These are matrixes often used for Text Embeddings. Each column should represent the embedding value for each word. Each input to the matrix should be a One Hot Encoded vector.