~/talk

Cross-dataset Time Series Anomaly Detection for Cloud Systems

on YouTube 21m
Talkat USENIX ATC '19 · on Friday, 2019-07-12· at 12:50· USENIX ATC Track II: Grand Ballroom VII–IX· [machine-learning]

speakers

name:Xu Zhang
links:none on file
name:Qingwei Lin
links:none on file
name:Yong Xu
links:none on file
name:Si Qin
links:none on file
name:Hongyu Zhang
links:none on file
name:Bo Qiao
links:none on file
name:Yingnong Dang
links:none on file
name:Xinsheng Yang
links:none on file
name:Qian Cheng
links:none on file
name:Murali Chintalapati
links:none on file
name:Youjiang Wu
links:none on file
name:Ken Hsieh
links:none on file
name:Kaixin Sui
links:none on file
name:Xin Meng
links:none on file
name:Yaohai Xu
links:none on file
name:Wenchi Zhang
links:none on file
name:Furao Shen
links:none on file
name:Dongmei Zhang
links:none on file

resources

video:https://youtu.be/lZKKmCkH9bc
slides:none on file

summary

ATAD trains anomaly detectors on labelled cloud datasets and applies them to new unlabelled services by combining transfer learning with active learning; features include statistical summaries, forecasting features (ARIMA/seasonality), and time-proximity deltas. A clustering-based mapping aligns source feature groups to target data before per-cluster transfer, then active queries pick diverse, high-uncertainty windows for human labels. Experiments on Yahoo/Twitter and Microsoft traces raise F1 significantly and need only 1%-5% labels to reach strong performance.

AI-written summary · about

chapters (13)

  1. 0:00Start
  2. 1:33Anomaly detection for cloud systems: a common problem
  3. 5:58Our Approach - ATAD
  4. 7:35Feature Identification
  5. 8:42Transfer between Source & Target Domain
  6. 10:08Making Transfer Learning More Effective
  7. 10:57Active Learning Component
  8. 11:49Identifying high uncertainty instances
  9. 12:29Maximizing the context diversity of candidates to be labeled
  10. 13:18Experiments
  11. 14:01Setup
  12. 15:41How effective is the Transfer Learning component?
  13. 16:47How effective is the Active Learning

related talks

~/search
type to search 6,249 conferences, 138,987 speakers and 185,172 talks
~/help
cmdctrl k
search every conference, speaker and series
/
focus the filter search
h
home
c
conferences
s
speakers
e
series
t
light / dark — the status bar button does the same
?
this help
esc
close this, then clear every filter — or just the search, while you are typing in it

Ignored while you are typing. Pages without a filter rail have no search to focus.