Computational and Sample Tradeoffs via Convex Relaxation