Multimodal Representation Learning:Cross-modality and Shared Representation

Various kinds of Information retrieval or processing task can be difficult basis on how the information is viewed or represented. Representation learning is a technique that allows a system to discover the representation of the data which used for detecting features and classifying things. As compared with learning from one representation, learning from different representations make the task easier. For example, learning something from images and their texts or learning from video and their audio produces good results than learning from one source (either image or text)or (video or audio). Learning from multiple modalities helps to achieve the target correctly within short time. This paper tries to design a Multimodal Autoencoder with minimum loss that can produce shared representation of images and its texts