主講人：鄭銀強

主講人簡介：鄭銀強教授於2013年在日本東京東京工業大學機械與控製工程系獲得工程博士學位。他目前是日本東京大學下一代人工智能研究中心的正教授，領導光學傳感和相機系統實驗室（OSCARS實驗室）。他發表了一系列關於光學成像和機器學習的研究論文。在與佳能和日立的合作中，他為多波段光聲成像系統和顯微熒光成像系統的開發和商業化做出了重大貢獻。他曾擔任 CVPR、ICCV⚱️、MM🛬、3DV📩、ACCV、ISAIR、DICTA 和 MVA 等CCF A類會議的區域主席。他曾獲得久負盛名的船井學術獎和柯尼卡美能達成像科學激勵獎。

講座摘要: AI algorithms for computer-based visual understanding have advanced significantly, due to the prevalence of deep learning and large-scale visual datasets in the RGB domain, which have also been proven vulnerable to digital and physical adversarial attacks. To deal with complex scenarios, many other imaging modalities beyond the visibility scope of human eyes, such as near infrared (NIR), thermal infrared (TIR), polarization, have been introduced, yet the vulnerabilities of visual AI based on these non-RGB modalities have not received due attention. In this talk, we will show that typical AI algorithms, like object detection and segmentation, can be more fragile than in the RGB domain. We showcase two physical attackers onto the YOLO-based human detector in the NIR and TIR domain, and one projection-based attacker onto the glass segmentation algorithm in the polarization-color domain, all of which are sufficiently concealing to human eyes.

 時間⏱：2024年5月24日下午15:00-18:00

地點：1號沐鸣樓204