Inductive Orientation-enabled Model Characterization

As models grow more complex, interpretable black-box characterization techniques are increasingly relevant. Based on the algorithmic search framework, we present estimation methods for model-theoretic quantities, such as algorithm flexibility, sensitivity to data, and ability to specialize. We compute these quantities across a wide variety of classification algorithms, observing trends matching known heuristics and theoretical properties. We further utilize these metrics to compare algorithms of different architectures and hyperparameter configurations. These findings validate uses for model evaluation, comparison, and hyperparameter tuning.

This project resulted in two papers: one establishing the use of inductive orientation-enabled metrics in a small range of individual classical models, and another paper performing a much more rigorous analysis on a greater range of models that includes meta-analyses and comparisons between models. The subsequent paper also demonstrates that, in practice, the bias-expressivity trade-off follows a more strict linear relationship.

[Paper #1] [Presentation slides]: Pang-Naylor, K., Chen, E., Montañez, G. (2025). Model Characterization with Inductive Orientation Vectors. \textit{International Conference on Agents and Artificial Intelligence (ICAART 2025)}. doi:10.5220/0013304400003890.

[Paper #2] (publish date: Dec. 5th, 2025): Pang-Naylor, K., Chen, E., Montañez, G. (in press). Analyzing and Comparing Machine Learning Models via Inductive Orientation. In International Conference on Agents and Artificial Intelligence. Springer International Publishing.