Machine learning for the rest of us

Dr Chris Hines1

1Monash Eresearch Centre, Clayton, Australia


Neural Networks are the new hawtness in machine learning and more generally in any field that  relies heavily on computers and automation. Many people feel its promise is overhyped, but  there is no denying that the automated image processing available is astounding compared to  ten years ago. While the premise of machine learning is simple, obtaining a large enough  labeled dataset, creating a network and waiting for it to converge before you see a modicum of  progress is beyond most of us. In this talk I consider a hypothetical automated kiosk called  “Beerbot”. Beerbot’s premise is stated simply: keep a database of how many beers each person  has taken from the beer fridge. I show how existing open source published networks can be  chained together to create a “good enough” solution for a real world situation with little data  collection or labeling required by the developer and no more skill than a bit of basic python. I  then consider a number of research areas where further automation could significantly improve  “time to science” and encourage all eResearch practitioners to have a go.


Chris has been kicking around the eResearch sector for over a decade. He has a background in quantum physics and with the arrogance of physicists everywhere things this qualifies him to stick his big nose into topics he knows nothing about.

Recent Comments