BEGIN:VCALENDAR VERSION:2.0 PRODID:IEEE vTools.Events//EN CALSCALE:GREGORIAN BEGIN:VTIMEZONE TZID:America/Denver BEGIN:DAYLIGHT DTSTART:20240310T030000 TZOFFSETFROM:-0700 TZOFFSETTO:-0600 RRULE:FREQ=YEARLY;BYDAY=2SU;BYMONTH=3 TZNAME:MDT END:DAYLIGHT BEGIN:STANDARD DTSTART:20241103T010000 TZOFFSETFROM:-0600 TZOFFSETTO:-0700 RRULE:FREQ=YEARLY;BYDAY=1SU;BYMONTH=11 TZNAME:MST END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTAMP:20240621T165007Z UID:4C914FF1-1600-4FAD-9038-231CB11BF579 DTSTART;TZID=America/Denver:20240620T180000 DTEND;TZID=America/Denver:20240620T200000 DESCRIPTION:We consider the intersection of networks and machine learning i n two contexts. In the first\,[]\nthe data of interest for ML is in the fo rm of a large\, complex network. Here\, Graph Neural\nNetworks (GNNs) util ize message passing for neighborhood aggregation to capture graph\ntopolog y\, while Graph Embedding based Neural Networks (GENNs) distill essential graph\ninformation into a concise representation suitable for traditional neural architectures. The\nsecond is when the training and/or the evaluati on phase of ML must be carried out over a\ndistributed environment such as an IoT network. These environments pose challenges due to\nlimited storag e\, communication and power\, complicating the deployment of complex ML\nm odels and impeding real-time decision-making. Our innovations in graph-coo rdinate based\nstrategies\, TCNN and DVCNN\, help sidestep the computation al challenges faced by competing\nalgorithms. Experimental results\, bench marked against the Open Graph Benchmark\nLeaderboard\, demonstrate that TC NN and DVCNN require orders of magnitude fewer\nparameters than any neural network method currently listed in the OGBN Leaderboard for both\nOGBN Pr oteins and OGBN-Products datasets.\n\nSpeaker(s): Anura\, \n\nAgenda: \n6: 00 pm Doors Open\n\n6:30 pm Online Broadcast starts\n\n6:45 pm Main Presen tation\n\n8:00 End\n\nRoom: C120\, Bldg: Engineering\, 400 Isotope Drive\, Colorado State University\, Fort Collins\, Colorado\, United States\, 805 21\, Virtual: https://events.vtools.ieee.org/m/423298 LOCATION:Room: C120\, Bldg: Engineering\, 400 Isotope Drive\, Colorado Stat e University\, Fort Collins\, Colorado\, United States\, 80521\, Virtual: https://events.vtools.ieee.org/m/423298 ORGANIZER:rtoftness@gmail.com SEQUENCE:25 SUMMARY:Machine Learning and Networks - Challenges\, Solutions and Tradeoff s URL;VALUE=URI:https://events.vtools.ieee.org/m/423298 X-ALT-DESC:Description: <br /><p>We consider the intersection of networks a nd machine learning in two contexts. In the first\,<img style="float: righ t\;" src="https://events.vtools.ieee.org/vtools_ui/media/display/98b41db7- 6f56-436b-8ea6-be5c6bac6d6c" alt="" width="210" height="315"><br>the data of interest for ML is in the form of a large\, complex network. Here\, Gra ph Neural<br>Networks (GNNs) utilize message passing for neighborhood aggr egation to capture graph<br>topology\, while Graph Embedding based Neural Networks (GENNs) distill essential graph<br>information into a concise rep resentation suitable for traditional neural architectures. The<br>second i s when the training and/or the evaluation phase of ML must be carried out over a<br>distributed environment such as an IoT network. These environmen ts pose challenges due to<br>limited storage\, communication and power\, c omplicating the deployment of complex ML<br>models and impeding real-time decision-making. Our innovations in graph-coordinate based<br>strategies\, TCNN and DVCNN\, help sidestep the computational challenges faced by comp eting<br>algorithms. Experimental results\, benchmarked against the Open G raph Benchmark<br>Leaderboard\, demonstrate that TCNN and DVCNN require or ders of magnitude fewer<br>parameters than any neural network method curre ntly listed in the OGBN Leaderboard for both<br>OGBN Proteins and OGBN-Pro ducts datasets.</p><br /><br />Agenda: <br /><p>6:00 pm Doors Open</p>\n<p >6:30 pm Online Broadcast starts</p>\n<p>6:45 pm Main Presentation</p>\n<p >8:00 End</p> END:VEVENT END:VCALENDAR