* I've been studying Brazilian Portuguese for a couple of years. It's the first Romance language that I've studied in depth (I took French in high school but ne

* I've been studying Brazilian Portuguese for a couple of years. It's the first Romance language that I've studied in depth (I took French in high school but never progressed very far). I've become fascinated by Romance language grammars, and Portuguese has grammar in spades, e.g. tons of conjugations like the future subjunctive, personal infinitive, imperfect, etc. The great thing is that once you've studied one Romance language, you almost always get an automatic headstart in another. (except for outliers like Romanian). For example, through my study of Portuguese I've found that I automatically gotten (written) Spanish for free. I can now read simple Spanish texts even though I've never studied the language (though I still have to beware of false cognates; also, more complex articles like those in El País still stump me.) * The other thing I've been investigating is meta-learning techniques for predictive modeling, like ensembling, stacking and boosting (the kinds of techniques used in ML competitions) -- basically ways to combine statistical models to extract more signal without overfitting. At some point individual statistical models are limited in their ability to extract signal from data (even though the signals are there) -- and neural networks are a step too far because of the data volume requirements -- so learning how to combine multiple models is the way to go. None of this stuff is new and I'm a few years late to the game but better late than never.