Invariant Transformation and Resampling based Epistemic-Uncertainty Reduction
This work offers a strategy to improve inference accuracy for AI models by reducing epistemic uncertainty, potentially benefiting practitioners seeking to balance model size and performance.
The paper addresses inference errors in AI models caused by epistemic uncertainty, observing that errors from invariant transformations of an input show partial independences. They propose a resampling-based inferencing method that aggregates outputs from multiple transformed input versions to achieve more accurate results.
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.