aiqbench analysis
This arXiv preprint proposes a probabilistic framework that splits a neural classifier or regressor into a predictable component and a per-input reliability score, aiming to address the black-box criticism. The approach combines EM, SGD, and maximum likelihood, and its claims rest on synthetic and real-world experiments, so practitioners should check how the reliability score performs against baselines before relying on it. It is a research-stage method, not a deployed product.
Decomposing A Neural Learner
arXiv:2610.14388v1 Announce Type: cross Abstract: Neural learners for classification and regression are often criticized as black boxes. We propose a probabilistic decomposition of neural learners that accounts for heterogeneous predictability across the feature space. A training procedure combining the EM algorithm, stochastic gradient descent, and maximum likelihood estimation is developed to jointly learn the predictable component and an input-dependent reliability score. We show that the optimal decomposed learner recovers the response distribution within the predictable component, while the reliability score characterizes the predictability of individual observations. We further extend the framework to incorporate localized feature transformations. Experiments on synthetic and real-world datasets demonstrate the effectiveness of the proposed framework.
Abstract:Neural learners for classification and regression are often criticized as black boxes. We propose a probabilistic decomposition of neural learners that accounts for heterogeneous predictability across the feature space. A training procedure combining the EM algorithm, stochastic gradient descent, and maximum likelihood estimation is developed to jointly learn the predictable component and an input-dependent reliability score. We show that the optimal decomposed learner recovers the response distribution within the predictable component, while the reliability score characterizes the predictability of individual observations. We further extend the framework to incorporate localized feature transformations. Experiments on synthetic and real-world datasets demonstrate the effectiveness of the proposed framework.
Machine Learning (stat.ML); Machine Learning (cs.LG)
arXiv:2610.14388 [stat.ML]
(or arXiv:2610.14388v1 [stat.ML] for this version)
https://doi.org/10.48550/arXiv.2610.14388 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 201 words
Related coverage
What's Continual Learning, and Why Might We Expect To See It In Advanced LLM Agents?
LessWrong · Jun 12, 2026
NPBoost: Neural Processes with Gradient-Boosted Fixed Effects
arXiv · Sep 24, 2026
Generalization and infinite width
LessWrong · Aug 1, 2026
MetaLearnNCA: Few-Shot Offline Meta-Learning via Interacting Neural Cellular Automata
arXiv · Oct 7, 2026