MIND: Decoupling Model-Induced Label Noise via Latent Manifold Disentanglement

TL;DR

MIND introduces a theoretically grounded framework that decouples model-induced label noise into manageable components using latent manifold disentanglement, improving robustness in noisy annotation scenarios.

Contribution

The paper proposes a novel Latent Manifold Disentanglement approach to address systematic label noise, enabling effective noise correction without ground-truth anchors.

Findings

MIND outperforms state-of-the-art methods on complex benchmarks.

It effectively corrects zero-shot hallucinations in Vision-Language Models.

Demonstrates robustness across controlled and real-world datasets.

Abstract

The paradigm of learning from automatic annotations driven by pre-trained experts and Foundation Models dominates data-hungry applications. However, it introduces a critical challenge: model-induced label noise. Unlike stochastic noise in classical robust learning, this noise stems from annotator inductive biases, manifesting as systematic errors tightly coupled with local feature manifolds. Existing methods relying on global transition matrices underfit these structural patterns, while learning instance-specific matrices remains mathematically intractable. We propose Model-Induced Noise Decoupling (MIND), a theoretically grounded framework addressing this dilemma. We demonstrate that the high-dimensional noise manifold can be decoupled into tractable, subspace-dependent components via Latent Manifold Disentanglement. Specifically, our Latent Decoupling Estimator (LDE) dynamically projects samples into latent structural clusters with consistent error modes, facilitating noise identifiability without ground-truth anchor points. To rigorously evaluate robustness, we adopt a hierarchical protocol: moving from controlled noise on CIFAR-100 to a structural stress test on large-scale real-world 3D datasets (S3DIS, ScanNet), where error patterns explicitly couple with geometric manifolds. Empirically, MIND significantly outperforms state-of-the-art methods on these complex benchmarks and effectively corrects zero-shot hallucinations from Vision-Language Models (e.g., OpenSeg), highlighting its potential as a robust distillation framework for Foundation Models.