Learning to Optimize Plan Execution in Information Agents Craig A. - PowerPoint PPT Presentation

Learning to Optimize Plan Execution in Information Agents Craig A. Knoblock Knoblock Craig A. University of Southern California University of Southern California Craig Knoblock University of Southern California 1

Acknowledgements Acknowledgements � Theseus Theseus Agent Agent � Electric Elves Electric Elves � � Execution Execution • Jose Luis Jose Luis Ambite Ambite • • Greg • Greg Barish Barish • Maria Maria Muslea Muslea • • Steve Minton • Steve Minton • Hans Hans Chalupsky Chalupsky • • Maria • Maria Muslea Muslea • Yolanda Gil • Yolanda Gil � Speculative Execution Speculative Execution • Jean Oh Jean Oh • � • Greg • Greg Barish Barish • David V. David V. Pynadath Pynadath • • Thomas A. Russ Thomas A. Russ • � Funding Funding � • Milind • Milind Tambe Tambe • DARPA • DARPA • AFOSR • AFOSR • NSF • NSF Craig Knoblock Craig Knoblock University of Southern California 2 2 University of Southern California

Introduction Introduction � The Web is a tremendous resource, but The Web is a tremendous resource, but � designed for browsing designed for browsing • Sites provide limited capabilities for Sites provide limited capabilities for • personalization personalization • Few sites are designed to be integrated with Few sites are designed to be integrated with • others others � Goal: Develop technology to rapidly Goal: Develop technology to rapidly � construct personal software agents construct personal software agents • Build agents that can perform retrieval, Build agents that can perform retrieval, • integration, and monitoring tasks on any integration, and monitoring tasks on any online source online source Craig Knoblock Craig Knoblock University of Southern California 3 3 University of Southern California

Outline Outline Motivating Application: The Electric Elves Motivating Application: The Electric Elves 1. 1. Efficiently executing agent plans Efficiently executing agent plans 2. 2. Speculative plan execution Speculative plan execution 3. 3. Value prediction for speculative execution Value prediction for speculative execution 4. 4. Related Work Related Work 5. 5. Conclusions Conclusions 6. 6. Craig Knoblock Craig Knoblock University of Southern California 4 4 University of Southern California

Electric Elves Project Electric Elves Project [Chalupsky Chalupsky et al, 2001] et al, 2001] [ Elves project goal: Apply agent technology to Elves project goal: Apply agent technology to support human organizations support human organizations • Develop software agents that automate routine tasks • Enable software agents and humans to work together • Support coordination of tasks � Applications: Office Elves and Travel Elves Applications: Office Elves and Travel Elves � W W W A g e n t I n f o r m a t i o n O n t o l o g y - b a s e d P r o x i e s A g e n t s M a t c h m a k e r s F o r GRID P e o p l e Craig Knoblock Craig Knoblock University of Southern California 5 5 University of Southern California

Agents for Monitoring Travel Agents for Monitoring Travel [Ambite et al, 2002] [Ambite et al, 2002] • Travel Elves created as an application of the Travel Elves created as an application of the • Electric Elves Electric Elves • Given travel itinerary, generates set of agents for Given travel itinerary, generates set of agents for • anticipating travel- -related failures and related failures and anticipating travel opportunities: opportunities: • Price changes Price changes • • Schedule changes Schedule changes • • Flight delays & cancellations Flight delays & cancellations • • Earlier and close connections Earlier and close connections • • Finding the closest restaurant given GPS coordinates Finding the closest restaurant given GPS coordinates • Craig Knoblock Craig Knoblock University of Southern California 6 6 University of Southern California

Monitoring Travel Plans Monitoring Travel Plans Craig Knoblock Craig Knoblock University of Southern California 7 7 University of Southern California

Agents Deployed to Agents Deployed to Monitor Travel Itinerary Monitor Travel Itinerary Travel Itinerary W W W A g e n t I n f o r m a t i o n O n t o l o g y - b a s e d P r o x i e s A g e n t s M a t c h m a k e r s F o r GRID P e o p l e Flight Prices & Restaurants Schedules Flight Status Weather Craig Knoblock Craig Knoblock University of Southern California 8 8 University of Southern California

