BEGIN:VCALENDAR VERSION:2.0 PRODID:IEEE vTools.Events//EN CALSCALE:GREGORIAN BEGIN:VTIMEZONE TZID:America/New_York BEGIN:DAYLIGHT DTSTART:20240310T030000 TZOFFSETFROM:-0500 TZOFFSETTO:-0400 RRULE:FREQ=YEARLY;BYDAY=2SU;BYMONTH=3 TZNAME:EDT END:DAYLIGHT BEGIN:STANDARD DTSTART:20241103T010000 TZOFFSETFROM:-0400 TZOFFSETTO:-0500 RRULE:FREQ=YEARLY;BYDAY=1SU;BYMONTH=11 TZNAME:EST END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTAMP:20240802T184135Z UID:770A8BFF-5EE1-456E-882B-BFE08679C5DD DTSTART;TZID=America/New_York:20240801T110000 DTEND;TZID=America/New_York:20240801T120000 DESCRIPTION:Federated learning (FL) is an efficient and privacy-preserving distributed learning paradigm that enables massive edge devices to train m achine learning models collaboratively. Although various communication sch emes and algorithm designs have been proposed to expedite the FL process i n resource-limited wireless networks\, the unreliable nature of wireless c hannels\, device heterogeneity\, and data heterogeneity are still less exp lored. In this talk\, number of solutions solutions will be discussed for addressing the above practical challenges in wireless FL. Firstly\, to tac kle the unreliable wireless channels\, a novel FL framework\, namely FL wi th gradient recycling (FL-GR)\, which recycles the historical gradients of unscheduled and transmission-failure devices to improve the learning perf ormance of FL will be discussed. Secondly\, to solve the heterogeneity iss ues\, partial model aggregation\, knowledge aided learning and adaptive mo del pruning-based FL framework will be explained. Based on our research ex perience\, some open problems of wireless FL will be provided.\n\nSpeaker( s): Professor Arumugam \, \n\nRoom: 460\, Bldg: ENG\, 245 Church Street\, Toronto\, Ontario\, Canada\, M5B 1Z4 LOCATION:Room: 460\, Bldg: ENG\, 245 Church Street\, Toronto\, Ontario\, Ca nada\, M5B 1Z4 ORGANIZER:l5zhao@torontomu.ca SEQUENCE:32 SUMMARY:Federated Learning in Resource Limited Wireless Networks URL;VALUE=URI:https://events.vtools.ieee.org/m/427860 X-ALT-DESC:Description: <br /><p class="MsoNormal" style="margin-bottom: 0c m\; text-align: justify\; text-justify: inter-ideograph\; line-height: nor mal\; mso-layout-grid-align: none\; text-autospace: none\;"><span lang="EN -US" style="mso-bidi-font-family: Calibri\; mso-ansi-language: EN-US\;">Fe derated learning (FL) is an efficient and privacy-preserving distributed l earning paradigm that enables massive edge devices to train machine learni ng models collaboratively. Although various communication schemes and algo rithm designs have been proposed to expedite the FL process in resource-li mited wireless networks\, the unreliable nature of wireless channels\, dev ice heterogeneity\, and data heterogeneity are still less explored. In thi s talk\, number of solutions solutions will be discussed for addressing th e above practical challenges in wireless FL. Firstly\, to tackle the unrel iable wireless channels\, a novel FL framework\, namely FL with gradient r ecycling (FL-GR)\, which recycles the historical gradients of unscheduled and transmission-failure devices to improve the learning performance of FL will be discussed. Secondly\, to solve the heterogeneity issues\, partial model aggregation\, knowledge aided learning and adaptive model pruning-b ased FL framework will be explained. Based on our research experience\, so me open problems of wireless FL will be provided.</span></p>\n<p class="Ms oNormal" style="margin-bottom: 0cm\; text-align: justify\; text-justify: i nter-ideograph\; line-height: normal\; mso-layout-grid-align: none\; text- autospace: none\;">&nbsp\;</p>\n<p class="MsoNormal" style="margin-bottom: 0cm\; text-align: justify\; text-justify: inter-ideograph\; line-height: normal\; mso-layout-grid-align: none\; text-autospace: none\;"><span lang= "EN-US" style="mso-bidi-font-family: Calibri\; mso-ansi-language: EN-US\;" >&nbsp\;</span></p> END:VEVENT END:VCALENDAR