Fourier Analysis
and Boolean Function Learning
Jeff Jackson
Fourier analysis has informed a
number of positive and negative results
regarding the learnability
of Boolean functions. In turn, learning
research has led to the discovery of Fourier
properties of various
Boolean function
classes. I will
survey Fourier properties and
associated learning results for several
interesting Boolean classes,
including monotone functions, disjunctive
normal form (DNF) expressions,
constant-depth circuits more generally, and halfspaces. Some flavor of
the underlying Fourier analysis will be
included.