The Traveling salesrat: Insights into optimal spatial...
Transcript of The Traveling salesrat: Insights into optimal spatial...
Introduction
Methods
Acknowledgments
References
Conclusions
0.2
0.4
0.6
0.8
1
1 2 3 40
Last Trial
Configuration Type
First Trial
1 2 3 40
0.2
0.4
0.6
0.8
1
Configuration Type
Exploration Trial First Trial Last Trial
Optimality Ratio = Path taken/min path length
Theoretical Distribution of Path Lengths
Experimental Distribution of Path Lengths
32.521.510.50
20
40
60
80
32.521.510.50
100
200
300
Trials
Op
tima
lity
Ra
tio
Performance across trials
2.0
1.8
1.6
1.4
1.2
1.0
0.8
42 6 8 10
Days
Op
tima
lity
Ra
tio
Performance across days
1 2 3 4 5 6
T2T1
T3 T4
T5
T7 T8
T9
T6
T10Traveling Salesman Problem
Optimization on the basis of spatial and reward information
Last Trial
Last TrialFirst Trial
First Trial
10 trials to find the optimal path for a given configuration
A total of 24 configurations over a six day period were used.
Re-optimization due to reward contingencies modification of
First Trial Last Trial
Reward removed from city after10 trials.
Trials continued until rat was able to re-optimize its path.
Spatially optimal, reward optimal, or ambiguous configurations were presented. Each configurations: Exploration trial+ 3 decision trials.
? The traveling salesman problem is a classic problem in artificial intelligence and theoretical computer science in which an agent has to plan visits to a fixed set of cities. It can be solved by calculating the total distance traveled for every possible tour and sorting the solutions. Finding the best solution is computationally expensive (NP-complete problem) because each city added increases the complexity of the problem exponentially.
? Heuristic methods allow humans to find near optimal solutions.
? Understanding the neural mechanisms underlying these heuristic processes can give insights into how complex choices are made.
? We propose a rodent model to investigate problem solving strategies at both behavioral and neural levels.
? We study how rats use a combination of spatial and reward information to optimize their decisions.
The authors would like to profusely thank Brian Gereke for assistance with experimental design, conduction of experiments and data analysis , Peter Hirshman and Nadia Corral-Frías for their help running experiments.
Traveling Salesman Problem
Optimization on the basis of spatial and reward information
Re-Optimization because of reward removal
T1 T4 T7 T10
Reward Removed
T11 T14
T17 T20Waveforms of VTA neurons:
T21 T22 T23 T24
Re-optimization of path after a reward is removed from a learned location
0º 30º 90º 150º 180º
-150º -90º -30º
Applegate, D., Bixby, R., Chvátal, V., et al. (2006) The Traveling Salesman Problem: A ComputationalStudy. Princeton, NJ: Princeton University Press.
Bures, J.,Buresova, O., and Nerad, L. (1992). Can rats solve a simple version of the traveling salesman problem? Behavioural Brain Research. 52, 133-142.
Fields, H., et al. (2007). Ventral Tegmental Area Neurons in Learned Appetitive Behavior and Positive Reinforcement. Annu. Rev. Neurosci. 30, 289-316.
?The ability of rats to optimize their path based on reward contingencies suggests that reinforcement may contribute to the rats’ ability to shift their strategy towards the optimal path.
?This shift is likely to depend at least in part on the interaction between brain structures that are involved in reward processing (VTA), spatial navigation (hippocampus) and planning (prefrontal cortex).
?Route optimization occurs within a configuration, not over sessions, which suggests that this task involves planning and short term memory, not long term memory.
?Rats will choose the optimal spatial or reward solution if presented separately – this choice is made more decisively with training. When both reward and spatial options are presented together, rats will shift strategy from reward to spatial optimization with training.
?Rats are able re-optimize when reward contingencies are changed.
?Optimization based on reward availability and quantity suggests the involvement of the dopamingeric system of the ventral tegmental area (VTA).
?Unlike in most TSP problems, spatial choice is biased by the current position and orientation of the rat.
Optimality Ratio = Path taken/min path length
Nu
mb
er
of P
ath
sN
um
be
r o
f P
ath
s
Starting Orientation (degrees)
0 30 90 150 180 -150 -90 -300
0.2
0.4
0.6
0.8
1
Pro
ba
bili
ty o
f C
ho
osi
ng
Rig
ht C
ity
City preference based on initial starting position
n=45 (4 rats)
n=3 rats
Influence of initial orientation in cases of ambiguous choices
Last TrialFirst Trial
Rats were randomly started at eight angles (0, 30, 90, 120, 180, 210, 270, 330) relative to center of arena.
Rats were allowed to visit only one city.
Type 1:Spatially Ambiguous
Reward Disambiguous
Type 2:Spatially Disambiguous
Reward Ambiguous
Type 3:Spatially Disambiguous Reward Disambiguous
Type 4:Spatially Disambiguous Reward Disambiguous
Worst Choice Best spatial + reward Reward Choice
Spatial Choice
Choose R = 1Choose L = 0
Influence of initial orientation in ambiguous choices
(Fields et al., 2007)
Intermediate Trial
Sample paths of a rat running trials of a Type 4 configuration
Spatially Optimal
Reward Optimal
Non-optimal
Reward + SpatiallyOptimal
Spatially Optimal
Reward Optimal
Non-optimal
Reward + SpatiallyOptimal
Spatially Optimal
Reward Optimal
Non-optimal
Reward + SpatiallyOptimal
Spatially Optimal
Reward Optimal
Non-optimal
Reward + SpatiallyOptimal
679.22
1 2 3Laurel Watkins de Jong , Gerard M. Martin , Jean-Marc Fellous1. 2. 3.
Undergraduate Biology Research Program (UBRP), Univ. of Arizona, Tucson, AZ Dept. of Psychology, Memorial University, CA Dept of Psychology and program in applied Mathematics, Univ. of Arizona, Tucson, AZ.
The Traveling salesrat: Insights into optimal spatial navigation and the role of the dopaminergic system.
Sample Paths
Sammy. Jr
Darwin
DarwinSammy Jr.
Indiana
0 5 10 15 20 timed-out0
1
2
3
4
5
Trials until re-optimization
Nu
mb
er
of o
ccu
ren
ce
s
n=25 (4 rats)
679.22