Complexity Theoretic Limitations on Learning Halfspaces

Abstract: We study the problem of agnostically learning halfspaces which is defined by a fixed but unknown distribution $\mathcal{D}$ on $\mathbb{Q}^n\times {\pm 1}$. We define $\mathrm{Err}{\mathrm{HALF}}(\mathcal{D})$ as the least error of a halfspace classifier for $\mathcal{D}$. A learner who can access $\mathcal{D}$ has to return a hypothesis whose error is small compared to $\mathrm{Err}{\mathrm{HALF}}(\mathcal{D})$. Using the recently developed method of the author, Linial and Shalev-Shwartz we prove hardness of learning results under a natural assumption on the complexity of refuting random $K$-$\mathrm{XOR}$ formulas. We show that no efficient learning algorithm has non-trivial worst-case performance even under the guarantees that $\mathrm{Err}_{\mathrm{HALF}}(\mathcal{D}) \le \eta$ for arbitrarily small constant $\eta>0$, and that $\mathcal{D}$ is supported in ${\pm 1}^n\times {\pm 1}$. Namely, even under these favorable conditions its error must be $\ge \frac{1}{2}-\frac{1}{n^c}$ for every $c>0$. In particular, no efficient algorithm can achieve a constant approximation ratio. Under a stronger version of the assumption (where $K$ can be poly-logarithmic in $n$), we can take $\eta = 2^{-\log^{1-\nu}(n)}$ for arbitrarily small $\nu>0$. Interestingly, this is even stronger than the best known lower bounds (Arora et. al. 1993, Feldamn et. al. 2006, Guruswami and Raghavendra 2006) for the case that the learner is restricted to return a halfspace classifier (i.e. proper learning).