Active Learning with Sentence Error Classification
in Machine Translation
NLP Grad Research
We explore a model-based Active Learning method
in Neural Machine Translation, for translating from two low-resource languages,
Gujarati and Somali, to English, which utilizes the multi-headed multi-layered encoder-decoder
attention weights of the Transformer architecture.
real-time Super Resolution GAN
Undergrad Research Project
In the last few years, several deep learning models, especially Generative Adversarial Networks have
received a lot of attention for the task of Single Image Super-Resolution (SISR). This paper focuses
on improving an existing deep-learning based method to perform Super-Resolution Microscopy in
real-time using a standard GPU. We compare the quality and the running time for the outputs produced by our model, opening its
applications in different areas like low-end benchtop and even mobile microscopy
Working on Semi autonomous agricultural vehicles
Self Project
Just my forray into the field of visual SLAM
Automating Calibration process for Biomedical devices
project for Major requirement
Calibration and incertainty estimation for any device to ensure that the output that they are giving matches the reading shown under certain thresholds, especially in medical sector. In this project I calculate uncertainty for different biomedical devices like defibrilator and Infusion Pump.
I also hack together setup using arduino which which automates this process of uncertainty calculation and directly connects devices of different standards to a standard computer system.
