Papers with re-constructive and associative nature of human memory
Associative Multichannel Autoencoder for Multimodal Word Representation (D18-1)
Copied to clipboard
| Challenge: | Existing models that represent word meanings from word co-occurrences ignore associations between modalities and lack ability to transfer information between . |
| Approach: | They propose a novel associative multichannel autoencoder that integrates textual, visual and auditory inputs to learn multimodal word representations. |
| Outcome: | The proposed model outperforms strong unimodal models and state-of-the-art models on six benchmark concepts similarity tests. |