Geometric Routing Enables Causal Expert Control in Mixture of Experts

📅 2026-04-15
📈 Citations: 0
Influential: 0
📄 PDF

career value

215K/year
🤖 AI Summary
This work addresses the lack of interpretability and controllability in expert specialization mechanisms within sparse Mixture-of-Experts (MoE) models. The authors propose a low-dimensional geometric routing method based on cosine similarity, which endows each rank-1 expert with inherent mono-semanticity. For the first time, expert-level specialization is established as a fundamental unit of interpretability, exhibiting structural mono-semanticity, causal verifiability, and zero-overhead inference controllability. Experiments demonstrate that 15% of experts can be explicitly mapped to 10 distinct semantic categories. Furthermore, causal interventions enable significant modulation of target class output probabilities—e.g., increasing temporal category likelihood by 321% or decreasing geographic category likelihood by 23%—with effects that accumulate across multiple layers.

Technology Category

Application Category

📝 Abstract
Sparse Mixture-of-Experts (MoE) models scale parameters while fixing active computation per token, but the specialization of individual experts remains opaque. In a companion paper we showed that routing topology is quality-neutral: five structurally different configurations converge to statistically equivalent language modeling quality. Here we show that expert identity is nonetheless causally meaningful: individual rank-1 experts are monosemantic by construction, and cosine-similarity routing in a low-dimensional metric space makes their specialization directly inspectable. We present four lines of evidence. First, projecting expert output vectors through the unembedding matrix yields a Semantic Dictionary: 15% of experts are monosemantic specialists spanning 10 categories (temporal, geographic, cardinal, discourse, emotional, financial, military, scientific). Second, routing exhibits a frequency-to-syntax gradient: early layers separate tokens by word frequency, deeper layers by syntactic class (Zipf-confound controls, all $p < 0.001$). Third, causal interventions confirm these labels: steering toward a temporal expert's centroid increases P(temporal) by +321% (median across 44 prompts); suppressing a geographic expert drops P(geographic) by -23%; rewriting an expert's output vector halves target-category probability, and effects compose additively across layers. Fourth, the interventions are not unique to cosine routing: linear routers support comparable steering, but only cosine routing provides geometric transparency -- expert specialization is readable directly from the centroid matrix. MoE expert-level specialization is a first-class interpretability primitive: architecturally monosemantic, causally validated, and controllable at inference with zero overhead.
Problem

Research questions and friction points this paper is trying to address.

Mixture of Experts
expert specialization
model interpretability
geometric routing
monosemantic experts
Innovation

Methods, ideas, or system contributions that make the work stand out.

Geometric Routing
Mixture of Experts
Monosemantic Experts
Causal Interpretability
Cosine Similarity
🔎 Similar Papers
No similar papers found.
I
Ivan Ternovtsii
Department of Software Systems, Uzhhorod National University
Y
Yurii Bilak
Department of Software Systems, Uzhhorod National University