![Page 1: Weighted synergy graphs for effective team formation with heterogeneous ad hoc agents](https://reader035.vdocument.in/reader035/viewer/2022070422/568165bb550346895dd8b7c0/html5/thumbnails/1.jpg)
WEIGHTED SYNERGY GRAPHS FOR EFFECTIVE TEAM FORMATION WITHHETEROGENEOUS AD HOC AGENTS
Somchaya Liemhetcharat, Manuela Veloso
Presented by:Raymond Mead
![Page 2: Weighted synergy graphs for effective team formation with heterogeneous ad hoc agents](https://reader035.vdocument.in/reader035/viewer/2022070422/568165bb550346895dd8b7c0/html5/thumbnails/2.jpg)
Problem• Written for RoboCup Rescue Simulator, where teams of
robots are used to solve tasks.• We want to choose the best team of robots to tackle a disaster.• Around 50 possible agents.
• How can we form the best team when everyone’s abilities, and how well people work together, are known?
• Given observations of groups and their performances, how can we generate a graph to model each person’s ability, and how well people work together?
![Page 3: Weighted synergy graphs for effective team formation with heterogeneous ad hoc agents](https://reader035.vdocument.in/reader035/viewer/2022070422/568165bb550346895dd8b7c0/html5/thumbnails/3.jpg)
Modeling Teams• For forming teams, we want to look at:
• The compatibility between members of the team.• Each person’s ability.
• Using a weighted graph:• Each vertex represents a person, who has a certain ability• Edges are used to show similarity between people
• A person’s ability is modeled as a normal distribution • For someone, , their ability is
![Page 4: Weighted synergy graphs for effective team formation with heterogeneous ad hoc agents](https://reader035.vdocument.in/reader035/viewer/2022070422/568165bb550346895dd8b7c0/html5/thumbnails/4.jpg)
Example Graph
![Page 5: Weighted synergy graphs for effective team formation with heterogeneous ad hoc agents](https://reader035.vdocument.in/reader035/viewer/2022070422/568165bb550346895dd8b7c0/html5/thumbnails/5.jpg)
Compatibility
• is the minimum distance between
• , is a compatibility function.• Models how well people work together.
• Larger distance → Less compatible• • , exponential decay
![Page 6: Weighted synergy graphs for effective team formation with heterogeneous ad hoc agents](https://reader035.vdocument.in/reader035/viewer/2022070422/568165bb550346895dd8b7c0/html5/thumbnails/6.jpg)
Synergy of a Pair• A pair of people: • For a pair’s Synergy, add their abilities, , and scale it by
how compatible they are, .
• Normal distribution ~
![Page 7: Weighted synergy graphs for effective team formation with heterogeneous ad hoc agents](https://reader035.vdocument.in/reader035/viewer/2022070422/568165bb550346895dd8b7c0/html5/thumbnails/7.jpg)
Synergy of a Team• Average the Synergy between all pairs in a team •
• Normal Distribution ~
![Page 8: Weighted synergy graphs for effective team formation with heterogeneous ad hoc agents](https://reader035.vdocument.in/reader035/viewer/2022070422/568165bb550346895dd8b7c0/html5/thumbnails/8.jpg)
Example Synergies
![Page 9: Weighted synergy graphs for effective team formation with heterogeneous ad hoc agents](https://reader035.vdocument.in/reader035/viewer/2022070422/568165bb550346895dd8b7c0/html5/thumbnails/9.jpg)
Evaluating a Team• -value of a team is s.t. .
• Probability of a team’s performance being is .• If , then
• high risk, high reward• low risk, low reward
• is better than if
• -optimal team: • Has largest
![Page 10: Weighted synergy graphs for effective team formation with heterogeneous ad hoc agents](https://reader035.vdocument.in/reader035/viewer/2022070422/568165bb550346895dd8b7c0/html5/thumbnails/10.jpg)
Problem: Finding the -Optimal Team• Among all possible teams, find the best team for given .
• Need to check all possible sizes of teams• Need to check most, if not all teams for each team size.
