BEGIN:VCALENDAR VERSION:2.0 PRODID:IEEE vTools.Events//EN CALSCALE:GREGORIAN BEGIN:VTIMEZONE TZID:America/New_York BEGIN:DAYLIGHT DTSTART:20230312T030000 TZOFFSETFROM:-0500 TZOFFSETTO:-0400 RRULE:FREQ=YEARLY;BYDAY=2SU;BYMONTH=3 TZNAME:EDT END:DAYLIGHT BEGIN:STANDARD DTSTART:20231105T010000 TZOFFSETFROM:-0400 TZOFFSETTO:-0500 RRULE:FREQ=YEARLY;BYDAY=1SU;BYMONTH=11 TZNAME:EST END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTAMP:20231030T152347Z UID:33C1686B-64A6-4FDE-AD42-9C1E4EBACBD1 DTSTART;TZID=America/New_York:20231027T140000 DTEND;TZID=America/New_York:20231027T150000 DESCRIPTION:The typical hardware platform for neural networks operates base d on clocked computation and consists of advanced parallel graphics proces sing units (GPU) and/or application specific integrated circuits (ASIC)\, which are reconfigurable\, multi-purpose and robust. However\, for such pl atforms the input data often needs to be converted to electrical domain\, digitized\, and stored. Furthermore\, a clocked computation system typical ly has a high power consumption\, suffers from a limited speed\, and requi res a large data storage device. To address the ever-increasing demand for more sophisticated and complex AI based systems\, deeper neural networks with a large number of layers and neurons are required\, which result in e ven higher power consumption and longer computation time. Photonic deep ne tworks could address some of\nthese challenges by utilizing the large band width available around the optical carrier and low propagation loss of CMO S-compatible photonic devices and blocks. In this talk\, a low-cost integr ated highly-scalable photonic architecture for implementation of deep neur al networks for image/video/signal classification is presented\, where the input images are taken using an array of pixels and directly processed in the optical\ndomain. The implemented system performs computation by propa gation and\, as such\, is several orders-of-magnitude faster than state-of -the-art clocked based systems and operates at a significantly lower power consumption. This system\, which is scalable to a network with a large nu mber of layers\, performs in-domain processing (i.e. processing in the opt ical domain) and as a result\, opto-electronic conversion\,\nanalog-to-dig ital conversion\, and requirement for a large memory module are eliminated .\n\nCo-sponsored by: Glenn Cowan\n\nSpeaker(s): Dr. Firooz Aflatouni\n\nR oom: 1.162\, Bldg: EV\, 1515 Saint-Catherine St W\, Montreal\, Quebec\, Ca nada\, H3G 1S6 LOCATION:Room: 1.162\, Bldg: EV\, 1515 Saint-Catherine St W\, Montreal\, Qu ebec\, Canada\, H3G 1S6 ORGANIZER:gcowan@ece.concordia.ca SEQUENCE:18 SUMMARY:Integrated photonic deep networks for image classification URL;VALUE=URI:https://events.vtools.ieee.org/m/378537 X-ALT-DESC:Description: <br /><p>The typical hardware platform for neural n etworks operates based on clocked computation and consists of advanced par allel graphics processing units (GPU) and/or application specific integrat ed circuits (ASIC)\, which are reconfigurable\, multi-purpose and robust. However\, for such platforms the input data often needs to be converted to electrical domain\, digitized\, and stored. Furthermore\, a clocked compu tation system typically has a high power consumption\, suffers from a limi ted speed\, and requires a large data storage device. To address the ever- increasing demand for more sophisticated and complex AI based systems\, de eper neural networks with a large number of layers and neurons are require d\, which result in even higher power consumption and longer computation t ime. Photonic deep networks could address some of<br />these challenges by utilizing the large bandwidth available around the optical carrier and lo w propagation loss of CMOS-compatible photonic devices and blocks. In this talk\, a low-cost integrated highly-scalable photonic architecture for im plementation of deep neural networks for image/video/signal classification is presented\, where the input images are taken using an array of pixels and directly processed in the optical<br />domain. The implemented system performs computation by propagation and\, as such\, is several orders-of-m agnitude faster than state-of-the-art clocked based systems and operates a t a significantly lower power consumption. This system\, which is scalable to a network with a large number of layers\, performs in-domain processin g (i.e. processing in the optical domain) and as a result\, opto-electroni c conversion\,<br />analog-to-digital conversion\, and requirement for a l arge memory module are eliminated.</p> END:VEVENT END:VCALENDAR