Dynamic TMoE dynamically adjusts expert count in a MoE architecture for non-stationary time series via MMD shift detection and a temporal memory router, reporting SOTA results with 10.4% MSE and 7.8% MAE reductions on nine benchmarks.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Dynamic TMoE: A Drift-Aware Dynamic Mixture of Experts Framework for Non-Stationary Time Series Forecasting
Dynamic TMoE dynamically adjusts expert count in a MoE architecture for non-stationary time series via MMD shift detection and a temporal memory router, reporting SOTA results with 10.4% MSE and 7.8% MAE reductions on nine benchmarks.