Computational Biology, and ML interop

Our meeting in June has two exciting young Julians talk about their recent work with the language.

Schedule:
6:45pm Pizza, networking
7:15pm Ayush Shridhar - ONNX.jl: Interoperating ML models
8:00pm Joe Greener - Julia: a natural language for computational biology
8:45pm End

Joe Greener is a research associate (postdoc) in the Bioinformatics Group at UCL. The group is based at the Francis Crick Institute. Previously he was a PhD student in the Structural Bioinformatics Group at Imperial College London. His interests include protein structures, software development and open science. Currently his research involves developing computational methods to predict and design protein structures.

Joe will speak about "Julia: a natural language for computational biology". He will discuss some of the BioJulia organisation, show packages such as BioStructures.jl, Bio3DView.jl and Molly.jl, and talk about how he uses Julia and its ecosystem in his research.

Ayush Shridhar is a final year undergraduate majoring in Computer Science. He has been involved with the julia community for over 2 years and he's been contributing to the FluxML machine learning ecosystem since. His main areas of interest are Machine learning and developing ML related software.

Ayush will demonstrate ONNX.jl, which is an ONNX backend for the Flux machine learning framework. It reads ONNX graphs and produces the Flux code for these high quality pretrained models. Open Neural Network eXchange is a format of representing machine learning models. With the growth of machine learning frameworks, a set of specifications were needed that could help us in removing the framework specific barrier from the world of machine learning. Thus ONNX was born, with the aim of implementing cross framework transfer learning. In essence, a simple example can be training a model in Pytorch but deploying it as a Tensorflow model. There are a number of pretrained models available online, ONNX helps one load these models built in any framework into any other framework. ONNX.jl is a package Ayush developed during his summer of code project last year. Unlike other ONNX backends, ONNX.jl actually produced Julia code for the model, and thus this can be used for a variety of tasks, from normal classification, to fine-tuning for segmentation, style transfer. This eliminates the need of training the model from scratch, thus reducing time and need for additional computational resources.

Image Credit: The Amazing @cormullion