New arXiv paper proposes training neural networks to be locally adaptive
A paper on arXiv (2611.22198) argues that deep neural networks fitted via global empirical risk minimization commit to a single model across the covariate space, and proposes training networks to be locally adaptive in regression problems.
Read the full story at arXiv stat.ML ↗
Timeline · 1 report
- 2026-11-18 05:00 · arXiv stat.ML
Training Neural Networks to be Locally Adaptive