EIDOS

Summary

EIDOS is a time-series foundation model family that trains on latent-space predictive dynamics with observation-space grounding and a SiGLU point-wise scalar tokenizer.

Role In The Wiki

EIDOS anchors the time-series version of latent predictive learning, showing how representation dynamics can improve robustness and forecasting. It is also the local anchor for point-wise numeric value embeddings in time-series models: each scalar observation is converted to a latent token before causal prediction.

Artifact Status

The source page records an expected local official code drop at external/EIDOS/ and checkpoint at external/EIDOS/eidos 1.pt, but those artifacts are absent in this checkout. Treat the paper source page as the current canonical local reference until the external artifacts are available.

Evidence

Relation To Foundation TSFM Agenda

Use the source-level agenda mapping in eidos-2026 rather than duplicating verdict rows here.

At the entity level, EIDOS anchors the time-series version of latent predictive learning, showing how representation dynamics can improve robustness and forecasting. It is also the local anchor for point-wise numeric value embeddings in time-series models: each scalar observation is converted to a latent token before causal prediction. This page should stay as the object card; source pages carry slot-level verdicts, evidence, and missing pieces.