Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Alert Generation in Execution Monitoring Using ...

Hong Xu
May 21, 2018

Alert Generation in Execution Monitoring Using Resource Envelopes

The presentation slides of the paper "T. K. Satish Kumar, Hong Xu, Zheng Tang, Anoop Kumar, Craig Milo Rogers, and Craig A. Knoblock. Alert generation in execution monitoring using resource envelopes. In Proceedings of the 31st International Florida Artificial Intelligence Research Society Conference (FLAIRS), 38–43. 2018."

More details: http://www.hong.me/papers/kumar2018.html
Link to published paper: https://aaai.org/ocs/index.php/FLAIRS/FLAIRS18/paper/view/17615/16853

Hong Xu

May 21, 2018
Tweet

More Decks by Hong Xu

Other Decks in Research

Transcript

  1. Alert Generation in Execution Monitoring Using Resource Envelopes T. K.

    Satish Kumar Hong Xu Zheng Tang Anoop Kumar Craig Milo Rogers Craig A. Knoblock [email protected], [email protected], {zhengtan, anoopk, rogers, knoblock}@isi.edu May 21, 2018 Information Sciences Institute, University of Southern California the 31st International FLAIRS Conference (FLAIRS 2018) Melbourne, Florida, the United States of America
  2. Executive Summary Execution monitoring is an important aspect of AI.

    We apply the idea of resource envelope to alert generation in execution monitoring. We study three applications in details. Kumar et al. (Information Sciences Institute, USC) Alert Generation in Execution Monitoring Using Resource Envelopes 1 / 15
  3. Agenda Alert Generation in Execution Monitoring Simple Temporal Network with

    Resources (STNR) and Resource Envelope Applying Resource Envelopes for Alert Generation in Execution Monitoring Conclusion Kumar et al. (Information Sciences Institute, USC) Alert Generation in Execution Monitoring Using Resource Envelopes 2 / 15
  4. Agenda Alert Generation in Execution Monitoring Simple Temporal Network with

    Resources (STNR) and Resource Envelope Applying Resource Envelopes for Alert Generation in Execution Monitoring Conclusion
  5. Alert Generation in Execution Monitoring For a given execution plan,

    adaptation are often required in response to a changing environment. Alerts for such adaption are important. Ships need to adapt their routes in response to unexpected hurricane. (Image: (Skillet n.d.)) A team needs to adapt its plan in response to unexpected illness of its members. (Image: (U.S. Department of Defense Current Photos 2013)) Kumar et al. (Information Sciences Institute, USC) Alert Generation in Execution Monitoring Using Resource Envelopes 3 / 15
  6. Potential Future Resource Depletion as Alerts • One important type

    of alert relates to potential future resource depletion. • Examples include depletion of fuel in transportation applications, labor hours in human resource management applications, supplies in manufacturing applications, and so on. • We apply the idea of resource envelope to generate this type of alert under the framework of Simple Temporal Network with Resources (STNR), a temporal reasoning framework. Kumar et al. (Information Sciences Institute, USC) Alert Generation in Execution Monitoring Using Resource Envelopes 4 / 15
  7. Agenda Alert Generation in Execution Monitoring Simple Temporal Network with

    Resources (STNR) and Resource Envelope Applying Resource Envelopes for Alert Generation in Execution Monitoring Conclusion
  8. Simple Temporal Network (STN) • A graphical representation of a

    collection of simple temporal constraints between the execution times of various events. • First proposed in (Dechter et al. 1991). • Formally, it is defined on a directed graph G = X, E , where • X is a set of events, and • each edge eij = (Xi , Xj ) ∈ E represents a simple temporal constraint LB(eij ) ≤ Xj − Xi ≤ UB(eij ). Kumar et al. (Information Sciences Institute, USC) Alert Generation in Execution Monitoring Using Resource Envelopes 5 / 15
  9. Simple Temporal Network with Resources (STNR) Each event in an

