CSE 598 • 6. Social Insects & Collective Computation

• The decay parameter simulates natural evaporation of pheromone from the trail. • The decay parameter prevents pheromone from accumulating indefinitely on edges. Pheromones are molecules that evaporate over time, so this parameter captures that phenomenon. If the trail is too long compared to the value of the decay parameter, the pheromone may evaporate before other ants find the trail.

As seen in the application of ant colony optimization to the Traveling Salesman Problem, what is the purpose of the decay parameter in the pheromone update? • The decay parameter simulates the consumption of pheromone by ants. • The decay parameter allows the pheromone on the preferred trail to accumulate infinitely. • The decay parameter simulates natural evaporation of pheromone from the trail. • The decay parameter prevents pheromone from accumulating indefinitely on edges.

c) When a path breaks, ants prune the paths that lead to dead ends. Pruning refers to the removal over time of paths that lead to dead ends.

How do turtle ants use pruning to repair trails? a) Once a path is repaired, ants prune all other paths that are non-optimal. b) Once a path is repaired, ants prune the optimal path to make it as short as possible. c) When a path breaks, ants prune the paths that lead to dead ends. d) When a path breaks, ants prune the paths closest to the break to redirect search in the opposite direction.

d) The colony will fail to switch to the stronger feeder. In Dr. In Pratt's lecture he presents this finding.

If a stronger feeder is introduced after the majority of the ant colony (other than Tenmothorax Regulatus colonies) has committed to a weaker feeder, what will happen? a) The ants uncommitted when the new feeder is introduced will commit to the new feeder. b) The colony will successfully switch to the stronger feeder. c) The colony will divide evenly among the two feeders. d) The colony will fail to switch to the stronger feeder.

0.5 When feeders that are equal distance from the nest and are of equal quality, the ants will randomly select one feeder to feed from with 50/50 chance of selecting either feeder.

If an ant nest is equal distance from feeder A with 0.1M of sucrose and feeder B with 0.1M of sucrose what is the probability that the ants collectively commit to feed from feeder A over time?

a) True In ACO algorithms, ants can have local memory, known as its tabu list. In the Traveling Salesman Problem, we saw an example of this.

In ant colony optimization algorithms, each ant can have local memory. a) True b) False

a) Feeder A was completely consumed faster than feeder B. The final plot of this assignment showed the amount of food in the feeder over time. Feeder A was consumed faster than feeder B.

In the assignment "Modeling Collective Behavior in Ants", how did we show that feeder A was the better feeder? a) Feeder A was completely consumed faster than feeder B. b) Feeder B was completely consumed by faster than feeder A. c) More ants were recruited to feeder A than feeder B. d) All ants were committed to feeder A at the end of the experiment.

b) The ant moves to the cell nearest to the feeder it is committed to. If the ant is tandem running, it has perfect memory of the shortest path to the feeder it is committed to. It is then able to move in the direction of the feeder.

In the assignment "Modeling Collective Behavior in Ants," what happens when an ant is tandem running and not at a food location? a) The ant moves randomly. b) The ant moves to the cell nearest to the feeder it is committed to. c) The ant un-commits from its feeder. d) The ant checks if there is a potential recruit at its location.

a) This parameter allows edges with zero pheromone to be traversed. With probability equal to q_explore, an ant will explore an edge with zero pheromone.

In the network route repair algorithm, what is the purpose of the q_explore parameter? a) This parameter allows edges with zero pheromone to be traversed. b) This parameter is a rate coefficient that determines how strong an effect pheromone level has on the ants decision to traverse any given edge. c) This parameter determines how many edges are traversed in a single iteration. d) This parameter selects a random edge to traverse with probability equal to q_explore.

d) Junctions Junctions between separate paths are represented by nodes as explained in the Chandrasekhar et. al paper.

In the network route repair algorithm, what physical element of the trails was represented by nodes in the graph? a) Leaves b) Branches c) Twigs d) Junctions

• Twigs • Branches Branches and twigs were represented by edges as explained in the Chandrasekhar et. al paper.

In the repairing network routes algorithm, what physical element of the trails were represented by edges? • Twigs • Junctions • Branches • Leaves

• Heuristic information Both algorithms use heuristics information to select a path. The traveling salesman algorithm uses path length and the bayesian network uses a probabilistic metric. • Pheromone level Both algorithms use the pheromone level on the edge to select a path.

In the traveling salesman and bayesian networks examples for ant colony optimization, what do ants consider when adding an edge to their path? • Heuristic information • Random choice • Shortest path • Number of ants on the trail • Pheromone level

b) Feeder A has higher quality / better food than Feeder B. If Feeder A is of higher quality then it is more likely that tandem running ants will recruit others to the feeder.

Recall the "Simple model of forager behavior" as seen in Dr. Pratt's lecture to answer the following question. Under what conditions will the recruitment rate for Feeder A be greater than for Feeder B? a) Feeder A is further from the nest than feeder B. b) Feeder A has higher quality / better food than Feeder B. c) Feeder A has lower quality food than feeder B. d) Feeder A and feeder B always have equal recruitment rates by definition.

b) The discovery rate for feeder A will be equal to that of feeder B. The discovery rates for both feeders are equal by definition.

Recall the "Simple model of forager behavior" described in Dr. Pratt's lecture to answer the following question: If an ant nest is equal distance from feeder A with 0.1M of sucrose and feeder B with 0.8M sucrose, which of the following statements about the discovery rate for these feeders is true? a) The discovery rate of feeder A and feeder B change with respect to time. b) The discovery rate for feeder A will be equal to that of feeder B. c) The discovery rate for feeder A will be less than that of feeder B. d) The discovery rate for feeder A will be greater than that of feeder B.

a) Trail A Pheromone laying ants will lay more pheromone on a trail leading to a better food source. In this case that feeder is feeder A.

When considering a pheromone laying ant experiment, suppose that feeder A has 0.8M sucrose and feeder B has 0.1M sucrose. Let trail A be the shortest path to feeder A and trail B be the shortest path to feeder B. Which trail would most likely have a larger pheromone quantity after a food source is selected? a) Trail A b) Trail B c) Not enough information is given.

c) Optimization problems ACO algorithms were first introduced by Marco Dorigo to solve optimization problems such as, traveling salesman, quadratic assignment, and job-shop scheduling.

When the ant colony optimization algorithm was first introduced, what kinds of problems was it used to solve? a) Search problems b) Counting problems c) Optimization problems d) Decision problems

c) Linear In Dr. Pratt's lecture he describe this function as the "Function relating recruitment effectiveness to recruitment effort" and said that it is linear for tandem running.

Which of the following functions best describes the relationship between the number of recruiters for a feeder and probability of choosing that feeder for tandem running ants, for the majority of ant species? a) Hyperbolic b) Exponential c) Linear d) Sigmoidal

a) Bi-directional search Bi-directional search had an average success rate of 70% on the full grid, which was much higher than uni-directional search.

Which search method was found to be the most effective for repairing networks on a full grid? a) Bi-directional search b) Uni-directional search c) Multi-directional search d) Random-directional search

c) At low values of q_decay the original path is preserved for longer allowing for repair that starts close to the break. The ants performed better with local search so preserving the pheromone on the paths close to the break was necessary for their success.

Why was the network repair algorithm more successful for low values of q_decay? a) At low values of q_decay new paths are preserved for longer allowing for ants to effectively prune paths. b) At low values of q_decay new paths are preserved for longer allowing more ants to explore them. c) At low values of q_decay the original path is preserved for longer allowing for repair that starts close to the break. d) At low values of q_decay the original path is preserved for longer allowing for repair that starts further from to the break.