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:20240626T180002Z
LOCATION:3001\, 3rd Floor
DTSTART;TZID=America/Los_Angeles:20240625T103000
DTEND;TZID=America/Los_Angeles:20240625T120000
UID:dac_DAC 2024_sess103@linklings.com
SUMMARY:Empowering Edge Intelligence: When IoT Devices Meet AI
DESCRIPTION:Research Manuscript\n\nReal-life application scenarios raise t
 hree important demands on IoT devices: higher accuracy (smarter), higher p
 erformance (faster), and lower resource consumption (more efficient). This
  session explores the hardware-software co-optimization of intelligent alg
 orithms for diverse IoT devices and application scenarios, making IoT devi
 ces smarter, faster, and more efficient. The papers address low-power data
  processing on edge devices, Processing-In-Memory architecture design, rea
 l-time DNN scheduling on MCUs, and a practical deployment of learning-base
 d Heating, Ventilation, and Air Conditioning (HVAC) controllers.\n\nDeep H
 armonic Finesse: Signal Separation in Wearable Systems with Limited Data\n
 \nWe present a method, referred to as Deep Harmonic Finesse (DHF), for sep
 aration of non-stationary quasi-periodic signals when limited data is avai
 lable. The problem frequently arises in wearable systems in which, a combi
 nation of quasi-periodic physiological phenomena give rise to the sensed s
 ignal,...\n\n\nMahya Saffarpour, Weitai Qian, Kourosh Vali, Begum Kasap, H
 erman Hedriana, and Soheil Ghiasi (University of California, Davis)\n-----
 ----------------\nGraph Neural Networks Automated Design and Deployment on
  Device-Edge Co-Inference Systems\n\nThe key to device-edge co-inference p
 aradigm is to partition models into computation-friendly and computation-i
 ntensive parts across device and edge, respectively. However, for Graph Ne
 ural Networks (GNNs), partitioning without architecture exploration is ine
 ffective due to various computational-com...\n\n\nAo Zhou, Jianlei Yang, T
 ong Qiao, and Yingjie Qi (Beihang University); Zhi Yang (Peking University
 ); and Weisheng Zhao and Chunming Hu (Beihang University)\n---------------
 ------\nEPIM: Efficient Processing-In-Memory Accelerators based on Epitome
 \n\nThe exploration of Processing-In-Memory (PIM) accelerators has garnere
 d significant attention within the research community. However, the utiliz
 ation of large-scale neural networks on Processing-In-Memory (PIM) acceler
 ators encounters challenges due to constrained on-chip memory capacity. To
  tackle ...\n\n\nChenyu Wang (Princeton University); Zhen Dong (University
  of California, Berkeley); Daquan Zhou (Bytedance Inc.); Zhenhua Zhu and Y
 u Wang (Tsinghua University); Jiashi Feng (Bytedance Inc.); and Kurt Keutz
 er (University of California, Berkeley)\n---------------------\nRT-MDM: Re
 al-Time Scheduling Framework for Multi-DNN on MCU Using External Memory\n\
 nAs the application scope of DNNs executed on microcontroller units (MCUs)
  extends to time-critical systems, it becomes important to ensure timing g
 uarantees for increasing demand of DNN inferences. To this end, this paper
  proposes RT-MDM, the first real-time scheduling framework for multiple DN
 N tas...\n\n\nSukmin Kang, Seongtae Lee, and Hyunwoo Koo (Sungkyunkwan Uni
 versity); Hoon Sung Chwa (Daegu Gyeongbuk Institute of Science and Technol
 ogy); and Jinkyu Lee (Sungkyunkwan University)\n---------------------\nLEA
 F: An Adaptation Framework against Noisy Data on Edge through Ultra Low-Co
 st Training\n\nIn real-world neural network deployments, incoming data oft
 en contains noise and imperfections. Retraining on resource-constrained ed
 ge devices becomes essential to maintain performance. To tackle this chall
 enge, we introduce LEAF, a hardware-efficient framework designed for adapt
 ing to degraded ima...\n\n\nZihan Xia (University of California, San Diego
 ); Jinwook Kim (SK hynix); and Mingu Kang (University of California, San D
 iego)\n---------------------\nGo Beyond Black-box Policies: Rethinking the
  Design of Learning Agent for Interpretable and Verifiable HVAC Control\n\
 nRecent research has shown the potential of Model-based Reinforcement Lear
 ning (MBRL) to enhance energy efficiency of Heating, Ventilation, and Air 
 Conditioning (HVAC) systems. However, existing methods rely on black-box t
 hermal dynamics models and stochastic optimizers, lacking reliability guar
 antee...\n\n\nZhiyu An, Xianzhong Ding, and Wan Du (University of Californ
 ia, Merced)\n\nTopic: AI\n\nKeyword: AI/ML Algorithms\n\nSession Chairs: H
 ongyang Jia (Tsinghua University) and Grace Li Zhang (Technische Universit
 ät Darmstadt)
END:VEVENT
END:VCALENDAR