An unreviewed technical study identifying a phenomenon in foundation models with a proposed solution, lacking validation on held-out test data or independent benchmarks.
Reported
Sample size (assets)1000
Models testedtwelve deep-learning forecasting models
Benchmark configurations97