running! I also enjoy debating socio-political and economic issues and host/participate in informal debates every two weeks.
What are you looking forward to doing with Intel?
As part of the Intel Ambassador program, I hope to continue working on interesting experiments, share upcoming research, and be part of an amazing community. The ambassador program has given a wide array of resources otherwise unavailable to graduate students. It is crucial to keep on learning and sharing work!
Upon acceptance into the Intel® Student Ambassador Program for AI, Graduate and PhD students from top universities worldwide can access newly-optimized frameworks and technologies, hands-on training, and technical resources provided by the Intel® AI Academy. To be considered for the program, apply today.
Intel's compilers may or may not optimize to the same degree for non-Intel microprocessors for optimizations that are not unique to Intel microprocessors. These optimizations include SSE2, SSE3, and SSSE3 instruction sets and other optimizations. Intel does not guarantee the availability, functionality, or effectiveness of any optimization on microprocessors not manufactured by Intel. Microprocessor-dependent optimizations in this product are intended for use with Intel microprocessors. Certain optimizations not specific to Intel microarchitecture are reserved for Intel microprocessors. Please refer to the applicable product User and Reference Guides for more information regarding the specific instruction sets covered by this notice.
Notice revision #20110804