Fig 1.
Connectionist diagram of RWDDM.
Each CS unit is connected to a summing junction (labelled Σ) via a modifiable link V. The output of the summing junction is the CR. The US is represented as a teaching signal with a fixed weight H. Each CS unit has its own timer Ψ and representation x. The bottom panel shows a zoomed-in view of the timer Ψl and CS representation xl associated with CSl. The timer slope Al is tuned to a 5-second CS duration.
Fig 2.
RWDDM timer and CS representation during three 12-trial timing scenarios.
Top two rows: timing a novel 6 second stimulus. Timer starts with a low baseline slope (A = 0.001) on trial 1 and gradually adapts over training to reach approximately the required slope. Middle two rows: stimulus duration change from 6 to 3 seconds. Bottom two rows: stimulus duration change from 6 to 12 seconds. Parameters: αt = 0.215, θ = 1, σ = 0.25, m = 0.15.
Table 1.
Summary of the main features of the models.
Table 2.
Model features and the experimental findings they can explain.
Table 3.
Simulation designs.
Fig 3.
Acquisition and reacquisition.
Top left: simulated associative strength V in acquisition and extinction. Top middle: adaptation of RWDDM slope A. CR extinction began at trial 80 but has no effect on the RWDDM slope. Top right panel: simulated V curves in acquisition and reacquisition. Bottom left panel: response strength data from an experiment in acquisition and extinction, redrawn from Fig 1 in [77]. Bottom right panel: data from an experiment in acquisition and reacquisition, redrawn from the top panel of Fig 3 in [77]. Model parameters: m = 0.15, θ = 1, σ = 0.3, αt = 0.1, αV = 0.1, H = 4 in acquisition and H = 0 in extinction.
Fig 4.
Left column: simulated response strength averaged over trials in extinction short-long (top) and long-short (bottom). Middle column: time estimate adaptation of the model during extinction short-long (top) and long-short (bottom). Right column: experimental data from an experiment where the CS duration changed from 12-sec in acquisition to either 24-sec (top) or 6-sec (bottom) in extinction. Data plots redrawn from Figure 10 in [80]. Model parameters: m = 0.25, θ = 1, σ = 0.35, αt = 0.08, αV = 0.09, H = 30.
Fig 5.
Left panel: model V values for each CS duration in extinction. Middle panel: simulated CR values calculated only for the first 10 seconds of the CS. Each data point is calculated by summing the output of eq (10) over the first 10 sec of each trial, then averaging these trial values two by two, and dividing by 100 to rescale. Right panel: actual CR data for the first 6 sec of the CS in extinction, redrawn from Figure 8 (C) in [80].
Fig 6.
Top row: simulated associative strength in latent inhibition (left), simulated CR averaged over the first 30 trials of conditioning phase (middle), and simulated CR averaged over the last 30 trials of conditioning phase (right). Bottom row: acquisition curves from an actual experiment in latent inhibition (left), and response rate data during the CS (right). Data plots redrawn from Figures 1 and 2 respectively in [70]. Model parameters: αt = 0.1, αV = 0.08, μ = 1, σ = [0.6 − 0.35], m = 0.2, H = 4, αPH = 0.4, γ = 0.03.
Fig 7.
Experimental designs from two blocking experiments.
CS X was blocked (B) in rows 1 and 2, and not blocked (NB) in rows 3 and 4. Blue bar indicates US presence.
Fig 8.
Blocking with different durations.
Left column: simulation (top) with a 15 sec blocking CS and 10 sec blocked CS, and animal data (bottom) from an experiment with the same design. Right column: simulation (top) with a 10 sec blocking CS and 15 sec blocked CS, and animal data (bottom) from an experiment with the same design. Data panels redrawn from the top right panel in Figure 5 in [86]. Model parameters: αt = 0.2, αV = 0.1, μ = 1, σ = 0.35, m = 0.2, H = 10.
Fig 9.
Left column: simulation (top) and data (bottom) from conditioned inhibition with a long inhibitor. Right column: simulation (top) and data (bottom) from conditioned inhibition with a short inhibitor. Data plots redrawn from Figure 4 in [92]. Model parameters: αt = 0.09, αV = 0.06, μ = 1, σ = 0.35, m = 0.16, H = 30.
Fig 10.
Disinhibition of delay and compound peak procedure.
Top row: simulation (left) and data (right) of disinhibition of delay. Bottom row: simulation (left and middle) and data (right) of a compound peak procedure. The middle panel is a normalized (proportion of maximum response strength) version of the left panel. Data plot redrawn from Figure 13 in [44]. Model parameters: m = 0.25, θ = 1, σ = 0.18, αt = 0.75, αV = 0.1, H = 5.
Fig 11.
Top row: simulated average response rate during CSs (left), associative strength over trials (middle), and superimposition of response curves (right). Bottom row: average response rate data from an FI experiment, redrawn from bottom right panel of figure 4 in [102]. Model parameters: m = 0.15, θ = 1, σ = 0.3, αt = 0.2, αV = 0.1, H = 5.
Fig 12.
Left: simulated response strength during long trials. Right: response strength data from a mixed FI experiment, redrawn from Figure 3 in [104]. Model parameters: αt = 0.2, αV = 0.1, μ = 1, σ = 0.425, m = 0.2, H = 30.
Fig 13.
Top row: simulated average response strength during peak trials (left), and the same data plotted after both axes are normalized (right). Bottom row: average response strength data from an experiment in VI and FI, redrawn from Figure 1 in [108]. Model parameters: αt = 0.1, αV = 0.1, μ = 1, σ = 0.3, m = 0.2, H = 40.
Fig 14.
Top row: simulated response strength averaged over peak trials in temporal averaging (left), and the same data normalized by maximum response strength and peak time (right). Bottom row: peak trial response strength data from an experiment in temporal averaging, redrawn from Figure 1 in [110]. Model parameters: αt = 0.2, αV = 0.1, μ = 1, σ = 0.35, m = 0.2, H = 30.
Table 4.
Summary of main simulation results and comparison with other models.
Notes: (1) if learning rate is allowed to vary.