amritpurshotam Goto Github PK

My name is Amrit Purshotam and I'm a machine learning scientist.

Here are some projects that I work on in my spare time.

• ImageNet Papers: I'm implementing the best performing architectures from every year of the ImageNet challenge since 2012 as an exercise in reproduction and to step through time to see how the approaches evolved first-hand. What I'm doing differently is implementing the architectures with the full training pipelines and making the training runs public with wandb.
• FancyPCA: I built a demo of the PCA Colour Augmentation technique with Streamlit which can be seen live here.
• Baobab: I'm helping to build an event management platform that enables grass-roots community builders to scale their efforts when organising conferences that have complex requirements for selecting attendees. The application is primarily built for and used by the Deep Learning Indaba who I passionately support and helping build this is my way of contributing to their mission.

I'm also currently incubating the below projects while I figure out some final details before getting started in earnest.

• DeepSARA: A revisit of my MSc project where I built classification and detection models for southern African rock art figures using classical ML and CV techniques. I'm now curious to see how well some deep learning approaches work. But first I need to re-label the dataset.
• park2vec: I want to train general location vectors with the aim to use them to build a parkrun course similarity tool. I'm currently trying to wrangle OpenStreetMap data into a format that would be useful to train on.