Research map / Learning theory and algorithms / Sample complexity
Projected Neural Additive Models as Universal Approximators
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Establishes that projected neural additive models (NAMs augmented with a learnable linear transformation T capturing cross-variable interactions) are universal approximators.
Problem
Neural additive models are interpretable but lack expressivity because of their rigid additive structure, while MLPs are universal approximators but not transparent.
Approach
Augments NAMs with a learnable linear transformation T so that cross-variable interactions can be captured while retaining an interpretable structure, and provides a theoretical proof that the resulting PNAMs are universal approximators.
Key results
- Theoretical result establishing that PNAMs are universal approximators, addressing the expressivity gap of NAMs relative to MLPs.
Limitations
- Only the abstract was available; empirical validation details are not stated in the provided text.
Experimental setup
Baselines: neural additive models multi-layer perceptrons
Builds on: neural additive models multi-layer perceptron
Closest papers by meaning
- Improving Neural Additive Models with Bayesian Principles (ICML 2024 · similarity 0.86)
- Additive Models Explained: A Computational Complexity Approach (NeurIPS 2025 · similarity 0.86)
- Uncovering a Universal Abstract Algorithm for Modular Addition in Neural Networks (NeurIPS 2025 · similarity 0.85)
- OrthoReg: Orthogonal Regularization for Hybrid Symbolic-Neural Dynamical Systems (NeurIPS 2026 · similarity 0.84)
- Expressivity of Neural Networks with Random Weights and Learned Biases (ICLR 2025 · similarity 0.84)
- Explainable Neural Networks with Guarantee: A Sparse Estimation Approach (AAAI 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.