Decoupled Representation Learning
A paradigm that treats representation learning as an independent pre-training stage decoupled from downstream task optimization, producing representations that can be reused across models and scenarios and reducing engineering and computational costs.
Timeline (last 90 days)
The semantic channel design draws on the channel idea from multi-task learning and disentangled representation learning, splitting multiple semantic factors originally entangled in a single embedding space into several independent subspaces for separate modeling.
LUMIR separates representation learning from the optimization objective of the downstream recommendation model, so the learned representations can be trained and reused independently of a specific model.
All Facts (2)
The semantic channel design draws on the channel idea from multi-task learning and disentangled representation learning, splitting multiple semantic factors originally entangled in a single embedding space into several independent subspaces for separate modeling.
50%UnverifiedLUMIR separates representation learning from the optimization objective of the downstream recommendation model, so the learned representations can be trained and reused independently of a specific model.
50%