Meta-complexity to learning beyond P/poly

Ascertain whether easiness results for meta-complexity problems that currently yield agnostic learnability of P/poly from random examples can be extended to obtain agnostic learning algorithms for typical restricted circuit classes such as ACC^0, TC^0, and NC^1.

Background

Recent results show that certain meta-complexity easiness assumptions imply agnostic learnability of P/poly over polynomially samplable distributions. Extending such implications to restricted classes would deepen the connections between meta-complexity and learning and potentially yield new algorithms for long-standing open learning problems for circuit classes.