Neural Network Approaches Advance Forecasting of Sparse, Bursty Time Series Data
Two new machine learning studies present methods for forecasting intermittent time series—data with long periods of inactivity punctuated by sudden bursts—common in network operations and supply chain management. NetBurst uses an event-centric pipeline that separates burst timing from magnitude, while a comparative study finds that simpler global neural network models (TiDE) outperform traditional local models and complex architectures. These advances address a critical operational challenge where standard time-series models fail because they are designed for dense, regular data rather than sparse, heavy-tailed patterns.
Two recent machine learning papers tackle the problem of forecasting intermittent time series—datasets characterized by long stretches of low or zero activity interrupted by rare, extreme events. NetBurst, presented in the first study, introduces an event-centric pipeline that collapses inactive periods and separately models burst timing and magnitude, achieving 1.3–116× reductions in forecasting error on network telemetry data compared to strong baselines including Amazon's Chronos-2 and Datadog's Toto. The second study conducts the first comprehensive comparison of local versus global forecasting models on intermittent data, testing on over 40,000 real-world time series from supply chains. It finds that TiDE, a relatively simple neural network architecture with a Tweedie distribution head, consistently outperforms both traditional local models and larger global models while requiring fewer computational resources. Both studies highlight that existing time-series foundation models, optimized for dense periodic data, perform poorly on the 'wild regime' of sparse, heavy-tailed real-world data.
What's missing
Both papers are preprints on arXiv and have not undergone peer review at a published venue. The NetBurst study does not provide details on computational complexity or scalability to very large networks. The comparative study does not discuss how results might generalize to non-supply-chain domains or whether the findings hold for different types of sparsity patterns.
What different sources said
- arXiv cs.LGCenter
Intermittent time series forecasting: local vs global models
- arXiv cs.LGCenter
NetBurst: Event-Centric Forecasting of Bursty, Intermittent Time Series
Related
Genetic Drift, Not Selection, Drives Rapid Feather Color Evolution in Island Bird Radiation
A new study of an island bird radiation found that rapid evolution of feather coloration is driven primarily by genetic drift in small populations rather than sexual or ecological selection. The research integrated whole-genome data with detailed plumage measurements across complete species sampling to test whether signaling trait evolution correlates with speciation rates. The findings suggest that neutral demographic processes play a central role in generating phenotypic diversity during island radiations, challenging assumptions about the mechanisms driving rapid evolution.
New AI Model Improves Prediction of Therapeutic Peptide Function from Protein Sequences
Researchers developed a lightweight CNN classifier that predicts whether peptide sequences have therapeutic properties, trained on a database of 54,655 peptides across 48 functional categories. The model uses a novel negative sampling strategy to reduce false positive rates from over 60% in previous approaches to 2.1%. This advancement could accelerate drug discovery by enabling faster computational screening of peptide candidates before expensive experimental testing.
Study Shows Different Metabolic Stress Models Produce Distinct Effects on Human Neuronal Networks
Researchers tested three common in vitro metabolic stress models on human-derived neuronal networks and found each produced different patterns of neuronal activity and cell damage. The models tested were hypoxia alone, oxygen-glucose deprivation (OGD), and hypoxia combined with glutamate exposure. The findings suggest that choice of experimental model significantly affects results and that combining electrophysiological and structural analyses is important for accurately assessing metabolic stress in stroke research.