• NP-Hard• Reduce the Max-Clique problem to Finding the Optimal Team.• Max-Clique: Find the largest subgraph, where there is an edge
between every pair of vertices.• NP-Complete
![Page 11: Weighted synergy graphs for effective team formation with heterogeneous ad hoc agents](https://reader035.vdocument.in/reader035/viewer/2022070422/568165bb550346895dd8b7c0/html5/thumbnails/11.jpg)
Algorithm: -optimal team of size • Branch and Bound Algorithm:
• is a team used for exploring possible teams.• Bound performance of to decide to keep exploring or not.• is the current known best team, with .• Initially, , and .
• Check all pairs, unless a new best is not possible with the current members.
• if the best is known• otherwise
![Page 12: Weighted synergy graphs for effective team formation with heterogeneous ad hoc agents](https://reader035.vdocument.in/reader035/viewer/2022070422/568165bb550346895dd8b7c0/html5/thumbnails/12.jpg)
Algorithm: -optimal team of size
If , compare and Return if is better, otherwise.
For , where
• All nodes that can be added are assumed to be worst or best case• Min compatibility with min ability → worst• Max compatibility with max ability → best
![Page 13: Weighted synergy graphs for effective team formation with heterogeneous ad hoc agents](https://reader035.vdocument.in/reader035/viewer/2022070422/568165bb550346895dd8b7c0/html5/thumbnails/13.jpg)
Reducing the Max-Clique Problem• , is unweighted - want to find the max-clique.
• The max-clique in will be the largest optimal team.
• Create to run with • Each edge in corresponds to an edge of weight 1 in • Everyone’s ability is • , Evaluating a team only depends on mean, always 1.•
![Page 14: Weighted synergy graphs for effective team formation with heterogeneous ad hoc agents](https://reader035.vdocument.in/reader035/viewer/2022070422/568165bb550346895dd8b7c0/html5/thumbnails/14.jpg)
Max-Clique → Best Team• Evaluating :, definition, only mean matters
• only when there is an edge between a pair in • otherwise
• Maximized when there is an edge between every pair of
![Page 15: Weighted synergy graphs for effective team formation with heterogeneous ad hoc agents](https://reader035.vdocument.in/reader035/viewer/2022070422/568165bb550346895dd8b7c0/html5/thumbnails/15.jpg)
Approximation Algorithm• Simulated Annealing
• Looking at teams similar to the current best, and comparing them
• Generate a random team• Repeat constant times:
• Find a new team similar to the current best, swap a node in • Evaluate both teams
• Replace if the new team is better
• Return the best team found
• Runs in if is known.• Evaluating is , where
• if n is unknown
![Page 16: Weighted synergy graphs for effective team formation with heterogeneous ad hoc agents](https://reader035.vdocument.in/reader035/viewer/2022070422/568165bb550346895dd8b7c0/html5/thumbnails/16.jpg)
Approximation Algorithm
Repeat times
Compare and Replace if is better
Return
![Page 17: Weighted synergy graphs for effective team formation with heterogeneous ad hoc agents](https://reader035.vdocument.in/reader035/viewer/2022070422/568165bb550346895dd8b7c0/html5/thumbnails/17.jpg)
Comparison• Effectiveness of team is
• Where ’s performance fits between best and worst.
![Page 18: Weighted synergy graphs for effective team formation with heterogeneous ad hoc agents](https://reader035.vdocument.in/reader035/viewer/2022070422/568165bb550346895dd8b7c0/html5/thumbnails/18.jpg)
Learning the Synergy Graph
• We have observations, , containing all people, .• Each observation is , team , performance, .
• Find a synergy graph that best fits the observations.• Need to find ability of each person.• Need to find the compatibility between people.
