Space: The final frontier Crisis space/resource allocation

1
Space The final frontier Crisis
space/resource allocation
December 18, 2003

• LTI/CSD Jaime Carbonell, Scott Fahlman, Eugene
  FinkLTI Bob Frederking, Greg Jorstad, Ulas
  Bardak,Thuc Vu, Richard WangImplicitly Yiming
  Yang, William Cohen, Lori Levin, Steve Smith,

2
People
3
Purpose

• Automated allocation of office space and
• related resources to office users, in both
• crisis and routine situations.

4
Urgent space allocation
Wean Hall
Toxic Cloud!
5
Main steps

• Elicitation of user preferences

• Near-optimal allocation based onpartial
  knowledge of preferences

• Negotiations with users

• Mediation of trades among users

6
Demo
7
Main challenges

• How to represent and reason about space

• How to optimize space allocation conditioned on
  resources, constraints, preferences, and forecasts

• How to cope with surprise, such as crises,
  degraded space, new constraints, new preferences,
  new utility functions, and new optimization
  criteria

• How to cope with uncertainty, such as partial
  knowledge of preferences, contingency planning
  based on possible exogenous events, and
  prediction of negotiation outcomes

• How to learn what works and why

8
Knowledge representation

• Need to represent
• Facts People, departments, equipment, affinity
  groups,
• Spatial relations 2D and 3D maps, connectivity,
  functions,
• Constraints Minimal space/person, labs
  w/plumbing and power,
• Preferences Proximity relations, windows,
  equipment,
• Utilities Cost and benefit of satisfying
  preferences, elasticities,
• Episodes Past space planning with utilities and
  outcomes, including justifications for decisions,
  retractions,
• Optimization criteria As first-class objects, so
  as to reason about what to optimize and how to
  assign weights and priorities

• Need to reflect on
• What does not the system know that it needs to
  know To trigger active learning, user
  interactions,
• What-if scenarios For multi-user negotiation, to
  assess the completeness of the systems
  knowledge,

9
Pervasive learning

• Learning at the factual level
• By being told Constraints, preferences, facts,
• By negotiation and examples Preference weights,
  utilities,

• Learning at the planning level
• Mode selection When to plan, when to seek
  information, when to negotiate, when to optimize,
  when to validate,
• Operator selection How to select best actions
  within modes
• Historical learning What to reuse/transfer
  longitudinally

• Learning at the meta level
• Self assessment Utility of the learning (e.g.
  idiosyncratic versus general), accuracy of the
  learning, permanence,
• Targeting the learner On maximizing expected
  future discounted utility, on correcting flaws
  (unlearning),

10
First fifteen months

• Non-crisis space allocation
• Add users to an existing occupied building
• Allocate offices in a new building

• Respecting constraints and preferences
• Unary Size, windows, internet, bio-isolation,
• N-ary Proximity to co-workers, quiet,

• Optimizing global utility
• Maximize preference satisfaction
• Minimize moving users already in place

• Coping with uncertainty
• Use ranges, defaults, what-if planning
• Elicit preferences and trade-offs
• Support single-user and multi-user negotiations

11
Architecture
12
Main modules

• Natural-language e-mail communications

• Representation of user preferences, which
  includes defaults and learned knowledge

• Representation of (uncertain) knowledgeof
  available space and related resources

• Optimization based on available knowledge

• Intelligent elicitation of user preferencesand
  information about available space

• Single-user and multi-user negotiations

• Bartering office space among users

• Speed-up and quality learning

13
Initial results

• Understanding of space-related e-mail

• Limited representation of space and related
  resources without uncertainty

• Optimization based on simple preferences

14
Initial results E-mail understanding
Extraction Rules
Extraction System
Space E-mail
Extracted Text
Template Generator
Space Template
15
E-mail example
Johnson wants to move to wean, he prefers the
room 5102. He wants that room for conducting his
experiments. His room will be filled with
chemical bottles and equipment. He would like to
be on the 5th floor, or higher than the 6th
floor, but definitely not lower than the fourth
floor please. He prefers the size of his room to
be between 1025 square meters. He will be moving
into the room starting 2/28/2004 until May 24,
2004. He likes to have at least a window in his
room, if possible. His room should have at least
2 doors. His room does not need internet, but
definitely need electricity and 510 sinks. He
would like to be above the Wean Engineering
Library, and below his advisor's office. He would
also like to be around 50 to 100 yards away from
the building's entrance.
16
Extraction rules

• Noun phrase identifier
• A rule-based noun phrase chunker utilizing the
  part-of-speech tags from the tagger

17
Extraction rules

• Negative scope identifier
• Determines what part of the sentence has
  negated meaning
• but not too far away from his classmates

• Quantity identifier
• Identifies quantities along with logical
  attributes
• must be a hundred fifty five square feet

18
Extracted text
requester Johnson filler chemical bottles and
equipment purpose conducting his
experiments building wean room
5102 date_start 2/28/2004 date_end May 24,
2004 floor_min the 6th floor the fourth
floor floor the 5th floor size_range 1025
square meters window_min a window entrance_min
2 doors entrance the building's
entrance plumbing_range 510 internet_neg
internet electric electricity rel_above the
Wean Engineering Library rel_beneath his
advisor's office dist3_range 50 to 100
yards dist3_from the building's entrance
19
Space template after normalization
Request_Allocation requester
(Johnson) filler (chemical bottles
and equipment) purpose (conducting his
experiments) building (WEH)
room (5102) floor (5gt6gt4)
size (gt108lt269) start_date (Sat Feb 28
2004) end_date (Mon May 24 2004)
window () entrance (gt2)
internet (-) plumbing (gt5lt10)
electric () above (the Wean
Engineering Library) beneath (his
advisor's office) distance
((gt150lt300)(the building's entrance))
20
Initial results Representation

• List of available offices

• Database of basic office properties(size,
  windows, internet connections,)

• On-demand computation of other properties(distanc
  e between offices, accessibility,)

21
Initial results Conversion from AutoCad
DB
RADAR/SpaceRepresentation
AutoCad
AutoCad maps include line drawings
andfree-floating text for office numbers.

