On the well-spread property and its relation to linear regression
This addresses a theoretical limitation in robust linear regression for researchers in statistics and machine learning, but it is incremental as it builds on prior work on well-spreadness.
The paper shows that consistent recovery in robust linear regression is impossible for certain design matrices lacking the well-spread property, and it proves that certifying well-spreadness for Gaussian matrices is efficient with quadratic observations but computationally hard with fewer observations.
We consider the robust linear regression model $\boldsymbol{y} = Xβ^* + \boldsymbolη$, where an adversary oblivious to the design $X \in \mathbb{R}^{n \times d}$ may choose $\boldsymbolη$ to corrupt all but a (possibly vanishing) fraction of the observations $\boldsymbol{y}$ in an arbitrary way. Recent work [dLN+21, dNS21] has introduced efficient algorithms for consistent recovery of the parameter vector. These algorithms crucially rely on the design matrix being well-spread (a matrix is well-spread if its column span is far from any sparse vector). In this paper, we show that there exists a family of design matrices lacking well-spreadness such that consistent recovery of the parameter vector in the above robust linear regression model is information-theoretically impossible. We further investigate the average-case time complexity of certifying well-spreadness of random matrices. We show that it is possible to efficiently certify whether a given $n$-by-$d$ Gaussian matrix is well-spread if the number of observations is quadratic in the ambient dimension. We complement this result by showing rigorous evidence -- in the form of a lower bound against low-degree polynomials -- of the computational hardness of this same certification problem when the number of observations is $o(d^2)$.