OceanMoE: Conditional Sparse MoE Cuts Error in Long-Horizon Ocean Forecasting
burny_tech · x · 2026-09-21
A new arXiv paper proposes OceanMoE, a structured conditional sparse Mixture-of-Experts framework for long-horizon multivariate ocean forecasting.
The core idea balances cross-variable sharing with target-specific specialization in a unified model:
- Fuses cross-variable information into target-specific local representations and performs content-conditioned sparse routing at each spatial location, with the number of active experts adapting to router confidence.
- Decoder routing adds a learned geographic bias parameterized by spherical-harmonic spatial bases.
- Shared residual and seasonal pathways supply common cross-variable and month-dependent context.
On long-horizon autoregressive ORAS5 forecasting, OceanMoE lowers aggregate error in both evaluated settings and keeps lower geometric-mean normalized RMSE than the corresponding baselines.
More from Research
- Loopback Transformer matches 3B model with 75% of parameters, gains mostly from decay — pbaylies · 2026-11-23
- RoboQuest benchmark: best AI agent solves only 23.2% of embodied exploration tasks — omarsar0 · 2026-11-23
- OpenAI publishes internal pretraining perplexity curve, proposes perplexity as shared safety bar — merettm · 2026-11-23
- AI math swarm proves a 962-variable quartic is minimal; 4,973-variable version spans 1.62 MB of text — gleech · 2026-11-23
- Even OpenAI's quasi-Riemann result already spawned vast expository work, argues AI-proof skeptic — gleech · 2026-11-23
- CMU's ExploitBench ranks LLM cyber agents on a capability ladder, not pass/fail — goyal__pramod · 2026-11-23