Benchmarking time-series foundation models (Chronos-Bolt, zero-shot) vs. supervised (PatchTST) and classical (seasonal-naive, Croston) baselines on the M5 Walmart dataset, scored with MASE and WQL. No single model dominates: foundation/deep models win on dense SKUs, classical methods win on the intermittent tail.

Tokenization Matters: A Fair Ablation of Point-wise, and Variate-wise Transformers for Financial Time Series. (includes PatchTST, iTransformer, Crossformer, Autoformer, Fedformer, Informer, TimeNet, and non-stationarity extensions.

Heuristics-free self-supervised representation learning for time series with SIGReg (LeJEPA). Disentangles time-axis collapse, positional structure, and representation richness across PatchTST, TCN, and bag-of-patches encoders. Reproducible, seeded, significance-tested.

chatbot designed to allow users to interact with transformer models

Comparison of Return Forecasting Methods for Markowitz Portfolio Optimization: Historical Mean, AutoARIMA, PatchTST Transformer

Time-series foundation model fine-tuning toolkit with GPU acceleration

SOTA time-series models for industrial sensor streams · ETT, SWaT · IEEE TII

9. PatchTST: Channel-Independent Patching Transformer. Hibridne arhitekture za predikciju koje kombinuju deep learning i klasične time-series modele.

Disease Forecasting in a Tropical Context: A Comparative Evaluation of Model Performance and Generalizability for Dengue Fever and Influenza in Vietnam

PatchTST discharge forecasting with conformal prediction bands — Reno river, station 425 (Progea/TOPKAPI)

Agentic multi-strategy hedge fund: PatchTST forecasts, 4-agent LangGraph debate, CPCV-OOS + DSR validation, HRP with Ledoit-Wolf shrinkage. 10-year OOS Sharpe=0.766.