Seol mar théacs é seo: Speeding up convolutional neural networks with low rank expansions