Goal

Implement and test simple models of learning (habituation) and attention shifting, and test them for stereotyped looking time behavior (e.g., the Hunter & Ames 1988 model).

Pelz et al. 2015 Model

The Pelz model has five components:

  1. the Gompertz learning curve that the learner follows when attending to a stimulus,
  2. decay of short term memory operating on unattended stimuli,
  3. the cost of switching attention between objects (0.5 bits), and
  4. the amount of information that the learner knows about the objects at the start of test, and
  5. softmax probabilistic decision rule to choose which attention between objects.

We plot the learning curve for an attended stimulus and decay curve (\(\beta = 0.1\)) for unattended stimuli below. The Gompertz learning curve has parameters \(a = 100\), the maximum amount of information (bits) that can be learned from each object; \(b = 5\), the amount of initial information, and learning rate \(c = 1\).

Simple simulations

We run a simple simulation with three objects, each with starting information of 1.

sim1 = run_sim(c(1,1,1))
long <- sim1 %>% pivot_longer(cols = starts_with("Object."), 
                              names_to="Object", 
                              names_prefix="Object.",
                              values_to="Information")
long %>% ggplot(aes(x=time, y=Information, group=Object, color=Object)) +
  geom_point() + geom_line() +
  theme_bw()

Poli et al. 2020 Model

Word Learning Models