There is no specific prompt I am trying to answer. So this is a general statement of purpose.
The advent of the computer, together with Turing's theory of universal computation, has revolutionized technology and science. However, for all this progress, one problem remains elusive to the computational paradigm - human intelligence. While computers are astoundingly good at solving formally specified problems, the kind of intuitive reasoning that humans perform each second is harder to reproduce. At the same time, this type of reasoning creates immense value for society - a large fraction of expenditures today go towards human labor, making the automation of "human" tasks an attractive target for economic growth. These considerations lead me to believe that artificial intelligence is the highest-impact technology that is currently being developed.
My career goals are therefore to develop intelligent automation systems that will benefit society with a particular focus on computer vision. I am especially interested in the problem of contextual scene understanding for autonomous driving where understanding the behavior in terms of agents, their goals, intentionality, and making predictions about what might happen is quite challenging.
Instance segmentation has come to be one of the relatively important, complex, and challenging areas in machine vision research. This realization came to me when I was working on a problem to automate detecting damages in vehicles for Sony Ericsson at Smart India Hackathon 2019 where we faced issues while applying instance segmentation for small objects and handling occlusions. Our approach was based on using Mask-RCNN and by transfer learning on the pre-trained weight for the MS-COCO dataset, we were able to achieve 83% accuracy on the test set provided by Ericsson. This project was ultimately awarded the 1st prize at the finals winning a cash prize of 100,000 INR. I also presented a paper based on this system in ICACTA 2020 which was eventually published in Springer's Advanced Computing Technologies and Applications.
Widespread deployment of automated systems will profoundly alter the pace of both economic and scientific progress, by allowing an increasingly large variety of presently labor-intensive tasks to be automated. One area is visual surveillance. This was a theory that I concentrated on proving during my final dissertation under Dr. Shashi Dugad at Tata Institute Of Fundamental Research(TIFR). As TIFR is located on the shorelines, it requires constant surveillance against foreign threats, especially after the 26/11 Mumbai attacks. The objective was to automate the task to identify any malicious ships or persons from the CCTV feeds. We optimized the YOLO object detection algorithm by introducing the idea of spatial pyramid pooling, redefining the loss function, and introducing a weight regularization. This improved the detection accuracy rate by 4.3% as well led to lower inference time.
Another problem in computer vision that I would like to explore is Face recognition. I think that the reason for the slow progress was due to factors like in person variability and the tendency to conflate familiar and unfamiliar face processing. My work as a participant in the Cisco ThingQbator program was directed at exploring this problem. I created a face recognition system for tracking criminals in very low-resolution CCTV cameras. We devised a novel approach using the existing paper on Facenet architecture to extract face embeddings instead of relying on the bottleneck layers in Convolutional Neural Network(CNN) and then feeding these embeddings through an ensemble learner. My research based on this approach has been accepted and presented in ASIANCON 2021 and is digitally available on IEEE Xplore.
People are afraid that the cognitive automation that artificial Intelligence brings will take away millions of jobs. But I am a firm believer that enhancing and collaborative potential that we envision stands in stark contrast to the zero-sum predictions of what artificial intelligence will do to our society and organizations. Instead, I think that greater productivity and the automation of cognitively routine work is a boon, not a threat.
This has become all the more evident in my current professional work experience at Tata Consultancy Services (TCS) on an automation research project called - "Intelligent Services Network" where we noticed a record year-on-year 100% increase in revenue and business growth during our pilot release. The project aimed to build a system that could derive contextual information about enterprises and provide proactive cues instead of reactive recommendations.. My work is directed at working with ontologies for knowledge representation and in turn utilize their model-theoretic semantics to gain better background information about business processes compared to knowledge graphs and graph embeddings. This background knowledge helped me expand and enrich features in machine learning tasks. Ultimately we were able to create intelligent business workflows which can lead to autonomous, cognitive-based decisions for business enterprises.
A graduate degree in CS from ___ would serve as a launchpad for my career aspirations. (Talk about University). I am fascinated by the work at Waymo LLC and Tesla which are using machine learning to improve vehicular vision for autonomous driving. After my master's, I would like to work as a vision developer for these companies. To achieve my goals, I am mindful of the need to bridge the gaps in my skills and acquire advanced knowledge through a Master's in Computer Science.
