Markov Decision Processes
Markov Decision Processes are Finite State Machines with four key components: state, action, transition function, reward function. They run into the curse of dimensionality.
Winograd Schema Challenge
The Winograd Schema Challenge aims to address the flaws in the Turing Test to determine if an AI is "human level".
Kalman filters are a simple but powerful technique for determining the most likely current state of an object in motion.
Game theory is the study of equilibria-based solutions.
False Discovery Rates
How to reduce the noise of making multiple comparisons.
Transfer learning is powerful and underutilized.
Reinforcement learning is a type of unsupervised state-action-transition-reward training mainly used to learn games right now.
Optimal decision-making with POMDPs
POMDPs are Markov decision processes that have to deal with a partially obervable game.
Be more rigorous in real life by thinking in terms of priors, posteriors, and updates.