Feature-wise transformations for conditioning models with a context. Neat! Could use entity embedding as "context" to inject a random effect into a model? distill.pub/2018/feature-w…

Interesting article on ways to define "fairness." Predictive parity? False positive rate? Something else? nature.com/articles/d4158…

Clustering in latent space, but using a discriminative loss function instead of one which minimizes reconstruction error. arxiv.org/abs/1805.10795

Using decision trees to estimate the fixed effect in a mixed model. towardsdatascience.com/mixed-effects-…

Reliable Uncertainty Estimates in Deep Neural Networks using Noise Contrastive Priors reddit.com/r/MachineLearn…

AutoKeras, an open-source neural architecture search which like previous approaches shares weights between candidate models, but uses Bayesian optimisation instead of RNNs/evolutionary algos to control the search github.com/jhfjhfj1/autok…

Cool project! Interpolating in latent space with splines is a neat idea too. rajatvd.github.io/Animating-Dood…