Lynker Analytics hosted the monthly meeting of the Machine Learning & Data Science meetup group in Wellington New Zealand on May 2nd. Over 50 practitioners and enthusiasts attended our BizDojo Market Lane, Wellington location to hear Principal Data Science, David Knox speak about Practical Active Learning for Image Classification.
In his talk David spoke about Supervised Machine Learning which of course requires a lot of data with which to train accurate models. He explained the conundrum whereby if you are starting a project with a lot of unlabelled data, no or very little ground truth labelled data and little budget to fund a labour-intensive labelling effort, where do you start?
He explored the practical use of Active Learning to help you gather an efficient training set of labelled data for an image classification project. He gave a brief overview of the concept of Active Learning and a demonstration of different sampling strategies - where you choose the results of a weak model to select samples to be hand labelled for further training.
David described and demonstrated a human-in-the-loop data labelling tool, looking first at the software components and then the experience of the human-in-the-loop working through a real-world image classification problem. Finally, David discussed the application of Active Learning to large datasets with very sparse features, of interest when extracting features from imagery.
It was a very informative evening and was thoroughly enjoyed by all. The presentation and code resources are available at Github. If you are based in Wellington and are interested in joining the group please register here.