The Dynamic Key-Value Memory Network (DKVMN) model works as follows: at time $t$, it first receives a knowledge component (KC) $q_t$, then predicts the probability of answering $q_t$ correctly, and eventually updates the memory using the question-and-answer interaction $(q_t, a_t)$.

Assuming there are $Q$ different KCs and $N$ latent concepts, these latent concepts are stored in a key memory $\mathbf{M}^k \in \mathbb{R}^{N \times d_k}$, where $d_k$ denotes the embedding size of the key memory slot. The corresponding knowledge states are stored in a value memory $\mathbf{M}^v \in \mathbb{R}^{N \times d_v}$.

The DKVMN operates in three major steps:

1. Getting Attention Weight
2. Making Prediction
3. Updating Value Memory