
Deep neural networks (DNNs) may develop abstract internal representations, termed "symbols," which can be extracted and used to understand, improve, and safeguard DNN decision-making.


coremsg

searching-for-internal-symbols-underlying-deep-learning-decisions


Searching for Internal Symbols Underlying Deep Learning Decisions


title_rewrite


This research paper proposes a novel framework and method for disambiguating multiple class label predictions in deep learning image recognition, determining whether predictions stem from distinct entities or a single entity misidentified, and provides verifiable counterfactual proofs for increased confidence in model interpretations.


disambiguating-multiple-class-labels-in-deep-learning-image-recognition-a-counterfactual-proof-framework


Disambiguating Multiple Class Labels in Deep Learning Image Recognition: A Counterfactual Proof Framework
