Learning to Predict: A Fast Re-constructive Method to Generate Multimodal Embeddings | Read Paper on Bytez