identity goals

from 2017: remember things I am at least somewhat an expert in, that other people may not have seen:

• philosophy of bayesian statistics. priors, likelihoods, subjective, objective, empirical bayes, the importance of subjective model choice as much as prior choice.
• graphical models. reasoning about probabilisies in a structured and computationally tractable way. message passing. variational inference. black-box variational inference.
• probabilistic programming. Types of PPLs in the abstract (declarative BLOG-style, imperative Church-style, propositional Infer.NET-stle, BayesDB), stochastic memoization, inference by program analysis, MCMC, BBVI, likelihood weighting, inference networks, etc.
• nonparametric Bayes. Stochastic processes, Gaussian processes, Dirichlet process models, etc.
• deep learning. backpropagation. convex and non-convex optimization (gradient ascent, Newton's method, lbfgs, natural gradient, preconditioning and adaptive preconditioning, momentum, Adagrad, Adam, etc.). convnets and fully convolutional nets. various nonlinearities. the history of supervised training and unsupervised pretraining.
• deep generative models. VAEs and GANs. generally optimizing things according to a variational bound. the wake-sleep algorithm, IWAEs, normalizing flows, structured VAEs, etc.
• RL and deep RL. Q learning, value iteration, policy iteration, policy gradient, TD learning, actor-critic methods, TRPO.
• 'basic' ML: linear/logistic regression, GLMs, PCA, SVMs, kernels.
• functional programming. category theory. Haskell.
• a long list of research ideas in bayesflow.org, many of which are still quite plausible.
• music theory: chords, progressions, harmony, melody, and applied skills in singing and violining.