Notes on Retroactive Interference Model of Forgetting

Mikhail Katkov

2023-12-26

Abstract:We present analytical derivation of the minimal and maximal number of items retained in recently introduced Retroactive Interference Model of Forgetting. Also we computed the probability that two items presented at different times are retained in the memory at a later time analytically.

Neurons and Cognition

What problem does this paper attempt to address?

The problem that this paper attempts to solve is about two key mathematical properties in the memory forgetting model. Specifically, the paper focuses on the following two issues in the Retroactive Interference Model of Forgetting:

1. **Minimum and maximum number of retained items**:
- The paper derives how many memory items can be retained at least and at most in different presentation order arrangements under a given memory implementation (i.e., a set of memory value vectors).
- In particular, the paper proves that under any presentation order, at most \( T \) items can be retained (where \( T \) is the number of time steps), and the distribution of the least retained items is the same as the distribution of retained items in a random presentation order with dimension \( n - 1 \).

2. **Pair - wise retention correlation**:
- Through numerical simulation, the paper observes that in cases where the dimension is greater than 1, the retention correlation between memory items presented at different times is not zero.
- The paper provides an analytical derivation of these correlations, which are purely caused by the partially ordered set (poset) properties of the model.

### Formula summary

- **Minimum and maximum number of retained items**:
- Maximum number of retained items: \[ \text{Maximal number of retained items} = T \]
- Distribution of the minimum number of retained items: \[ P(|M_n(T; V_{\sigma(T, V)})| = k) = P(|M_{n - 1}(T; V)| = k) \]

- **Pair - wise retention correlation**:
- Derivation of the retention function: \[ RC_n(T) = \int_0^1 du_1 \ldots \int_0^1 du_n \left(1 - \prod_{d = 1}^n (1 - u_d)\right)^T \]
- Matrix representation of the retention probability: \[ P(u; t + 1) = E_r(u) P(u; t) \]
- Transition matrix: \[ E_r(u) = \begin{pmatrix} 1 & \prod_{d = 1}^n (1 - u_d) \\ 0 & 1 - \prod_{d = 1}^n (1 - u_d) \end{pmatrix} \]

### Explanation

1. **Minimum and maximum number of retained items**:
- By analyzing the properties of the partially ordered set (poset), the paper proves that under any permutation order, at most \( T \) items can be retained. This is because if arranged in descending order of a certain dimension, each item will not be erased by subsequent items.
- The distribution of the least retained items is the same as the distribution of retained items in a random permutation with dimension \( n - 1 \). This is because if arranged in ascending order of a certain dimension, each item will not be erased by subsequent items.

2. **Pair - wise retention correlation**:
- Through the analysis of the Markov process, the paper derives the retention correlation between memory items presented at different times. These correlations reflect the partially ordered set (poset) properties of the model, that is, whether an item will be erased depends on the relationship between its values in various dimensions and the values of other items.

These results are helpful for designing experiments to verify the basic assumptions of the model and provide a theoretical basis for further research on the forgetting mechanism of human memory.