Balancing storage, efficient computing, accuracy, mapping to silicon and power consumption is a challenge when trying to use low bit-depth neural network. Case studies encompassing anomaly detector and classifier model design are complex tasks if neural networks are investigated targeting ultra-low power devices such as sensors and microcontrollers. Deeply Quantized Neural Networks (DQNNs) offer the most interesting approach to these tasks. The design and the training of DQNN also is not a trivial task. Unfortunately, current off the shelf microcontrollers are not yet able to exploit their potentialities. Realization of custom energy efficient hardware accelerators sometime may represent a viable alternative in terms of energy efficiency, especially applied to a raising field such as in-sensing neural computing. Hybrid Neural Networks variants developed with experimental deep learning tools, can achieve interesting accuracies compared to more traditional design approaches. In this talk all those aspects will be discussed with reference to latest efforts of ST including a) tools for efficient deployment on micro controllers for image processing b) custom ultra-low power hardware circuitry for real-time execution of the Hybrid Neural Network with traditional CMOS technologies and implemented with field-programmable gate array, c) latest ST solutions for in sensor deep learning computing. Part of the talk will include associated demo and code inspection.
Davide Denaro joined STMicroelectronics since 2004 after working for various IT companies.
He was engaged in numerous research projects in the Computer Vision, Wearables and Iot domains.
Currently He is Senior Software Designer Engineer in the Artificial Intelligent Software and Tools Group.
He got master degree in Computer Science at Università degli Studi di Catania.