Steven K Esser           

Technical lead for algorithms development in Brain-inspired Computing group.  Research focuses on developing deep learning methods to train networks suitable for deployment on emerging ultra low power inference hardware.  This has included creating networks with extremely low-precision and limited connectivity (spiking neurons, trinary weights, 128 inputs per neuron) that achieved near state-of-the-art accuracy while running on the 65 mW TrueNorth chip.  Recent work has extended low precision methods to improve performance as well as to operate at different precision levels, allowing customization for various hardware platforms.