CLaST Improves Probabilistic Time Series Forecasting with Context-Aware VAE

Alexander Marusov, Dmitry Anikin, Petr Sokerin, Vitaliy Pozdnyakov, Ilya Kuleshov, Alexey Zaytsev· August 21, 2026 View original

Key takeaways

  • CLaST significantly improves probabilistic time series forecasting accuracy.
  • It uses a contrastive VAE to learn context-aware temporal dependencies.
  • The framework shows strong performance in both short-term and long-term predictions.
  • Better latent representations lead to more expressive and reliable forecasts.

Who benefits

FinanceEnergyHealthcareTransportationSupply Chain

Summary

CLaST is a new VAE framework for probabilistic multivariate time series forecasting that learns contextually similar embeddings through a contrastive loss function. It consistently outperforms strong baselines, achieving significant improvements in both short-term and long-term prediction accuracy across various benchmarks.

Probabilistic forecasting for time series, crucial in fields like energy, finance, and medicine, often struggles to capture complex internal temporal dependencies, leading to less expressive latent representations in deep generative models. To overcome this, researchers introduce CLaST (Context-aware Contrastive VAE for Probabilistic Time Series Forecasting), a Variational Autoencoder (VAE) framework designed for multivariate time series. CLaST's innovation lies in its ability to learn embeddings that preserve contextual similarity between observations, achieved through a novel contrastive loss function. Extensive experiments across nine widely used benchmarks demonstrate CLaST's superior performance. It achieves up to 16.4% improvement in CRPS and 14.4% in NMAE for short-term forecasting, and even more substantial gains of up to 48.6% in CRPS and 25.1% in NMAE for long-term predictions compared to the second-best methods.

Why it matters

Professionals relying on accurate probabilistic time series forecasts in critical domains can benefit from CLaST's enhanced ability to capture temporal dependencies, leading to more reliable predictions and better decision-making.

How to implement this in your domain

  1. 1Investigate CLaST as a potential upgrade for existing probabilistic forecasting models in your domain.
  2. 2Experiment with integrating contrastive learning principles into other generative models for time series.
  3. 3Benchmark CLaST's performance against current forecasting solutions on your specific datasets.
  4. 4Explore how the learned context-aware embeddings can provide additional insights into time series dynamics.

Original post by Alexander Marusov, Dmitry Anikin, Petr Sokerin, Vitaliy Pozdnyakov, Ilya Kuleshov, Alexey Zaytsev

"arXiv:2608.20025v1 Announce Type: new Abstract: Probabilistic forecasting models are widely used for time series forecasting in domains such as energy systems, finance, medicine, and transportation. In recent years, deep generative models have shown strong results on probabilisti…"

View on X

Originally posted by Alexander Marusov, Dmitry Anikin, Petr Sokerin, Vitaliy Pozdnyakov, Ilya Kuleshov, Alexey Zaytsev on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses

More in AI Research

AI ResearchAI Engineering & DevTools

Decoding Silent Reading from Non-Invasive EEG

This research demonstrates that open-vocabulary word-level and semantic information can be reliably decoded from non-invasive EEG during silent reading. Using a contrastive decoder and a large dataset from a single participant, the study shows decoding scales log-linearly with training data and extends to rare words.

Ingo Marquardt, Anthilia Alchanat, Priyanka JainAug 21, 2026
AI ResearchAI Engineering & DevTools

Exact Learning Coefficients for Singular Models

This paper presents the first deterministic algorithm for exactly computing local learning coefficients (Real Log Canonical Thresholds) for two-dimensional singular models. This breakthrough provides ground truth for calibrating sampling-based estimators and reveals algebraic structure in learning coefficients, outperforming sampling in shallow regimes.

Gr\'egoire Sergeant-Perthuis (CQSB, Sorbonne Universit\'e), Elias Tsigaridas (Ouragan Team, INRIA), Jules Tsukahara (Ouragan Team, INRIA)Aug 21, 2026
AI Engineering & DevToolsAI Research

Standardized ML Evaluation for Power System Protection

This paper proposes a standardized framework for evaluating machine learning applications in power system protection, addressing inconsistencies in current research. It defines seven critical study dimensions and instantiates the framework with a case study on fault classification and localization using a public benchmark.

Julian Oelhaf, Georg Kordowich, Paula Andrea P\'erez-Toro, Christian Bergler, Johann J\"ager, Andreas Maier, Siming BayerAug 21, 2026