A seminal result in learning theory characterizes the PAC learnability of binary classes through the Vapnik-Chervonenkis dimension. Extending this characterization to the general multiclass setting has been open since the pioneering works on multiclass PAC learning in the late 1980s. This work resolves this problem: we characterize multiclass PAC learnability ... more >>>

We analyze the Fourier growth, i.e. the $L_1$ Fourier weight at level $k$ (denoted $L_{1,k}$), of read-once regular branching programs.
We prove that every read-once regular branching program $B$ of width $w \in [1,\infty]$ with $s$ accepting states on $n$-bit inputs must have its $L_{1,k}$ bounded by
$$
\min\left\{ ... more >>>

We prove that a modification of Andreev's function is not computable by $(3 + \alpha - \varepsilon) \log{n}$ depth De Morgan formula with $(2\alpha - \varepsilon)\log{n}$ layers of AND gates at the top for any $1/5 > \alpha > 0$ and any constant $\varepsilon > 0$. In order to do ... more >>>