Multimodal Neurons in Artificial Neural Networks

GPTKB entity

Statements (27)
Predicate Object
gptkbp:instanceOf scientific theory
gptkbp:defines Neurons in artificial neural networks that respond to multiple types of input modalities, such as images and text.
gptkbp:enables zero-shot learning
multimodal generation
cross-modal understanding
multimodal retrieval
semantic alignment
gptkbp:field gptkb:artificial_intelligence
gptkb:machine_learning
gptkbp:firstDescribed 2021
https://www.w3.org/2000/01/rdf-schema#label Multimodal Neurons in Artificial Neural Networks
gptkbp:inspiredBy biological neurons
gptkbp:notableExample gptkb:DALL-E
gptkb:GPT-4
CLIP model
gptkbp:notablePublication Multimodal Neurons in Artificial Neural Networks (OpenAI, 2021)
gptkbp:property can be probed with both text and image inputs
emerge in large-scale neural networks
respond to abstract concepts across modalities
gptkbp:relatedTo gptkb:artificial_neural_networks
deep learning
representation learning
multimodal learning
gptkbp:studiedBy gptkb:OpenAI
gptkb:Google_Research
gptkbp:bfsParent gptkb:Gabriel_Goh
gptkbp:bfsLayer 7