On optimal decision-making in brains and social insect colonies Marshall, Bogacz, Dornhaus, Planque, Kovacs,Franks Presented by Reffat Sharmeen, Kolli Sai Namratha
Contents Introduction Optimal decision-making Decision making in cortex Usher-McClelland model Decision making in social insect colonies Models of house-hunting Discussion
Introduction Animals constantly invest time and energy to make decisions. Need to compromise between speed and accuracy. Decision making model in primate brain is compared to house hunting models by social insect colonies. Striking parallels are evident between decision-making in primate brains and collective decision-making in social insect colonies.
Optimal decision-making Uncertain information are processed to choose among alternatives. Example : Follow a display filled with moving dots. Decision making process can be represented as Brownian motion on a line moving towards the correct hypothesis, known as diffusion model. Sequential probability ratio test(sprt) gathers evidence for two hypothesis until likelihood ratio reaches a positive or negative threshold.
Diffusion Model
Decision making in cortex Neurons in medial Temporal area (MT)are responsible to process the motions in visual field. Neurons in lateral intraparietal area(lip) and frontal eye field control eye movement. Over time LIP neurons integrate input from MT neurons and accumulate sensory evidence. When LIP neuron s activity is over a threshold, the decision is made and eye is moved to the corresponding direction.
Usher-McClelland model A decision making model in primate brains. Each neural population receives noisy input signal and inhibits activation of the other to a degree proportional to its own activation. These populations leak incoming evidence. If activity of either of the populations reaches a threshold, the decision is made. Based on parameters Usher McClelland model approximates optimal decision making.
Usher-McClelland model
Decision making in social insect colonies Honeybee and ant colonies hunt for new nest sites. Trade off between emigration duration and information about potential nest sites. Ant scouts discover site, recruit nest mates who teach others the route, thus making a collective decision based on positive feedback. Bee scouts discover sites, recruit others for positive feedback and switch to the new site after the decision has been made. No central control, individuals use only local information.
House-hunting in T.albipennis Ants switch directly from uncommitted to committed state by discovering site and becoming recruiters for the new site. Recruiters for a site can switch to recruit for other site or switch to being uncommitted to any site. Decision will be optimal if individuals have global knowledge about the alternatives available, which makes this model biologically unrealistic.
House hunting with indirect switching Committed scouts should be completely uncommitted to change their commitment. Uncommitted scouts can spontaneously discover alternative sites at a rate which is independent of site quality. This model cannot be reduced to two independent random processes. Also it does not asymptotically converge to the diffusion model, so it cannot be a statistically optimal decision making strategy.
House hunting with direct switching Scouts can directly switch their commitment between alternative sites. During the emigration process honeybees enter in decision making phase and number of alternative sites are reduced. Thus a close, inferior, easy to find site may be chosen due to positive feedback. When all scouts are committed in the colony, decision is optimal. Without decay this model can be described as asymptotically optimal.
Direct switching model
Discussion First optimality hypothesis for collective decision making during emigration for social insect colonies. Formally investigated similarities between neural decision making process and collective decision making in social insect colonies. Direct switching model approximates statistically optimal decision making. If direct switching does not occur their hypothesis can at least theoretically quantify the cost of deviation from optimality.
Discussion Only binary decision case was considered. In real world optimal decision making is harder for more than two alternatives. Information about all the alternatives is not available from the beginning, discovery of best available alternative may be quite late. Bandit problem- should scouts evaluate existing alternatives or discover unknown alternatives.