Monitoring Agents Monitoring Agents � Flight Flight- -Status Agent: Status Agent: � • Flight delayed message: Flight delayed message: • Your United Airlines flight 190 has been delayed. Your United Airlines flight 190 has been delayed. It was originally scheduled to depart at 11:45 AM It was originally scheduled to depart at 11:45 AM and is now scheduled to depart at 12:30 PM. and is now scheduled to depart at 12:30 PM. The new arrival time is 7:59 PM. The new arrival time is 7:59 PM. • Flight cancelled message: • Flight cancelled message: Your Delta Air Lines flight 200 has been cancelled. Your Delta Air Lines flight 200 has been cancelled. • Fax to hotel message: Fax to hotel message: • Attention: Registration Desk Attention: Registration Desk I am sending this message on behalf of David I am sending this message on behalf of David Pynadath, who has a reservation at your hotel. David , who has a reservation at your hotel. David Pynadath Pynadath is on United Airlines 190, which is now Pynadath is on United Airlines 190, which is now scheduled to arrive at IAD at 7:59 PM. Since the scheduled to arrive at IAD at 7:59 PM. Since the flight will be arriving late, I would like to request flight will be arriving late, I would like to request that you indicate this in the reservation so that the that you indicate this in the reservation so that the room is not given away. room is not given away. Craig Knoblock Craig Knoblock University of Southern California 9 9 University of Southern California

Monitoring Agents Monitoring Agents � Airfare Agent: Airfare dropped message Airfare Agent: Airfare dropped message � The airfare for your American Airlines itinerary The airfare for your American Airlines itinerary (IAD - - LAX) dropped to $281. LAX) dropped to $281. (IAD � Earlier Earlier- -Flight Agent: Earlier flights message Flight Agent: Earlier flights message � The status of your currently scheduled flight is: The status of your currently scheduled flight is: # 190 LAX (11:45 AM) - - IAD (7:29 PM) 45 minutes Late IAD (7:29 PM) 45 minutes Late # 190 LAX (11:45 AM) If you would like to return earlier, the following If you would like to return earlier, the following United Airlines flights will arrive earlier than your United Airlines flights will arrive earlier than your scheduled flights: scheduled flights: # 946 LAX (8:31 AM) - - IAD (3:35 PM) 11 minutes Late IAD (3:35 PM) 11 minutes Late # 946 LAX (8:31 AM) -------- -------- # 388 LAX (9:25 AM) - - DEN (12:25 PM) 10 minutes Late DEN (12:25 PM) 10 minutes Late # 388 LAX (9:25 AM) # 1534 DEN (1:20 PM) - - IAD (6:06 PM) On Time IAD (6:06 PM) On Time # 1534 DEN (1:20 PM) Craig Knoblock Craig Knoblock University of Southern California 10 10 University of Southern California

Outline Outline Motivating Application: The Electric Elves Motivating Application: The Electric Elves 1. 1. Efficiently executing agent plans Efficiently executing agent plans 2. 2. Speculative plan execution Speculative plan execution 3. 3. Value prediction for speculative execution Value prediction for speculative execution 4. 4. Related Work Related Work 5. 5. Conclusions Conclusions 6. 6. Craig Knoblock Craig Knoblock University of Southern California 11 11 University of Southern California

Efficiently Executing Efficiently Executing Agent Plans Agent Plans � Problem Problem � • Information gathering may involve accessing Information gathering may involve accessing • and integrating data from many sources and integrating data from many sources • Total time to execute these plans may be large Total time to execute these plans may be large • � Why? Why? � • Slow remote sources Slow remote sources • • Unpredictable network latencies Unpredictable network latencies • • Binding patterns • Binding patterns � Source cannot be queried until a previous query has Source cannot be queried until a previous query has � been answered been answered • Result: execution is often I/O Result: execution is often I/O- -bound bound • Craig Knoblock Craig Knoblock University of Southern California 12 12 University of Southern California