Is Class Signal Clustered or Routed in Task-Induced Implicit Neural Representation Weight Spaces?

2026-05-08

AI Summary

This work investigates the origin of class-discriminative signals in task-conditioned implicit neural representation (INR) weights—specifically, whether such signals arise from geometric clustering in weight space or from a reader-dependent dynamic routing mechanism. Leveraging the SIREN architecture within a Meta Weight Transformer framework, the study employs token-flow diagnostics, bias-column interventions, and inner-loop optimization to demonstrate that class separability does not stem from intrinsic clustering of the weights themselves, but is instead constructed through the reader’s causal routing of low-dimensional bias columns. A routing-augmented strategy derived from this insight outperforms baseline approaches under specific conditions, confirming the dynamic nature of class-signal generation and revealing non-additive interaction effects among different intervention modalities.

Abstract

Implicit neural representations (INRs) encode images as neural-network weights, making image classification a problem of weight-space classifiability. A natural geometric hypothesis is that classifier feedback should make image-specific weights cluster by class in the shared-anchor coordinate. We test this hypothesis in the SIREN-based Meta Weight Transformer (MWT) regime, where end-to-end training meta-learns a shared initialization and inner-loop update schedule for fitting image-specific SIRENs. We find that this prediction fails. Exposed weight-space geometry and supervised clustering pressure do not reliably track trained-reader accuracy; clustering can even make local neighborhoods more class-consistent while making the trained reader worse. Crucially, the reader constructs rather than inherits class-aligned geometry: token-flow diagnostics show that class-aligned neighborhoods become strongly predictive of trained-reader accuracy only after late reader interactions, not in the input coordinate. We further identify the native SIREN bias column in the augmented weight token as a low-dimensional, sample-dependent causal readout route for the trained reader; targeted controls rule out generic scalar-column and marginal-distribution artifacts. The diagnosis motivates interventions that strengthen reader routing, add an explicit bias route, or use denser inner-loop fitting; under the lane-specific training conventions used here, route-directed variants often outperform the shared-anchor baseline but interact non-additively. Task-induced INR weights are classifiable not because they form raw geometric clusters, but because their class signal is routed through the reader.

Problem

• Implicit Neural Representations
• Weight Space
• Class Signal
• Clustering
• Routing

Innovation

• implicit neural representations
• weight-space geometry
• signal routing
• Meta Weight Transformer
• classifiability