Hardware Architectures for Deep Learning

Masoud Daneshtalab is a tenured associate professor at Mälardalen University (MDH) in Sweden, an adjunct professor at Tallinn University of Technology (TalTech) in Estonia, and sits on the board of directors of Euromicro. His research interests include interconnection networks, brain-like computing, and deep learning architectures. He has published over 300-refereed papers.Mehdi Modarressi is an assistant professor at the Department of Electrical and Computer Engineering, University of Tehran, Iran. He is the founder and director of the Parallel and Network-based Processing research laboratory at the University of Tehran, where he leads several industrial and research projects on deep learning-based embedded system design and implementation.

• Part I: Deep learning and neural networks: concepts and modelsChapter 1: An introduction to artificial neural networksChapter 2: Hardware acceleration for recurrent neural networksChapter 3: Feedforward neural networks on massively parallel architectures Part II: Deep learning and approximate data representationChapter 4: Stochastic-binary convolutional neural networks with deterministic bit-streamsChapter 5: Binary neural networks Part III: Deep learning and model sparsityChapter 6: Hardware and software techniques for sparse deep neural networksChapter 7: Computation reuse-aware accelerator for neural networks Part IV: Convolutional neural networks for embedded systemsChapter 8: CNN agnostic accelerator design for low latency inference on FPGAsChapter 9: Iterative convolutional neural network (ICNN): an iterative CNN solution for low power and real-time systems Part V: Deep learning on analog acceleratorsChapter 10: Mixed-signal neuromorphic platform design for streaming biomedical signal processingChapter 11: Inverter-based memristive neuromorphic circuit for ultra-low-power IoT smart applications