We develop low-power AI on the edge, hardware and software solutions for the retail world.

If you believed event-based sensing and processing are the future, this is where we engineer the theories into reality.

I developed the first adversarial technique that successfully fooled HMAX, a convolution neural network, with my intern Maysun Hasan. The technique was published in Maysun's 2010 MIT theis: "Using Genetic Algorithm to 'Fool' HMAX Object Recognition Model." Our work preceded the CVPR2015 paper "Deep Neural Networks are Easily Fooled" by more than 5 years, which used exactly the same technique.My technical specialty is in low-power algorithm design and low-power learning rules. I am a trained neuroscientist both in electrophysiology and event-based/spiking neural network. I am very keen on identifying and developing talents. Some of the individuals that I have recruited directly out of academia included all 3 future founders of Nervana Systems, and a future 30-under-30 recipient.

Exhibiting artist in photography.

I was in charge of the international team that was responsible for camera tuning of all camera phones from Qualcomm. I developed and patented the industry-first machine learning based automatic camera tuning tool with my intern Kathik Ravirala who later joined the team full-time.