Invariant Transformation and Resampling based Epistemic-Uncertainty Reduction

2026-02-26

AI Summary

This work addresses the susceptibility of current AI models to erroneous predictions caused by epistemic uncertainty during inference. The authors propose a resampling-based inference strategy that requires no modification to the model architecture: by applying input-invariant transformations to generate multiple samples, they exploit the partial independence of epistemic uncertainty across these transformed inputs and aggregate the resulting outputs. This approach effectively mitigates reasoning errors stemming from epistemic uncertainty, yielding significant improvements in prediction accuracy across multiple tasks. Furthermore, it offers a novel perspective on the trade-off between model scale and performance, demonstrating that enhanced robustness can be achieved through inference-time strategies rather than solely relying on larger models.

Technology Category

Machine Learning: Calibration & Uncertainty QuantificationReasoning under Uncertainty: Other Foundations of Reasoning under UncertaintyPhilosophy and Ethics of AI: AI & Epistemology

Application Category

Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMsGraph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphsUser Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems

Abstract

An artificial intelligence (AI) model can be viewed as a function that maps inputs to outputs in high-dimensional spaces. Once designed and well trained, the AI model is applied for inference. However, even optimized AI models can produce inference errors due to aleatoric and epistemic uncertainties. Interestingly, we observed that when inferring multiple samples based on invariant transformations of an input, inference errors can show partial independences due to epistemic uncertainty. Leveraging this insight, we propose a"resampling"based inferencing that applies to a trained AI model with multiple transformed versions of an input, and aggregates inference outputs to a more accurate result. This approach has the potential to improve inference accuracy and offers a strategy for balancing model size and performance.

Problem

• epistemic uncertainty
• AI inference
• inference errors
• uncertainty reduction

Innovation

• invariant transformation
• resampling
• epistemic uncertainty
• inference accuracy
• uncertainty reduction