Connecting the Semantic Dots: Zero-shot Learning with Self-Aligning Autoencoders and a New Contrastive-Loss for Negative Sampling.
Mohammed Terry-JackNikolai RozanovPublished in: ICMLA (2022)
Keyphrases
- semantic information
- image registration
- natural language
- semantic relationships
- semantic web
- knowledge transfer
- positive and negative
- random sampling
- semantic network
- semantic similarity
- sampling strategy
- semantic representation
- semantic analysis
- semantic web technologies
- image search
- domain specific
- denoising
- high level
- information retrieval
- semantic annotation
- domain ontology
- monte carlo
- semantic features
- semantic knowledge
- black and white
- sampling methods
- feature selection