Statistical Physics of Learning

About these Michael Biehl lecture notes

These notes follow Michael Biehl’s mathematical treatment of learning in neural networks through statistical physics. They begin with classic neuron and perceptron models, develop teacher–student learning and replica methods, and continue through layered, deep, unsupervised, and competitive learning.

• McCulloch–Pitts neurons, Hebbian learning, and perceptrons
• Teacher–student models, learning curves, and generalization error
• Statistical mechanics, order parameters, and replica methods
• Layered and committee machines with hidden-unit specialization
• Overparameterization, deep, unsupervised, and competitive learning