Figure 5.

A hybrid memory model: (a) Hybrid learning rule was defined as the weighted sum of AR and SR models, and the portion of SR was denoted by α. (b) Temporal decay of the memory in the hybrid model was tested as in Fig. 3c. The hybrid model with α = 0.5 showed intermediate characteristics between the AR and SR models (orange line). The error bars represent the standard deviation for 100 trials. (c) Modulation of memory decay for various values of α: As α increased, the ratio of memory indices (MI ₈₀₀/MI ₀) increased linearly, suggesting that the characteristics of the hybrid model make a linear transition from the AR to the SR model. (d) The decay of old memory with newly appended memory in the hybrid model was tested as in Fig. 4c. The memory index in the hybrid models with α = 0.5 and 0.99 were intermediate values between those of the AR and SR models, but showed nonlinear transition as α varied. (e) The proportion of old memory was maintained as α varied: As α increased from 0 to 0.99, the ratio of memory indices (MI ₈₀₀/MI ₀) increased linearly. After this, the ratio abruptly increased as α changed from 0.99 to 1 (Wilcoxon signed rank test: *p < 10⁻⁹, n = 100). The data point at α = 1 was excluded from the linear regression (dashed line). The error bars represent the standard deviation.