Aman Krishna

Hi,

My name is Aman Krishna. Currently, I am working as a Machine Learning/AI researcher at Samsung Research India.

At Samsung, I worked on 3 major AI/ML projects, leading two of them. The first one was to fix the issues and add improvements to the current automatic Hand Wash Detection model available in Samsung Galaxy watches. The journey started with identifying biases in our data, followed by data augmentation, experimentation with different architectures, and generating more in-house data. In the end, we improved the existing model accuracy by 8.7%.

The second project was to improve the performance of wrist-up detection in Samsung smartwatches. A big challenge, the model size must be less than 1 KB. I collaborated with the lead on the project to perform exhaustive hyperparameter tuning to build the best model. We ran comprehensive tests to identify areas of improvement. As a result, we improved the wrist-up detection time by 130ms & reduced the false-positive rate in some scenarios by up to 60%.

The third project stemmed from the common problem faced in the first two projects, the lack of data. I proposed using Generative Adversarial Networks to create synthetic data for the team. Step one was to research the existing best approaches in academia. I followed by tweaking and improving the architecture for the team’s needs. It is my current work scope & has a solid potential to turn into a research paper.

Apart from these, I have worked in Recommender systems, Natural Language Processing (NLP), and Computer Vision with projects and code bases available on GitHub. My work experience spans around 3 years, with 1.5 years dedicated to machine learning and others distributed among Full Stack development, Android development & Data Science.

I have completed my Bachelor's in Computer Science from IIT Dhanbad and my Masters's in Computer Science from IIIT Hyderabad.