• Identify the position of each office
• Compute the office areas
• Identify the office numbers
• Find the shortest paths between offices

22
Initial results Optimization
Application of simulated annealing.

• State Assignment of users to offices a user may
  be in a specific office or have no office

• Objective function Weighted count of unsatisfied
  user preferences

• Transitions
• Assign a user to an office
• Remove a user from an office
• Exchange locations of two users

23
Future tasks

• Representation of relevant knowledge
• Uncertain knowledge of user preferences
• Uncertain knowledge of available space
• Allocation of space and related resources
• Possible communications with users
• Use of Scone for knowledge representation and
  inference of implicit information

24
Future tasks

• Representation of relevant knowledge
• Understanding of space-related e-mail
• Use of Scone knowledge base
• Handling multiple requests in one e-mail
• Handling user responses to earlier e-mails
• Identifying unclear places in e-mails and
  asking users for clarification

25
Future tasks

• Representation of relevant knowledge
• Understanding of space-related e-mail
• Generation of e-mail replies
• User-friendly explanations
• Politeness and diplomacy

26
Future tasks

• Representation of relevant knowledge
• Understanding of space-related e-mail
• Generation of e-mail replies
• Optimization based on partial knowledge
• Find an allocation with a (near-)largest
  expected value of the objective function
• Estimate the standard deviation of the
  resulting expected value
• Determine what additional information may
  reduce the standard deviation

27
Future tasks

• Representation of relevant knowledge
• Understanding of space-related e-mail
• Generation of e-mail replies
• Optimization based on partial knowledge
• Preference elicitation and negotiation
• Select questions that reduce uncertainty
• Estimate probabilities of possible replies
• Reduce the number of e-mails to users

28
Future tasks

• Representation of relevant knowledge
• Understanding of space-related e-mail
• Generation of e-mail replies
• Optimization based on partial knowledge
• Elicitation of preferences and negotiation
• Fairness of space allocation, with respect to
• Novice users
• Helpful users
• Busy users

29
Future tasks

• Representation of relevant knowledge
• Understanding of space-related e-mail
• Generation of e-mail replies
• Optimization based on partial knowledge
• Elicitation of preferences and negotiation
• Fairness of space allocation
• Soft commitments and cancellations
• Support different levels of commitment
• If breaking a commitment to a user, negotiate
  appropriate compensation

30
Future tasks

• Representation of relevant knowledge
• Understanding of space-related e-mail
• Generation of e-mail replies
• Optimization based on partial knowledge
• Elicitation of preferences and negotiation
• Fairness of space allocation
• Soft commitments and cancellations
• Bartering among users
• Allow users to offer office-space trades
• Identify prospective multi-user trades

31
Future tasks

• Representation of relevant knowledge
• Understanding of space-related e-mail
• Generation of e-mail replies
• Optimization based on partial knowledge
• Elicitation of preferences and negotiation
• Fairness of space allocation
• Soft commitments and cancellations
• Bartering among users
• Interaction with human administrators
• Providing relevant information
• Asking help with complex decisions
• Supporting multiple administrators

32
Future tasks

• Representation of relevant knowledge
• Understanding of space-related e-mail
• Generation of e-mail replies
• Optimization based on partial knowledge
• Elicitation of preferences and negotiation
• Fairness of space allocation
• Soft commitments and cancellations
• Bartering among users
• Interaction with human administrators
• Learning new knowledge and strategies
• User preferences
• Negotiation strategies
• E-mail understanding

33
Future tasks

• Representation of relevant knowledge
• Understanding of space-related e-mail
• Generation of e-mail replies
• Optimization based on partial knowledge
• Elicitation of preferences and negotiation
• Fairness of space allocation
• Soft commitments and cancellations
• Bartering among users
• Interaction with human administrators
• Learning new knowledge and strategies
• Graphical user interface
• Visualization
• Spatial input

34
Schedule of initial versions

• Knowledge representation (March 2004)
• Understanding e-mail (March 2004)
• Generation of e-mail replies (March 2004)
• Partial-knowledge optimization (May 2004)
• Elicitation of preferences (May 2004)
• Fairness of space allocation (December 2004)
• Soft commitments and cancellations (2005)
• Bartering among users (2005)
• Interaction with human administrators (2005)
• Learning new knowledge (long term)
• Graphical user interface (long term)

35
Interaction with other systems

• Scheduling
• Determine the availability of specific users
• Anticipate the needs of users and groups

• E-mail
• Identify and prioritize space-related e-mails
• Estimate response times of specific users

• Webmaster
• Provide on-line information about space

• User studies
• Evaluate user satisfaction
• Improve interaction with users

36
Evaluation

• Allocation quality

• Speed and scalability

37
To be continued
38
Allocation quality

• Maximizing quality of space allocation ai
• Subject to time constraints, available extrinsic
  data, available task information, and
  communication constraints
• Comparison with the results of omniscient
  unconstrained optimization aopt