主講人： Wujie Wen教授

時(shí)  間：2018年12月12日（周三）10:00

地  點(diǎn)：機(jī)電信息樓601

題目：Machine Vision, NOT Human Vision, Guided Compression towards Energy-Efficient and Robust Deep Learning Systems

內(nèi)容簡(jiǎn)介：

    In this talk, we first demonstrate that well-known data compression approaches (i.e. JPEG), which are often centered on ``human vision”, exhibit low efficiency to address those challenges. Interestingly, we find that root cause is that machine (or deep learning) and human view the compression quality of an image differently. Based on this critical observation, we for the first time develop a ``machine vision” guided image compression framework tailored for deep learning applications (e.g. attaining high accuracy at a much higher compression rate to lower data transmission cost), by embracing the nature of deep-cascaded information process mechanism of DNN architecture. Then we also show how to seamlessly integrate the defense into data compression to protect DNNs against emerging adversarial attacks, with good balance between accuracy and defense efficiency. We hope our advocated ``machine vision”, a radically different perspective to re-architecture existing techniques, can advance our understandings on developing more energy-efficient and robust deep learning systems.  

主講人簡(jiǎn)介：

    Wujie Wen is currently an assistant professor in ECE department at Florida International University (FIU), Miami, FL. He received his Ph.D. in Electrical and Computer Engineering from University of Pittsburgh in 2015. He earned his B.S. and M.S. degrees in electronic engineering from the Beijing Jiaotong University and Tsinghua University, Beijing, China, in 2006 and 2010, respectively. Before he joined FIU in 2015, he also worked with AMD and Broadcom for various engineer and intern positions. His current research interests include deep learning security/hardware acceleration, neuromorphic computing, and circuit-architecture design for emerging memory technologies. Dr. Wen servers as the associate editor of Neurocomputing, General Chair of ISVLSI 2019 (Miami), Technical Program Chair of ISVLSI 2018 (Hong Kong), as well as program committee for many conferences such as DAC, ICCAD, ASP-DAC etc. He received best paper nominations from ASP-DAC2018, DATE2016 and DAC2014. He was also the recipient of the 49th DAC A. Richard Newton Graduate Scholarship, the most prestigious Ph.D. scholarship (one awardee per year) in EDA society and 2015 DAC Ph.D. forum best poster presentation. His researches are funded by NSF and Florida Center for Cybersecurity etc.