<ul><li>An empirical validation of directional non-commutative monoidal embedding framework was presented.</li><li>The framework uses distinct non-commutative operators per dimension and generalizes classical one-dimensional transforms.</li><li>The study applied this framework to image classification on the MNIST dataset and compared it with fixed DFT-based embeddings.</li><li>Results show that the learned monoidal embeddings outperformed fixed DFT-based embeddings, confirming the effectiveness of directional non-commutative monoidal embeddings in representing image data.</li></ul>

Directional Non-Commutative Monoidal Embeddings for MNIST

Discover more