STATEMENT OF PURPOSE
The advent of the computer, together with Turing's theory of universal computation, has revolutionized technology and science. However, for all this progress, one problem remains elusive to the computational paradigm - human intelligence. While computers are astoundingly good at solving formally specified problems, the kind of intuitive reasoning that humans perform each second is harder to reproduce. At the same time, this type of reasoning creates immense value for society - a large fraction of expenditures today go towards human labor, making the automation of "human" tasks an attractive target for economic growth. These considerations lead me to believe that artificial intelligence is the highest-impact technology that is currently being developed.
My career goals are therefore to develop intelligent automation systems that will benefit society with a particular focus on computer vision. I am especially interested in the problem of contextual scene understanding for autonomous driving where understanding the behavior in terms of agents, their goals, intentionality, and making predictions about what might happen is quite challenging.
Instance segmentation has come to be one of the relatively important, complex, and challenging areas in machine vision research. This realization came to me when I was working on a problem to automate detecting damages in vehicles for Sony Ericsson at Smart India Hackathon 2019 where we faced issues while applying instance segmentation for small objects and handling occlusions. Our approach was based on using Mask-RCNN and by transfer learning on the pre-trained weight for the MS-COCO dataset, we were able to achieve 83% accuracy on the test set provided by Ericsson. This project was ultimately awarded the 1st prize at the finals winning a cash prize of 100,000 INR. I also presented a paper based on this system in ICACTA 2020 which was eventually published in Springer's Advanced Computing Technologies and Applications.
Widespread deployment of automated systems will profoundly alter the pace of both economic and scientific progress, by allowing an increasingly large variety of presently labor-intensive tasks to be automated. One area is visual surveillance. This was a theory that I concentrated on proving during my final dissertation under Dr. Shashi Dugad at Tata Institute Of Fundamental Research(TIFR). As TIFR is located on the shorelines, it requires constant surveillance against foreign threats, especially after the 26/11 Mumbai attacks. The objective was to automate the task to identify any malicious ships or persons from the CCTV feeds. We optimized the YOLO object detection algorithm by introducing the idea of spatial pyramid pooling, redefining the loss function, and introducing a weight regularization. This improved the detection accuracy rate by 4.3% as well led to lower inference time.
Another problem in computer vision that I would like to explore is Face recognition. I think that the reason for the slow progress was due to factors like in person variability and the tendency to conflate familiar and unfamiliar face processing. My work as a participant in the Cisco ThingQbator program was directed at exploring this problem. I created a face recognition system for tracking criminals in very low-resolution CCTV cameras. We devised a novel approach using the existing paper on Facenet architecture to extract face embeddings instead of relying on the bottleneck layers in Convolutional Neural Network(CNN) and then feeding these embeddings through an ensemble learner. My research based on this approach has been accepted and presented in ASIANCON 2021 and is digitally available on IEEE Xplore.
People are afraid that the cognitive automation that artificial Intelligence brings will take away millions of jobs. But I am a firm believer that enhancing and collaborative potential that we envision stands in stark contrast to the zero-sum predictions of what artificial intelligence will do to our society and organizations. Instead, I think that greater productivity and the automation of cognitively routine work is a boon, not a threat.
This has become all the more evident in my current professional work experience at Tata Consultancy Services (TCS) on an automation research project called - "Intelligent Services Network" where we noticed a record year-on-year 100% increase in revenue and business growth during our pilot release. The project aimed to build a system that could derive contextual information about enterprises and provide proactive cues instead of reactive recommendations.. My work is directed at working with ontologies for knowledge representation and in turn utilize their model-theoretic semantics to gain better background information about business processes compared to knowledge graphs and graph embeddings. This background knowledge helped me expand and enrich features in machine learning tasks. Ultimately we were able to create intelligent business workflows which can lead to autonomous, cognitive-based decisions for business enterprises.
A graduate degree in CS from ___ would serve as a launchpad for my career aspirations. (Talk about University). I am fascinated by the work at Waymo LLC and Tesla which are using machine learning to improve vehicular vision for autonomous driving. After my master's, I would like to work as a vision developer for these companies. To achieve my goals, I am mindful of the need to bridge the gaps in my skills and acquire advanced knowledge through a Master's in Computer Science.