• Strategy: Simulated Annealing
![Page 19: Weighted synergy graphs for effective team formation with heterogeneous ad hoc agents](https://reader035.vdocument.in/reader035/viewer/2022070422/568165bb550346895dd8b7c0/html5/thumbnails/19.jpg)
Learning Algorithm:
Repeat constant times:
Compare scores of , and if is better
Return
![Page 20: Weighted synergy graphs for effective team formation with heterogeneous ad hoc agents](https://reader035.vdocument.in/reader035/viewer/2022070422/568165bb550346895dd8b7c0/html5/thumbnails/20.jpg)
Generating G and Finding Similar G’
• Vertices represent each person• Randomly put edges of random weights between vertices
• Do one of the following to :• Increase a random edge’s weight by 1• Decrease a random edge’s weight by 1• Remove a random edge• Add a random edge of random weight
![Page 21: Weighted synergy graphs for effective team formation with heterogeneous ad hoc agents](https://reader035.vdocument.in/reader035/viewer/2022070422/568165bb550346895dd8b7c0/html5/thumbnails/21.jpg)
Similar Graph:
![Page 22: Weighted synergy graphs for effective team formation with heterogeneous ad hoc agents](https://reader035.vdocument.in/reader035/viewer/2022070422/568165bb550346895dd8b7c0/html5/thumbnails/22.jpg)
Fitting Abilities to a Graph• Look at all teams of size 2 or 3 of , .
• Each , there are observations of , each with a performance.• Fit a normal distribution to the observed performance of .
• , is the observed distribution of • is the set of all
• We want the distribution of to match the distribution of .• Fit to as best we can choosing for each person
![Page 23: Weighted synergy graphs for effective team formation with heterogeneous ad hoc agents](https://reader035.vdocument.in/reader035/viewer/2022070422/568165bb550346895dd8b7c0/html5/thumbnails/23.jpg)
Fitting Abilities• For with of size 2:
• Similar for of size 3.
• Know , from the graph, and we want to fit to.• , matrix of , one row per team,
• Fit , for • matrix of , one row per team,
• Fit for
![Page 24: Weighted synergy graphs for effective team formation with heterogeneous ad hoc agents](https://reader035.vdocument.in/reader035/viewer/2022070422/568165bb550346895dd8b7c0/html5/thumbnails/24.jpg)
Code:
![Page 25: Weighted synergy graphs for effective team formation with heterogeneous ad hoc agents](https://reader035.vdocument.in/reader035/viewer/2022070422/568165bb550346895dd8b7c0/html5/thumbnails/25.jpg)
Log-Likelihood• Sum of log-likelihoods for each observation, given
synergy graph, and abilities.
• For an observation :
• Probability density of normal distribution at value .
![Page 26: Weighted synergy graphs for effective team formation with heterogeneous ad hoc agents](https://reader035.vdocument.in/reader035/viewer/2022070422/568165bb550346895dd8b7c0/html5/thumbnails/26.jpg)
Code
![Page 27: Weighted synergy graphs for effective team formation with heterogeneous ad hoc agents](https://reader035.vdocument.in/reader035/viewer/2022070422/568165bb550346895dd8b7c0/html5/thumbnails/27.jpg)
Evaluation
• Generate a hidden graph, with compatibility and abilities.• Generate a set of observations
• Run the learning Algorithm• Compare Log-Likelihood of learned graph with true graph.
![Page 28: Weighted synergy graphs for effective team formation with heterogeneous ad hoc agents](https://reader035.vdocument.in/reader035/viewer/2022070422/568165bb550346895dd8b7c0/html5/thumbnails/28.jpg)
Results
![Page 29: Weighted synergy graphs for effective team formation with heterogeneous ad hoc agents](https://reader035.vdocument.in/reader035/viewer/2022070422/568165bb550346895dd8b7c0/html5/thumbnails/29.jpg)
Results
![Page 30: Weighted synergy graphs for effective team formation with heterogeneous ad hoc agents](https://reader035.vdocument.in/reader035/viewer/2022070422/568165bb550346895dd8b7c0/html5/thumbnails/30.jpg)
Using for RoboCup
![Page 31: Weighted synergy graphs for effective team formation with heterogeneous ad hoc agents](https://reader035.vdocument.in/reader035/viewer/2022070422/568165bb550346895dd8b7c0/html5/thumbnails/31.jpg)
Thoughts:• Domain specific:
• Works well for the given problem, but may not be good for other applications.
• Tested for relatively small graphs.• May not be generalizable to large sparse graphs.
• Due to randomness of search.
• Modifying for learning large graphs:• Generate a better initial graph.• Make better choice for a similar graph.• More localized evaluation.