Small-scale initiative in Machine Learning 2021: how did it go?
Introducing a series of blog posts on our collaborative projects in machine learning
By Sonja Georgievska and Jisk Attema
In the last decade, the machine learning field has seen exponential growth, owing mostly to the success of its subfield “deep learning”, based on deep neural networks. The latter build on the traditional artificial neural networks by adding many layers and parameters. This, together with some mathematical tricks and availability of “big” data and appropriate hardware, have made the networks much better at problem-solving. At the same time, many new deep neural network architectures have been invented that show even better results and/or are able to use data efficiently. As a result, deep learning is now used universally in computer vision, speech recognition, natural language processing, machine translation, bioinformatics, drug design, medical image analysis, material inspection, etc.
For the modern scientist (e.g. a medical researcher, a climate scientist, a political scientist) who wants to take advantage of machine learning, it is becoming increasingly challenging to be able to stay up-to-date with their scientific domain development and the explosive growth of machine learning techniques at the same time. Even worse, many new developments require extensive programming and data management skills. In fact, we notice that the latest breakthroughs are increasingly accomplished through close collaborations between domain scientists and machine learning specialists.
To connect to this demand for machine learning expertise, the eScience Center organized deep learning courses for interested researchers in 2020 and 2021. However, we noticed that ‘just a course’ was not enough to get the participants to successfully apply machine learning in their research. Finding the best machine learning method for their research problem and data, analyzing, and interpreting the output of the machine learning algorithms, troubleshooting, as well as iterating on the experimental setup was essential. This led to the idea of the ‘Small Scale Initiatives’ (SSI): an open call for short to medium-sized projects with a focus on applying eScience to actual research questions.
Through the first SSI call, the eScience Center selected 12 teams, with which the eScience Research Software Engineers (RSEs) would work closely for several months in the second half of 2021. The aim was that the RSEs would advise the research teams on how to best apply machine learning to their concrete research questions, from initial idea to publication. The research teams come equipped with domain knowledge and programming experience, while the RSEs provide for nuances between various machine learning approaches and strategies towards solving the specific research question, and for continuous support throughout the process.
Well, how did this initiative go? What exciting new research came out? What breakthroughs did we achieve? In the following months, through a series of blog posts written by the 12 research teams, we will go through the ins and outs of these exciting collaborations. By writing short and accessible stories, the teams will describe their experience with applying ML and with the collaboration with the eScience Center. Keep an eye on this blog series to learn more!
Update
Here we go:
How to find your rubber duck: Using machine learning to understand a changing sea
Parsing Hebrew and Syriac morphology using Deep Learning
A machine learning approach to laughter
How machine learning could help Simone to play Ludo
Can machine learning help us improve stroke rehabilitation? A step towards personalized therapy
Intermezzo: from the perspective of an eScience Engineer
Using machine learning to tell apart rain, snow, hail and fog from cell tower data
The mystery of glass: why machine learning can help us
Studying political symbolism in Turkish TV dramas with machine learning