BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Los_Angeles
X-LIC-LOCATION:America/Los_Angeles
BEGIN:DAYLIGHT
TZOFFSETFROM:-0800
TZOFFSETTO:-0700
TZNAME:PDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0700
TZOFFSETTO:-0800
TZNAME:PST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20240626T180034Z
LOCATION:3001\, 3rd Floor
DTSTART;TZID=America/Los_Angeles:20240626T103000
DTEND;TZID=America/Los_Angeles:20240626T104500
UID:dac_DAC 2024_sess102_RESEARCH150@linklings.com
SUMMARY:Zeroth-Order Optimization of Optical Neural Networks with Linear C
 ombination Natural Gradient and Calibrated Model
DESCRIPTION:Research Manuscript\n\nHiroshi Sawada, Kazuo Aoyama, and Kohei
  Ikeda (NTT Corporation)\n\nOptical neural networks (ONNs) have attracted 
 great attention due to their low energy consumption and high-speed process
 ing. The usual neural network training scheme leads to poor performance fo
 r ONNs because of their special parameterization and fabrication variation
 s. This paper contributes to extend zeroth-order (ZO) optimization, which 
 can be used to train such ONNs, in two ways. The first is to propose linea
 r combination natural gradient, which mitigates the optimization difficult
 y caused by the special parameterization of an ONN. The second is to gener
 ate a guided direction vector by calibration for better guessing than rand
 om vectors generated in ZO optimization. Experimental results show that th
 e two extensions significantly outperformed the existing ZO optimization a
 nd related methods with little computational overhead.\n\nTopic: AI\n\nKey
 word: AI/ML Algorithms\n\nSession Chairs: Parivesh Choudhary (Synopsys) an
 d Anca Molnos (CEA-List)
END:VEVENT
END:VCALENDAR