Nearly optimal deterministic algorithm for sparse Walsh-Hadamard transform

Abstract

For every fixed constant α > 0, we design an algorithm for computing the k-sparse Walsh-Hadamard transform (i.e., Discrete Fourier Transform over the Boolean cube) of an iV-dimensional vector x ∈ ℝN in time k1+α(log N)°(1) Specifically, the algorithm is given query access to x and computes a k-sparse x ∈ ℝ.N satisfying ||x-◯||1 ≤ c||◯ - Hk(◯)\\1, for an absolute constant c > 0, where x is the transform of x and Hk(x) is its best k-sparse approximation. Our algorithm is fully deterministic and only uses non-adaptive queries to x (i.e., all queries are determined and performed in parallel when the algorithm starts). An important technical tool that we use is a construction of nearly optimal and linear lossless condensers which is a careful instantiation of the GUV condenser (Guruswami, Umans, Vadhan, JACM 2009). Moreover, we design a deterministic and non-adaptive l1/l1 compressed sensing scheme based on general lossless condensers that is equipped with a fast reconstruction algorithm running in time k1+α(logN)O(1) (for the GUV-based condenser) and is of independent interest. Our scheme significantly simplifies and improves an earlier expander-based construction due to Berinde, Gilbert, Indyk, Karloff, Strauss (Allerton 2008). Our methods use linear lossless condensers in a black box fashion; therefore, any future improvement on explicit constructions of such condensers would immediately translate to improved parameters in our framework (potentially leading to k(log N)(1) reconstruction time with a reduced exponent in the poly-logarithmic factor, and eliminating the extra parameter α). By allowing the algorithm to use randomness, while still using non-adaptive queries, the running time of the algorithm can be improved to Ō(k log3 N).