Barriers To Accelerating The Training Of Artificial Neural Networks – A Systemic Perspective – Meet Manuel Muro

manuel-muroThe real breakthrough for the modern Artificial Intelligence (AI) and Machine Learning (ML) technology explosions started back in 1943 when researchers McCulloch & Pitts came up with a mathematical model to represent that function of the biological neuron; nature’s gift that allows all life to operate and learn over time. Eventually this research would then give birth to the Artificial Neural Network (ANN).

While there are other approaches used today to pursue AI & ML related solutions, the ANN is the oldest and most fundamental approach. While the implementation, a.k.a. the inference, of an ANN is rather straightforward! However, it is the training of the ANN that can get rather complex and more importantly the training can get rather time consuming to the point that finding the answers to the major questions that can help and guide us to a better understand of ANNs is significantly impeded as the time needed to train ANNs can take several hours, weeks and months to complete! Manuel’s presentation takes a look at how our modern computing technology is structured, from a systemic “hardware” perspective, to help better understand the barriers to accelerating the training of ANNs and how those barriers can be properly managed both in the present and going forward.

Meet Manuel Muro (@the_engr_manuel). Formally trained as an electrical engineer at N.C. State, Manuel Muro has been involved in just about every aspect of the design, verification and testing of semiconductor devices particularly Microprocessors and SoC/ASIC/FPGAs along with some Software, RF, Memory, Analog and Mixed-Signal experiences as well. Since 2014, he has been working on the hardware implementation of A.I. systems and more recently has been doing research and development on coming up with novel hardware approaches to significantly accelerate how machine learning is carried out to address both the speed and power consumption issues associated with current approaches.