Four short links: 23 September 2016
On Reproducibility, Robot Monkey Startup, Stealing Predictive Models, and GPU Equivalence
- The Winds Have Changed — wonderfully constructed rebuttal to a self-serving “those nasty people saying they can’t reproduce our media-packaged research findings are just terrible stone-throwers, ignore them” editorial, which builds and builds and should have you reaching for a can of petrol and a lighter by the end.
- Kindred AI — using artificial intelligence and high-tech exoskeleton suits to allow humans—and, at least according to one description of the technology, monkeys, too—to control and train an army of intelligent robots. Planet of the Apes inches its way closer to being.
- Stealing Machine Learning Models via Prediction APIs — Unlike in classical learning theory settings, ML-as-a-service offerings may accept partial feature vectors as inputs and include confidence values with predictions. Given these practices, we show simple, efficient attacks that extract target ML models with near-perfect fidelity for popular model classes including logistic regression, neural networks, and decision trees. We demonstrate these attacks against the online services of BigML and Amazon Machine Learning.
- GPU Equivalence for Deep Learning — In our own testing, we’ve found that one GPU server is about as fast as 400 CPU cores for running the algorithms we’re using. The article itself is an unremarkable overview, but this anecdatum leapt out at me.