What question did this study set out to answer?

This research aims to introduce an efficient gating mechanism using TERA projection in mixture of experts models.

synapse

⌘+K

synapse

⌘+K

June 18, 2026Open Access

Vortex-Gated Mixture of Experts: Symbolic, Zero-Parameter Routing via TERA Projection

Key Points

This research aims to introduce an efficient gating mechanism using TERA projection in mixture of experts models.
Developed Vortex-Gated MoE (VG-MoE) with a deterministic, closed-form TERA projection.
Verified numerical claims using the harmonic-engine open-source package.
Conducted tests to ensure functionality and alignment with routing math.
Confirmed that VG-MoE routes correctly using zero gate parameters, enhancing processing efficiency.
Achieved passing tests within the harmonic-engine suite, validating the proposed mathematical model.

Abstract

Vortex-Gated MoE (VG-MoE): a drop-in MoE gate using a deterministic, closed-form TERA projection with zero gate parameters. This revision adds a Reference Implementation & Verification note to Data Availability: the paper's falsifiable claims are now numerically verified in the open-source harmonic-engine package (tests/testₙewₘath. py, passing on CPU). The VortexGate routing math and HeartScale alignment scalars are exercised by the harmonic-engine suite.

Read Full Paperexternally

Mark Helpful

Bookmark

Relay

View Full Paper