BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:IC Colloquium: Efficient Machine Learning via Data Summarization
DTSTART:20200427T150000
DTEND:20200427T160000
DTSTAMP:20260406T232510Z
UID:4e04d364e198c88b3082419ad29ee422b00009dd161e3c5a91b3c6f0
CATEGORIES:Conferences - Seminars
DESCRIPTION:This talk will take place via zoom. Please click on the follow
 ing link: https://epfl.zoom.us/j/637517307\n\nBy: Baharan Mirzasoleiman - 
 Stanford University\nIC Faculty candidate\n\nAbstract\nLarge datasets have
  been crucial to the success of modern machine learning models. However\, 
 training on massive data has two major limitations. First\, it is continge
 nt on exceptionally large and expensive computational resources\, and incu
 rs a substantial cost due to the significant energy consumption. Second\, 
 in many real-world applications such as medical diagnosis and self-driving
  cars\, big data contains highly imbalanced classes and noisy labels. In s
 uch cases\, training on the entire data does not result in a high-quality 
 model.\n \nIn this talk\, I will argue that we can address the above limi
 tations by developing techniques that can identify and extract the represe
 ntative subsets from massive datasets. Training on representative subsets 
 not only reduces the substantial costs of learning from big data\, but als
 o improves their accuracy and robustness against noisy labels. I will pres
 ent two key aspects to achieve this goal: (1) extracting the representati
 ve data points by summarizing massive datasets\; and (2) developing effici
 ent optimization methods to learn from the extracted summaries. I will dis
 cuss how we can develop theoretically rigorous techniques that provide str
 ong guarantees for the quality of the extracted summaries\, and the learne
 d models’ quality and robustness against noisy labels. I will also show
  the applications of these techniques to several problems\, including sum
 marizing massive image collections\, online video summarization\, and spee
 ding up training machine learning models.\n\nBio\nBaharan Mirzasoleiman is
  a Postdoctoral Research Scholar in Computer Science Department at Stanfor
 d University\, where she works with Prof. Jure Leskovec. Baharan’s resea
 rch focuses on developing new methods that enable efficient exploration an
 d learning from massive datasets. She received her PhD from ETH Zurich\, w
 orking with Prof. Andreas Krause. She has also spent two summers as an int
 ern at Google Research. She was awarded an ETH medal for Outstanding Docto
 ral Dissertation\, and a Google Anita Borg Memorial Scholarship. She was a
 lso selected as a Rising Star in EECS from MIT.\n\nMore information
LOCATION:
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR