pixel-based offline reinforcement learning

Tag "pixel-based offline reinforcement learning" — all articles

062026-11Two architecture studies advance: Loopback Transformer boosts efficiency, MAC-v2 strengthens pixel-based offline RLarXiv posted two machine learning papers on Nov 6, 2026, proposing the Loopback Transformer and MAC-v2 architectures. The former separates input strength from update weights to achieve lower validation loss with 40-75% of the parameters; the latter combines a reconstruction-free latent world model with action chunking and performs strongly across several vision benchmarks. Both are preprints and have not yet been peer-reviewed.arXiv cs.LG