Mixing Synthetic Data Sources Outperforms Selecting Single Generators for Time Series Foundation Models
A new study accepted at ICML 2026 finds that equally mixing multiple synthetic data generators during pretraining of time series foundation models matches or outperforms using any single generator alone. The research evaluated 11 generator families across two architectures—Chronos-T5-Mini and Moirai-Small—finding that generator rankings are unstable across model types, making principled selection difficult. The findings reframe synthetic pretraining as a corpus composition problem rather than a generator selection problem, with practical implications for how AI researchers build training datasets for forecasting models.
Researchers have found that the choice of synthetic data generator for pretraining time series foundation models can produce up to a 2× gap in forecasting error under identical training budgets, yet no principled method for making that choice has existed. Testing 11 generator families on two architectures trained from scratch—Chronos-T5-Mini and Moirai-Small—the study found that which generators perform best is not consistent across model architectures, undermining any universal selection strategy. Rather than solving the selection problem directly, the authors propose sidestepping it: an equal-weight mixture of all generators matches or beats the best individual generator for both architectures. Combining this mixture with real-world data produced the strongest pretraining corpora overall. The study concludes that synthetic pretraining should be treated as a corpus composition problem, and that composition choices must be validated per model family rather than assumed to generalize. The paper was accepted at the ICML 2026 Workshop on Foundation Models for Structured Data, to be held in Seoul, South Korea.
What's missing
The study was conducted on models trained from scratch rather than fine-tuned from existing large pretrained checkpoints, which may limit generalizability to production-scale foundation models. The paper does not report results across diverse downstream forecasting domains beyond the benchmarks used, leaving open whether the mixture advantage holds in specialized fields such as finance or healthcare time series. Computational cost of the mixture approach versus single-generator training is not discussed.
What different sources said
- arXiv cs.LGCenter
Mix, Don't Pick: Why Synthetic Corpus Composition Matters for Time Series Foundation Model Pretraining
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.