Search

AutoML Forecasting | ByteJournal

AutoML automates time series forecasting, especially in finance, by selecting algorithms and tuning hyperparameters with tools like Azure ML. It offers improved accuracy and efficiency but faces challenges in data quality, interpretability, and tuning effort.

Time Series Bootcamp: Forecasting with ML, Deep Learning, Foundation Models Tickets, Saturday, November 28  •  3 PM - 7 PM UTC | Eventbrite

This two-day bootcamp teaches participants to build and compare time series forecasting models using retail data and a shared evaluation framework. It covers methods from statistical forecasting to foundation models, with practical labs, reusable materials, and guidance on balancing forecast quality, business costs, and computing needs.

Time Series Bootcamp: Forecasting with ML, Deep Learning, Foundation Models Tickets, Saturday, November 28  •  3 PM - 7 PM UTC | Eventbrite

This two-day bootcamp teaches participants to build and compare time-series forecasting models on retail data, using statistical, machine-learning, deep-learning, and foundation-model methods. It includes practical labs, evaluation tools, notebooks, and guidance on choosing models for business needs.

Agriculture, artificial intelligence changes weather forecasts: up to 6 weeks to protect crops

Translated title. Original: Agricoltura, l’intelligenza artificiale cambia le previsioni meteo: fino a 6 settimane per proteggere i raccolti

An AI-based weather forecast method extends predictive horizons from two to six weeks, improving accuracy and enabling earlier planning for agriculture and related sectors. The Nature study, led by MIT, Microsoft Research, and collaborators, uses machine learning to correct past model errors and achieved top results in an ECMWF competition.

[2610.31248] Reducing Multi-Horizon Photovoltaic Forecast Error through Tree-Anchored Expert Combination

A tree-based PV forecast is improved by a nonnegative, MAE-optimized combiner that weights multiple temporal predictors; across 56k windows, the static combination beats the latest neural correction, suggesting selective ensemble use over equal averaging.

6-week weather forecasts: MIT revolutionizes precision

Translated title. Original: Previsioni meteo a 6 settimane: il MIT rivoluziona la precisione

MIT researchers unveil Probabilistic Bias Correction to extend forecast accuracy to 2-6 weeks, boosting sub-seasonal predictions and benefiting sectors like agriculture; the approach improves key weather variables, outperforms multiple centers in AI Weather Quest 2025, and enhances flood and heat-event forecasts with practical planning advantages.

Time Series Bootcamp: Forecasting with ML, Deep Learning, Foundation Models Tickets, Saturday, November 28  •  3 PM - 7 PM UTC | Eventbrite

This two-day bootcamp teaches participants to build and compare forecasting models using real-world retail data. It covers evaluation, backtesting, prediction intervals, and model selection, with practical labs and reusable materials.

Prediction accuracy improvement using ensemble methods of IBM Granite time series model | AIBloom

Translated title. Original: IBM Granite時系列モデルのアンサンブル手法による予測精度向上 | AIBloom

IBM Research presents Granite Timeseries Ensemble, integrating four diverse pre-trained models to better handle trends, seasonality, and local fluctuations without extra training. It offers probabilistic and IQR-based fusion methods, requires no weight fitting, and demonstrates strong MASE performance across many settings, with noted trade-offs in memory and cross-channel correlations.

Mistral AI launches Timestral: probabilistic forecasts on CPU

Translated title. Original: Mistral AI lance Timestral : prévisions probabilistes sur CPU

Mistral AI unveils Timestral, a CPU-optimized, probabilistic, multivariate time-series model available in four sizes (5M–209M parameters). It emphasizes efficiency, zero-shot deployment, and the ability to model past and future variables for dynamic systems, aiming for broad, cost-effective adoption over GPU-heavy architectures.

Can AI better predict what to stock before the season? - IT People Technology Platform

Translated title. Original: Dokáže AI lépe odhadnout, co naskladnit před sezonou? - Technologická platforma IT People

Forecasting retailer demand is needed to balance stock levels and avoid overstock or stockouts; new time-series models like Timestral show promise but must be tested against historical data and integrated with pricing, promotions, and replenishment rules to assess practical impact and uncertainty.