BEGIN:VCALENDAR VERSION:2.0 PRODID:IEEE vTools.Events//EN CALSCALE:GREGORIAN BEGIN:VTIMEZONE TZID:Europe/Stockholm BEGIN:DAYLIGHT DTSTART:20240331T030000 TZOFFSETFROM:+0100 TZOFFSETTO:+0200 RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=3 TZNAME:CEST END:DAYLIGHT BEGIN:STANDARD DTSTART:20241027T020000 TZOFFSETFROM:+0200 TZOFFSETTO:+0100 RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=10 TZNAME:CET END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTAMP:20240614T091223Z UID:313355A2-2EE2-4864-9117-19B612CC650D DTSTART;TZID=Europe/Stockholm:20240610T150000 DTEND;TZID=Europe/Stockholm:20240610T160000 DESCRIPTION:Title: Speeding up distributed learning: towards low-complexity communication-efficient algorithms with superlinear convergence\n\nSpeake r(s): Subhrakanti Dey\n\nAgenda: \nTitle: Speeding up distributed learning : towards low-complexity communication-efficient algorithms with superline ar convergence\n\nAbstract:\n\nNext generation of networked cyber-physical systems will support a number of application domains e.g. connected auton omous vehicular networks\, collaborative robotics in smart factories\, and many other mission-critical applications. With the advent of massive mach ine-to-machine communication and IoT networks\, huge volumes of data can b e collected and processed with low latency through edge computing faciliti es. Distributed machine learning enables cross-device collaborative learni ng without exchanging raw data\, ensuring privacy and reducing communicati on cost. Learning over wireless networks poses significant challenges due to limited communication bandwidth and channel variability\, limited comp utational resources at the IoT devices\, the heterogeneous nature of distr ibuted data\, and also randomly time-varying network topologies. In this t alk\, we will present (i) low-complexity communication efficient Federated Learning (FL) algorithms based on approximate Newton-type optimization te chniques employed at the local agents\, which achieve superlinear converge nce rate as opposed to linear rates achieved by state-of-the-art gradient descent based algorithms\, and (ii) fully distributed network Newton type algorithms based on a distributed version of the well-known GIANT algorith m. While consensus based distributed optimization algorithms are naturally limited to linear convergence rates\, we will show that one can design fi nite-time consensus based distributed network-Newton type algorithms that can achieve superlinear convergence\, albeit at the cost of increased numb ers of consensus rounds. We will conclude with some new directions on resu lts on zeroth order techniques that can also achieve superlinear convergen ce rates in Federated Learning.\n\nVirtual: https://events.vtools.ieee.org /m/420689 LOCATION:Virtual: https://events.vtools.ieee.org/m/420689 ORGANIZER:amlkel@utu.fi, elisa.barney@ltu.se SEQUENCE:14 SUMMARY:IEEE Sweden Signal Processing Society Chapter\, SPS Day Event - Sem inar by Professor Subhrakanti Dey URL;VALUE=URI:https://events.vtools.ieee.org/m/420689 X-ALT-DESC:Description: <br /><p><strong><span style="font-size: 12.0pt\; c olor: black\;">Title</span></strong><span style="font-size: 12.0pt\; color : black\;">: Speeding up distributed learning: towards low-complexity comm unication-efficient algorithms with superlinear convergence</span></p><br /><br />Agenda: <br /><p><strong><span style="font-size: 12.0pt\; color: b lack\;">Title</span></strong><span style="font-size: 12.0pt\; color: black \;">: Speeding up distributed learning: towards low-complexity communicati on-efficient algorithms with superlinear convergence</span></p>\n<p><stron g><span style="font-size: 12.0pt\; color: black\;">Abstract</span></strong ><span style="font-size: 12.0pt\; color: black\;">: </span></p>\n<p class= "MsoNormal"><span lang="EN-GB" style="font-size: 12.0pt\; color: black\; m so-ansi-language: EN-GB\; mso-fareast-language: EN-US\;">Next generation o f networked cyber-physical &nbsp\;systems will support a number of applica tion domains e.g. connected autonomous vehicular networks\, collaborative robotics in smart factories\, and many other mission-critical applications . With the advent of massive machine-to-machine communication and IoT netw orks\, huge volumes of data can be collected and processed with low latenc y through edge computing facilities.&nbsp\; Distributed machine learning&n bsp\; enables&nbsp\; cross-device&nbsp\; collaborative learning without ex changing raw data\, ensuring privacy and reducing communication cost. Lear ning over wireless networks poses significant challenges due to limited c ommunication bandwidth and channel variability\, limited computational res ources at the IoT devices\, the heterogeneous nature of distributed data\, and also randomly time-varying network topologies. In this talk\, we will present&nbsp\; (i) low-complexity communication efficient Federated Learn ing (FL) algorithms based on approximate Newton-type optimization techniqu es employed at the local agents\, which achieve superlinear convergence ra te as opposed to linear rates achieved by state-of-the-art gradient descen t based algorithms\, and (ii) fully distributed network Newton type algori thms based on a distributed version of the well-known GIANT algorithm. Whi le consensus based distributed optimization algorithms are naturally limit ed to linear convergence rates\, we will show that one can design finite-t ime consensus based distributed network-Newton type algorithms that can ac hieve superlinear convergence\, albeit at the cost of increased numbers of consensus rounds. We will conclude with some new </span><span lang="EN-GB " style="font-size: 12.0pt\; color: black\; mso-ansi-language: EN-GB\; mso -fareast-language: EN-US\;">directions on results on zeroth order techniqu es that can also achieve superlinear convergence rates in Federated Learni ng.</span></p>\n<p class="MsoNormal"><span style="font-size: 12.0pt\; colo r: black\;">&nbsp\;</span></p> END:VEVENT END:VCALENDAR