Machine learning guys do things like linear regression, neural networks and support vector machines. Logic guys do relational structures, formulas, automata and such stuff. These two type of guys hardly ever meet to work together. So few logic guys know that neural networks are sometimes considered with only binary weights and signals, which makes them exactly circuits. And few machine learning guys know quantitative logics good enough to recognize thier standard neural network in a formula. In this talk, I will define very basic concepts used in machine learning in a way that exposes their logic counterparts. I will then argue that such relationship can be beneficial for both fields - providing theorems and deeper understanding to machine learning and practically relevant questions and experimental results to logic.