Delivered invited talk at CRI - UCSD: Efficient Memory-usage Techniques in Deep Neural Networks