Efficient Inference via Universal LSH Kernel