Abstract:As Artificial Intelligence (AI) models are increasingly integrated into critical systems, the need for a robust framework to establish the trustworthiness of AI is increasingly paramount. While collaborative efforts have established conceptual foundations for such a framework, there remains a significant gap in developing concrete, technically robust methods for assessing AI model quality and performance. A critical drawback in the traditional methods for assessing the validity and generalizability of models is their dependence on internal developer datasets, rendering it challenging to independently assess and verify their performance claims. This paper introduces a novel approach for assessing a newly trained model's performance based on another known model by calculating correlation between neural networks. The proposed method evaluates correlations by determining if, for each neuron in one network, there exists a neuron in the other network that produces similar output. This approach has implications for memory efficiency, allowing for the use of smaller networks when high correlation exists between networks of different sizes. Additionally, the method provides insights into robustness, suggesting that if two highly correlated networks are compared and one demonstrates robustness when operating in production environments, the other is likely to exhibit similar robustness. This contribution advances the technical toolkit for responsible AI, supporting more comprehensive and nuanced evaluations of AI models to ensure their safe and effective deployment. Code is available at <a class="link-external link-https" href="https://github.com/aheldis/Cross-model-correlation.git" rel="external noopener nofollow">this https URL</a>.

Harmony in Diversity: Merging Neural Networks with Canonical Correlation Analysis

Non-Uniform Parameter-Wise Model Merging

Merging or ensembling: integrative analysis in multiple neuroimaging studies

The Theory of Canonical Correlation Analysis and Its Application to Feature Fusion

Generalized canonical correlation analysis for disparate data fusion

Multi-modality fusion using canonical correlation analysis methods: Application in breast cancer survival prediction from histology and genomics

$C^2M^3$: Cycle-Consistent Multi-Model Merging

SoK: On Finding Common Ground in Loss Landscapes Using Deep Model Merging Techniques

Deep Neural Network Fusion via Graph Matching with Applications to Model Ensemble and Federated Learning.

A Biologically Plausible Neural Network for Multichannel Canonical Correlation Analysis

Training-time Neuron Alignment through Permutation Subspace for Improving Linear Mode Connectivity and Model Fusion

Exploring Cross-model Neuronal Correlations in the Context of Predicting Model Performance and Generalizability

Channel Splitting-and-Merging Connection in Neural Networks

PLeaS -- Merging Models with Permutations and Least Squares

Canonical Correlation Guided Deep Neural Network

Orthogonal Canonical Correlation Analysis and Its Application in Feature Fusion

Towards Aligned Canonical Correlation Analysis: Preliminary Formulation and Proof-of-Concept Results

Using Correspondence Analysis to Combine Classifiers

A novel multiset integrated canonical correlation analysis framework and its application in feature fusion

Dynamically building diversified classifier pruning ensembles via canonical correlation analysis

Investigating Neuron Disturbing in Fusing Heterogeneous Neural Networks