Chaitanya Animesh

Namaste! I am Chaitanya Animesh and I currently work as a Machine Learning Engineer at Symbl.ai. My work involves research and engineering around Large Language Models. I completed my Master's in Computer Science from University of California San Diego with specialization in AI.

During my Master's I began my journey in deep learning research. Prior to my Master's, I worked as a Software Engineer at JP Morgan Chase & Co . I earned my bachelor's degree in Electrical Engineering from IIT (BHU) Varanasi. I like AI, maths, cricket and nothingness.

Email  /  CV  /  Google Scholar  /  Twitter  /  Github  /  LinkedIn  /  Medium

I am interested in – 1. Deep learning – self-supervised learning, generative models, domain adaptation, 2. Natural language processing – auto-regressive & large language models and 3. Computer vision – multi-view geometry and learning based recognition. The following is my list of research work currently (hopefully with more to come!).

We propose a novel contrastive loss function that uses multiple positives and multiple negatives and show how it can be applied in both supervised and self-supervised settings. We provide theoretical analysis of how our loss function's gradient response is superior. It beats SupCon loss and performs on par with various SOTA SSL methods.

We propose a novel framework for fast detection of text regions in a camera captured single hazy image. A novel text extraction technique using the channels of CMYK color space is used. A simple, fast, and efficient approach of using contrast enhancement technique suitable for smart devices is proposed. A maximum weighted matching based grouping approach is used to form words. We have also created a new dataset of hazy images that contains scene texts and manually annotated the words in it.

Proposed a non-linear fixed time controller in order to track the reference voltage within a fixed time period overcoming the limitations of a PID controller. Mathematically proved the asymptotic stability of the proposed controller using Lyapunov theory of stability.

Template taken from Jon Barron (the Nerf guy!).