Classifier-free diffusion with offline/online data-type labels improves aggregated D4RL fine-tuning scores in the reported runs, but the ablation table shows the full method performing worst of the compared variants.
The introduction of the diffusion model does indeed entail an inevitable increase in computational and time costs
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Offline-to-Online Reinforcement Learning with Classifier-Free Diffusion Generation
Classifier-free diffusion with offline/online data-type labels improves aggregated D4RL fine-tuning scores in the reported runs, but the ablation table shows the full method performing worst of the compared variants.