Open chufanchen opened 5 months ago
Cycle Consistency + Decision Transformer
Like world model, we learn a forward and back dynamic model. Then we use these DMs to constraint decision transformer training using cycle consistency constraints.
I'm curious about performance between World Model(MLP-based and transformer-base) and decision transformer.
crl: offline continual learning + online rehearsal
Cycle Consistency + Decision Transformer
Like world model, we learn a forward and back dynamic model. Then we use these DMs to constraint decision transformer training using cycle consistency constraints.