ARROW: New Model-Based Algorithm Addresses Catastrophic Forgetting in Continual Reinforcement Learning
Researchers have published ARROW, a model-based continual reinforcement learning algorithm that uses a dual-buffer replay system inspired by neuroscience to reduce catastrophic forgetting. ARROW extends the DreamerV3 framework by maintaining separate short-term and long-term memory buffers, matching the distribution of past experiences more efficiently than standard fixed-size FIFO buffers. The work, accepted in Transactions on Machine Learning Research, addresses a key scalability bottleneck in training AI agents that must learn new tasks without losing previously acquired skills.
ARROW (Augmented Replay for RObust World models) is a continual reinforcement learning algorithm designed to help AI agents acquire new skills while retaining old ones, a challenge known as catastrophic forgetting. Drawing on neuroscience research showing the brain replays experiences to a predictive world model rather than directly to a behavioral policy, ARROW extends DreamerV3 with a memory-efficient, distribution-matching replay buffer. The system uses two complementary buffers: a short-term buffer for recent experiences and a long-term buffer that preserves task diversity through intelligent sampling. The authors evaluated ARROW on Atari games, which lack shared structure across tasks, and Procgen CoinRun variants, which allow for knowledge transfer between tasks. Compared to model-free and model-based baselines using replay buffers of the same size, ARROW demonstrated substantially less forgetting on structurally unrelated tasks while maintaining comparable forward transfer on related ones. The paper, spanning 36 pages and 11 figures including appendices, was accepted to Transactions on Machine Learning Research in 2026. The authors conclude that bio-inspired, model-based approaches show meaningful promise for continual RL and merit further investigation.
What's missing
The paper does not report results on real-world or continuous-control benchmarks beyond Atari and Procgen, leaving open questions about generalization to robotics or other high-dimensional domains. Long-term scaling behavior as the number of tasks grows substantially remains an open question.
What different sources said
- arXiv cs.AICenter
ARROW: Augmented Replay for RObust World models
Related
Gut Bacteria Enzyme Found to Break Down Heat-Processed Food Compounds, Producing Novel Biogenic Amines
Researchers have discovered that an enzyme in common gut bacteria can degrade N-epsilon-carboxymethyllysine (CML), a compound formed during thermal food processing, producing previously unknown biogenic amines. The enzyme, ornithine decarboxylase SpeC from enterobacteria, acts on CML and related modified lysine derivatives through a low-level 'underground' catalytic activity. This finding suggests a previously unrecognized communication axis between thermally processed dietary compounds and gut microbial physiology, with potential implications for host health.
Full-Length Gene Sequencing Reveals Two Distinct Bacterial Communities in Black-Legged Ticks Expanding Into Canada
Researchers used Oxford Nanopore full-length 16S rRNA gene sequencing to characterize the microbiome of Ixodes scapularis black-legged ticks collected in Nova Scotia, Canada, distinguishing between tick-adapted bacteria and environmentally acquired bacteria. The study comes as I. scapularis — the primary vector of Lyme disease — is rapidly expanding northward into Canada due to climate change. The findings suggest that environmentally derived bacteria in tick microbiomes are not mere contamination, which has implications for how tick microbiome data is collected and interpreted across surveillance studies.
Study Identifies Metabolic Link Between Cell Envelope Stress and Biofilm Formation in Bacteria
Researchers have discovered that the metabolite acetyl-CoA directly inhibits enzymes that degrade the bacterial signaling molecule c-di-GMP, connecting cell envelope biosynthesis stress to biofilm formation in Pseudomonas aeruginosa. The study found that sub-inhibitory concentrations of antibiotics targeting early peptidoglycan biosynthesis — but not other antibiotic classes — elevate c-di-GMP levels by reducing phosphodiesterase activity, with acetyl-CoA competing for the enzyme active site. Because the relevant enzyme domain is broadly conserved across bacterial species, this checkpoint mechanism may be widespread and could have implications for understanding antibiotic-induced biofilm responses.