My reflections, news and interests, mostly around machine learning and technology in education. Hopefully some of it is educational too.
Reviewing my PhD student's final thesis draft left me with a question his models were never asked to answer: what happens when a model has to live through decades of drifting data? The map I wish someone had handed me at the start: catastrophic forgetting and which remedies to trust, how to read the literature without being misled, a reading path, and a two-minute experiment in which a network forgets everything and a 200-image memory saves it.