Dropout: A Simple Way to Prevent Neural Networks…

Dropout is a compact lesson in turning randomness into regularization. Read it to see how masking neurons during training discourages brittle co-adaptation and approximates an ensemble without paying the cost of running many networks at inference time.

Reading focus: How randomly masking units makes features useful across many different network configurations. Why dropout can be understood as training an ensemble of thinned networks with shared weights. How keep probability and test-time scaling connect stochastic training to deterministic inference.

JMLR 2014. Srivastava et al.. 35 min read, very easy difficulty.