Chronos-2 turns a time-series foundation model into one that models how series interact
AspireAI community · arXiv
Amazon's Chronos-2 extends time-series foundation models past the univariate case: rather than treating each series independently, it models interactions among multivariate series and heterogeneous covariates, inferring relationships directly from historical context and emitting probabilistic forecasts without task-specific treatment. The authors report that across various benchmarks Chronos-2 achieves state-of-the-art accuracy on average, improving skill scores by 5.2 percentage points relative to the strongest baselines and increasing win rates by 10 percentage points. It is also cheap to run: the paper reports forecasts for 100 time series in under one second on a single mid-range GPU, 50x and 1000x faster end-to-end than the best task-specific point and probabilistic forecasters respectively. The same pretrained model is shown matching or surpassing specialised methods in infectious-disease forecasting, macroeconomic modelling, carbon-flux prediction and runoff simulation. The honest limitation is scope: this is an arXiv preprint whose reported wins are averaged across benchmark suites, and a separate earlier Chronos-2 technical report (arXiv:2510.15821) covers the base model that this work extends.
Read the story and explore more from the AspireAI community.
Two sparse-autoencoder analyses have resisted scaling: matching semantically similar features across layers, and compressing large feature circuits into interpretable supernodes. This paper argues both are the same underlying problem, estimating semantic distances between SAE features that sit on different activation manifolds. Its representation treats each feature not as a single decoder vector but as an activation-weighted distribution over the hidden states expressing it, projects those distributions into a shared reference space, and compares them with Wasserstein distance. The representation is proved invariant to activation rescaling and stable under perturbation, and recovers true matches under finite-sample margin conditions. Empirically it outperformed decoder-vector and LLM-based baselines and captured subtle functional distinctions between related features, and it compressed large feature circuits into interpretable supernodes automatically. The limitation is that this is interpretability tooling: the paper demonstrates better feature matching and circuit compression, not that downstream interpretability claims become more accurate.