High Dimensional Regression with Binary Coefficients. Estimating Squared Error and a Phase Transtition

We consider a sparse linear regression model Y=X\beta^{*}+W where X has a Gaussian entries, W is the noise vector with mean zero Gaussian entries, and \beta^{*} is a binary vector with support size (sparsity) k. Using a novel conditional second moment method we obtain a tight up to a multiplicative constant approximation of the optimal squared error \min_{\beta}\|Y-X\beta\|_{2}, where the minimization is over all k-sparse binary vectors \beta. The approximation reveals interesting structural properties of the underlying regression problem. In particular, a) We establish that n^*=2k\log p/\log (2k/\sigma^{2}+1) is a phase transition point with the following "all-or-nothing" property. When n exceeds n^{*}, (2k)^{-1}\|\beta_{2}-\beta^*\|_0\approx 0, and when n is below n^{*}, (2k)^{-1}\|\beta_{2}-\beta^*\|_0\approx 1, where \beta_2 is the optimal solution achieving the smallest squared error. With this we prove that n^{*} is the asymptotic threshold for recovering \beta^* information theoretically. b) We compute the squared error for an intermediate problem \min_{\beta}\|Y-X\beta\|_{2} where minimization is restricted to vectors \beta with \|\beta-\beta^{*}\|_0=2k \zeta, for \zeta\in [0,1]. We show that a lower bound part \Gamma(\zeta) of the estimate, which corresponds to the estimate based on the first moment method, undergoes a phase transition at three different thresholds, namely n_{\text{inf,1}}=\sigma^2\log p, which is information theoretic bound for recovering \beta^* when k=1 and \sigma is large, then at n^{*} and finally at n_{\text{LASSO/CS}}. c) We establish a certain Overlap Gap Property (OGP) on the space of all binary vectors \beta when n\le ck\log p for sufficiently small constant c. We conjecture that OGP is the source of algorithmic hardness of solving the minimization problem \min_{\beta}\|Y-X\beta\|_{2} in the regime n