    STN is also associated with a resource level. A resource level is positive/negative if this event produces/consumes resources. Example: X0 : 0 X1 : 50 X2 : −50 X3 : 250 X4 : −150 [5,10] [5,8] [10,10] [2,4] [5,10] Kumar et al. (Information Sciences Institute, USC) Alert Generation in Execution Monitoring Using Resource Envelopes 6 / 15
  10. Resource Envelope The largest and smallest possible total resource levels

    at each time instant (Kumar 2003). Can be computed using a maxflow procedure. X0 : 0 X1 : 50 X2 : −50 X3 : 250 X4 : −150 [5,10] [5,8] [10,10] [2,4] [5,10] 0 5 10 15 20 25 30 Time Instant 0 50 100 150 200 250 300 Resource Level Upper Resource Envelope Lower Resource Envelope Kumar et al. (Information Sciences Institute, USC) Alert Generation in Execution Monitoring Using Resource Envelopes 7 / 15
  11. Agenda Alert Generation in Execution Monitoring Simple Temporal Network with

    Resources (STNR) and Resource Envelope Applying Resource Envelopes for Alert Generation in Execution Monitoring Conclusion
  12. Resource Envelopes for Alert Generation Basic Principle: If the minimum

    resource level is too low at some time instants, an alert should be generated. Example: If resource level below 110 is dangerous at time instant 17, an alert will be generated. 0 5 10 15 20 25 30 Time Instant 0 50 100 150 200 250 300 Resource Level Upper Resource Envelope Lower Resource Envelope Kumar et al. (Information Sciences Institute, USC) Alert Generation in Execution Monitoring Using Resource Envelopes 8 / 15
  13. Smart Home X0 (0,0) Heater starts (-400,0) Heater ends (400,300)

    Washer starts (-500,-70) Washer ends (500, 0) Dryer starts (-600, 0) Dryer ends (600, 0) Refrigerator starts (-100,0) Refrigerator ends (100,0) Refrigerator starts (-100,0) Refrigerator ends (100,0) Rice Cooker starts (-100,-3) Rice Cooker ends (100,0) Dishwasher starts (-1200,-100) Dishwasher ends (1200,0) Xf (0,0) [0,+∞] [30,35] [5,30] [45,50] [0,+∞] [0,0] [0,+∞] [0,40] [0,+∞] [0,+∞] [20,30] [1020,1080] [0,+∞] [50,70] [0,+∞] [0,240] [0,0] [0,+∞ ] [40,45] [0,+∞ ] [0,+∞ ] [0,+∞] 0 200 400 600 800 1000 1200 1400 Time Instant (elapsed time since 12:00am in minutes) 2000 1750 1500 1250 1000 750 500 250 0 Resource Level (electrical power in watts) Upper Resource Envelope Lower Resource Envelope Electrical Power Limit Kumar et al. (Information Sciences Institute, USC) Alert Generation in Execution Monitoring Using Resource Envelopes 9 / 15
  14. Smart Home X0 (0,0) Heater starts (-400,0) Heater ends (400,300)

    Washer starts (-500,-70) Washer ends (500, 0) Dryer starts (-600, 0) Dryer ends (600, 0) Refrigerator starts (-100,0) Refrigerator ends (100,0) Refrigerator starts (-100,0) Refrigerator ends (100,0) Rice Cooker starts (-100,-3) Rice Cooker ends (100,0) Dishwasher starts (-1200,-100) Dishwasher ends (1200,0) Xf (0,0) [0,+∞] [30,35] [5,30] [45,50] [0,+∞] [0,0] [0,+∞] [0,40] [0,+∞] [0,+∞] [20,30] [1020,1080] [0,+∞] [50,70] [0,+∞] [0,240] [0,0] [0,+∞ ] [40,45] [0,+∞ ] [0,+∞ ] [0,+∞] [0,+∞] 0 200 400 600 800 1000 1200 1400 Time Instant (elapsed time since 12:00am in minutes) 2000 1750 1500 1250 1000 750 500 250 0 Resource Level (electrical power in watts) Upper Resource Envelope Lower Resource Envelope Electrical Power Limit Kumar et al. (Information Sciences Institute, USC) Alert Generation in Execution Monitoring Using Resource Envelopes 10 / 15
  15. Food Delivery r1 c1 c2 c3 c6 c5 c4 r2

