On the Transferability of VAE Embeddings using Relational Knowledge with Semi-Supervision
This work addresses transfer learning challenges in semi-supervised VAEs, but it is incremental as it builds on existing relational and VAE methods.
The paper tackled the problem of balancing disentanglement and low complexity modeling in relational VAE semi-supervision, finding that enforcing structure on representations improves zero-shot transductive transfer but can negatively impact inductive transfer.
We propose a new model for relational VAE semi-supervision capable of balancing disentanglement and low complexity modelling of relations with different symbolic properties. We compare the relative benefits of relation-decoder complexity and latent space structure on both inductive and transductive transfer learning. Our results depict a complex picture where enforcing structure on semi-supervised representations can greatly improve zero-shot transductive transfer, but may be less favourable or even impact negatively the capacity for inductive transfer.