PBWM Neural Network Emulating Human Probability Matching
Solution Overview
Problem
Current models for probability matching and loss sensitivity among human subjects are not bio-inspired, limiting their ability to predict animal behavior and masking subgroups within the population, as they focus on aggregate behavior rather than biological roots.
Innovation Solution
A system that uses neural network models trained on human subject responses to extract features related to probability matching, incorporating a prefrontal cortex basal ganglia working memory (PBWM) model to emulate human performance and capture varying degrees of conservatism, allowing for the generation of models of human behavior for training and adversarial planning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If reinforcement learning algorithms or Bayesian graphical models are used to model probability matching, then the model can capture aggregate human behavior, but the model cannot predict animal behavior based on lesion studies or neurotransmitter imbalances because it is not bio-inspired
Solution Approach 1:
The patent copies the structure and function of biological neural circuits (prefrontal cortex and basal ganglia) to create a computational model that can predict both human and animal behavior. The PBWM model replicates the anatomical and functional organization of these brain regions, allowing it to generate predictions consistent with lesion studies and neurotransmitter research while maintaining predictive accuracy for probability matching behavior
Solution Approach 2:
The patent replaces abstract mathematical models (reinforcement learning algorithms, Bayesian graphical models) with a biologically-inspired neural network model that simulates the actual mechanical and chemical processes in the brain. This substitution allows the model to incorporate biological constraints and mechanisms, making it applicable to both human and animal systems while preserving its ability to model aggregate behavior
2Device complexity
If aggregate behavior modeling is used, then the model can simplify analysis, but it masks interesting subgroups or phenotypes within the overall population
Solution Approach 1:
The patent segments the population into distinct subgroups or phenotypes based on their probability matching behavior patterns. By dividing the overall population into heterogeneous subgroups, the model can analyze and predict behavior at both the individual and aggregate levels simultaneously, preventing the masking of interesting subgroups while maintaining manageable complexity through structured analysis
Data Source
AI summary
Described is a system for modeling probability matching in human subjects. Features related to probability matching are extracted from a set of human subject responses from behavioral tasks. Neural network model instances are trained on the set of features, resulting in a set of trained neural network model instances. A set of model parameters are derived from the set of trained neural network instances, and the set of derived model parameters are used to emulate human performance on novel data.