    0 2 4 6 8 10 12 14 16 Time Instant (elapsed time since being en route) 0 1 2 3 4 5 6 Resource Level (amount of menu items in car) Upper Resource Envelope Lower Resource Envelope Kumar et al. (Information Sciences Institute, USC) Alert Generation in Execution Monitoring Using Resource Envelopes 11 / 15
  16. Food Delivery r1 c1 c2 c3 c6 c5 c4 r2

    0 2 4 6 8 10 12 14 16 Time Instant (elapsed time since being en route) 0 1 2 3 4 5 6 7 Resource Level (amount of menu items in car) Upper Resource Envelope Lower Resource Envelope Kumar et al. (Information Sciences Institute, USC) Alert Generation in Execution Monitoring Using Resource Envelopes 12 / 15
  17. Service Management (System Boot) • During system boot, services are

    being started. • Traditionally, a service cannot start before all its dependencies have started. • A modern system service management system, such as systemd, can start a service and its dependencies at the same time. • For example, the web service depends on the database management system (DBMS). The web service • begins starting up without the DBMS (A1 begins), • then gets interrupted when it needs to connect to the DBMS (A1 ends), • continues starting up once the DBMS is ready (A2 begins), and • finishes starting up (A2 ends). Kumar et al. (Information Sciences Institute, USC) Alert Generation in Execution Monitoring Using Resource Envelopes 13 / 15
  18. Service Management (System Boot) X0 : 2 A1 begins :

    -1 A1 ends : 1 A2 begins : -1 A2 ends : 1 B1 begins : -1 B1 ends : 1 B2 begins : -1 B2 ends : 1 C begins : -1 C ends : 1 Xf [0,+∞] [0,+∞] [0,+∞] [15,15] [0,10] [20,20] [0,10] [40,40] [0,+∞] [0,+∞] [10,10] [15,15] [0,+∞] [0,+∞] Resource: CPU cores 0 10 20 30 40 50 60 Time Instant (elapsed time during system boot) 1.0 0.5 0.0 0.5 1.0 1.5 2.0 Resource Level (number of CPU cores) Upper Resource Envelope Lower Resource Envelope Kumar et al. (Information Sciences Institute, USC) Alert Generation in Execution Monitoring Using Resource Envelopes 14 / 15
  19. Agenda Alert Generation in Execution Monitoring Simple Temporal Network with

    Resources (STNR) and Resource Envelope Applying Resource Envelopes for Alert Generation in Execution Monitoring Conclusion
  20. Conclusion • Generating alerts for resource depletion in execution monitoring

    is important. • We apply the idea of resource envelope to generate this type of alert under the framework of Simple Temporal Network with Resources (STNR). • We demonstrate on three application domains: smart home, food delivery, and service management. • (Future work) Make use of more features in resource envelopes. Kumar et al. (Information Sciences Institute, USC) Alert Generation in Execution Monitoring Using Resource Envelopes 15 / 15
  21. References I R. Dechter, I. Meiri, and J. Pearl. “Temporal

    constraint networks”. In: Artificial Intelligence 49.1–3 (1991), pp. 61–95. doi: http://dx.doi.org/10.1016/0004-3702(91)90006-6. T. K. S. Kumar. “Incremental Computation of Resource-Envelopes in Producer-Consumer Models”. In: the International Conference on Principles and Practice of Constraint Programming. 2003, pp. 664–678. S. D. Skillet. Clipper Ship HURRICANE in the English Channel. url: https://commons.wikimedia.org/wiki/File:S._D._Skillet_- _Clipper_Ship_HURRICANE_in_the_English_Channel.jpg. U.S. Department of Defense Current Photos. Jan. 2013. url: https://commons.wikimedia.org/wiki/File:130105-A-GX635-442_(8364697465).jpg.