BEGIN:VCALENDAR VERSION:2.0 PRODID:IEEE vTools.Events//EN CALSCALE:GREGORIAN BEGIN:VTIMEZONE TZID:Asia/Shanghai BEGIN:STANDARD DTSTART:19910915T010000 TZOFFSETFROM:+0900 TZOFFSETTO:+0800 TZNAME:CST END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTAMP:20260213T023612Z UID:A445ED49-30A4-4050-892E-B11B2C7712E1 DTSTART;TZID=Asia/Shanghai:20260205T160000 DTEND;TZID=Asia/Shanghai:20260205T170000 DESCRIPTION:This Distinguished Lecturer presentation mainly discusses the m ulti-agentic AI collaboration. With the development of artificial intellig ence (AI)\, deep neural network (DNN) inference has become a crucial compu tational task in edge intelligence networks. However\, due to the limited computing capacity and energy supplying of IoT devices\, and the task offl oading latency between IoT devices and edge servers as well\, the local in ference or centralized inference can hardly meet the requirements of low l atency and high energy efficiency. For this\, distributed collaborative in ference provides a promising solution based on the multi-agentic AI collab oration among IoT devices and edge servers. Notably\, the multi-agentic AI collaboration has to face key challenges such as unbalanced resource sche duling\, redundant data exchange\, and heterogeneous accuracy demands acro ss different computational domains. In this talk\, we thus focus on how to realize the end-end collaboration among IoT devices and end-edge collabor ation between IoT devices and edge servers. For the end-to-end collaborati ve inference\, a padding-aware IoT device collaboration framework is propo sed to achieve efficient data interaction and synchronized computation amo ng devices. By jointly optimizing model partitioning and padding data tran smission strategies\, a latency minimization model is established and tran sformed into a solvable linear programming form. For the end–edge collab orative inference\, an accuracy-aware multi-branch collaborative inference model is proposed to cope with diverse accuracy requirements and heteroge neous computational capacities. A mixed-integer nonlinear optimization mod el is formulated to jointly optimize DNN branch selection\, task partition ing\, and resource allocation for computation and communication. To reduce the computational complexity\, an efficient algorithm based on hierarchic al decomposition and proportional–integral–derivative (PID) search is developed\, achieving a dynamic trade-off between energy consumption and i nference accuracy. Finally\, we develop the prototype system on the NVIDIA Jetson platform to validate the effectiveness and practicality of the pro posed schemes in collaborative inference scenarios.\n\nSpeaker(s): Prof. L i Ping Qian \, \n\nAcademic Lecture Hall\, Engineering Building No. 2\, Sc hool of Automation\, Guangdong University of Technology\, Guangzhou\, Guan gdong\, China LOCATION:Academic Lecture Hall\, Engineering Building No. 2\, School of Aut omation\, Guangdong University of Technology\, Guangzhou\, Guangdong\, Chi na ORGANIZER:zsl@szu.edu.cn SEQUENCE:16 SUMMARY: Edge intelligence networks: from independent computation offloadin g to multi-agentic AI collaboration URL;VALUE=URI:https://events.vtools.ieee.org/m/537033 X-ALT-DESC:Description: <br /><p class="MsoNormal"><span lang="EN-US" style ="font-size: 12.0pt\; mso-bidi-font-size: 11.0pt\; font-family: 'Times New Roman'\,serif\;">This Distinguished Lecturer presentation mainly discusse s the </span><span lang="EN-US" style="font-size: 12.0pt\; font-family: 'T imes New Roman'\,serif\;">multi-agentic AI collaboration</span><span lang= "EN-US" style="font-size: 12.0pt\; mso-bidi-font-size: 11.0pt\; font-famil y: 'Times New Roman'\,serif\;">. </span><span lang="EN-US" style="font-siz e: 12.0pt\; font-family: 'Times New Roman'\,serif\;">With the development of artificial intelligence (AI)\, deep neural network (DNN) inference has become a crucial computational task in edge intelligence networks. However \, due to the limited computing capacity and energy supplying of IoT devic es\, and the task offloading latency between IoT devices and edge servers as well\, the local inference or centralized inference can hardly meet the requirements of low latency and high energy efficiency. For this\, distri buted collaborative inference provides a promising solution based on the m ulti-agentic AI collaboration among IoT devices and edge servers. Notably\ , the multi-agentic AI collaboration has to face </span><span lang="EN-US" style="font-size: 12.0pt\; font-family: 'Times New Roman'\,serif\; mso-fa reast-font-family: Arial\; color: #252525\;">key challenges such as unbala nced resource scheduling\, redundant data exchange\, and heterogeneous acc uracy demands across different computational domains.<span style="mso-spac erun: yes\;">&nbsp\; </span>In this talk\, we thus focus on how to realize the end-end collaboration among IoT devices and end-edge collaboration be tween IoT devices and edge servers. For the end-to-end collaborative infer ence\, a padding-aware IoT device collaboration framework is proposed to a chieve efficient data interaction and synchronized computation among devic es. By jointly optimizing model partitioning and padding data transmission strategies\, a latency minimization model is established and transformed into a solvable linear programming form. For the end&ndash\;edge collabora tive inference\, an accuracy-aware multi-branch collaborative inference mo del is proposed to cope with diverse accuracy</span><span lang="EN-US" sty le="font-size: 12.0pt\; font-family: 'Times New Roman'\,serif\; color: #25 2525\;"> </span><span lang="EN-US" style="font-size: 12.0pt\; font-family: 'Times New Roman'\,serif\; mso-fareast-font-family: Arial\; color: #25252 5\;">requirements and heterogeneous computational capacities. A mixed-inte ger nonlinear optimization model is formulated to jointly optimize DNN bra nch selection\, task partitioning\, and resource allocation for computatio n and communication. To reduce the computational complexity\, an efficient algorithm based on hierarchical decomposition and proportional&ndash\;int egral&ndash\;derivative (PID) search is developed\, achieving a dynamic tr ade-off between energy consumption and inference accuracy. Finally\, we de velop the prototype system on the NVIDIA Jetson platform to validate the e ffectiveness and practicality of the proposed schemes in collaborative inf erence scenarios.</span></p> END:VEVENT END:VCALENDAR