The cognitive mechanisms of optimal sampling.

Area of Science:

• Behavioral Ecology
• Computational Neuroscience
• Animal Behavior

Background:

• Animals must adapt foraging strategies to unpredictable prey densities.
• The explore-exploit trade-off is crucial for survival and efficient resource acquisition.

Purpose of the Study:

• To simulate and compare various models explaining how animals learn and adapt to changing prey densities.
• To determine which computational models best predict animal performance in dynamic foraging environments.

Main Methods:

• Simulated several computational models: optimising, dynamic backward sampling, matching law, Rescorla-Wagner, neural network, epsilon-greedy, and rule of thumb.
• Used a two-armed bandit task simulating changing reward probabilities.
• Analyzed model performance under varying session lengths and reward differentials.

Main Results:

• All models identified the more profitable reward source, especially with greater probability differences.
• Only dynamic programming showed faster switching to exploitation with less available time.
• A rule of thumb model was most successful overall; neural network also performed well.

Conclusions:

• The rule of thumb model provides a robust explanation for animal foraging decisions in unpredictable environments.
• Animal foraging strategies involve complex learning and adaptation mechanisms, with simpler heuristics often being highly effective.