Feedback Prize - English Language Learning

The goal of this competition was to assess the language proficiency of 8th-12th grade English Language Learners (ELLs). Utilizing a dataset of essays written by ELLs will help to develop proficiency models that better supports all students. We created transformer based ensemble of models, code is available at https://github.com/rashmibanthia/feedback3 Ranked top 1% - 8th out of 2654 teams.

Sartorius Cell Instance Segmentation

Detected and delineate distinct objects of interest in biological phase contrast microscopy images depicting neuronal cell types commonly used in the study of neurological disorders. We used Detectron2 and MMdetection library, ranked top 2%. Demo is available here - https://huggingface.co/rashmi

Google Brain - Ventilator Pressure Prediction

Simulated a ventilator connected to a sedated patient's lung. This helps overcome the cost barrier of developing new methods for controlling mechanical ventilators. Used LSTM based TF TPU and Pytorch Models. Ranked in top 1% (33/2605)

G2NET

Detected Gravitational Wave signals from the mergers of binary black holes. Specifically, built models to analyze simulated Gravitational wave time-series data from a network of 3 Earth-based detectors. Used ensemble of Pytorch and TF TPU based models, ranked top 5%.

CommonLit Readability

Built algorithms to rate the complexity of reading passages for grade 3-12 classroom use. Utilizing large pretrained NLP models from Huggingface, pretrained Roberta Large model. Stacked multiple large models, final rank 18th out of 3600 competitors.