AINeutralarXiv – CS AI · Mar 47/102
🧠Researchers have derived tight bounds on covering numbers for deep ReLU neural networks, providing fundamental insights into network capacity and approximation capabilities. The work removes a log^6(n) factor from the best known sample complexity rate for estimating Lipschitz functions via deep networks, establishing optimality in nonparametric regression.
AINeutralarXiv – CS AI · May 116/10
🧠Researchers introduce DTSemNet, a novel neural network representation of oblique decision trees that enables approximation-free gradient-based training for both classification and regression tasks. The approach eliminates reliance on softening or quantized gradients, achieving superior performance on benchmark datasets and expanding decision tree applicability to reinforcement learning environments.
AINeutralarXiv – CS AI · Mar 55/10
🧠Researchers introduce zono-conformal prediction, a new uncertainty quantification method for machine learning that uses zonotope-based prediction sets instead of traditional intervals. The approach is more computationally efficient and less conservative than existing conformal prediction methods while maintaining statistical coverage guarantees for both regression and classification tasks.
AINeutralarXiv – CS AI · Mar 45/102
🧠Researchers developed a method to extract numerical prediction distributions from Large Language Models without costly autoregressive sampling by training probes on internal representations. The approach can predict statistical functionals like mean and quantiles directly from LLM embeddings, potentially offering a more efficient alternative for uncertainty-aware numerical predictions.
AINeutralarXiv – CS AI · Mar 264/10
🧠Researchers have extended Neural Collapse theory to regression problems, discovering that Deep Neural Regression Collapse (NRC) occurs across multiple layers in neural networks, not just the final layer. The study reveals that collapsed layers learn structured representations where features align with target dimensions and covariance, providing insights into the simple structures that deep networks learn for regression tasks.
AINeutralarXiv – CS AI · Feb 274/107
🧠Researchers developed smooth-basis regression models including anisotropic RBF networks and Chebyshev polynomial regressors that compete with tree ensembles in tabular regression tasks. Testing across 55 datasets showed these models achieve similar accuracy to tree ensembles while offering better generalization properties and gradual prediction surfaces suitable for optimization applications.
AINeutralarXiv – CS AI · Feb 274/106
🧠Researchers developed a new mathematical technique called 'anchoring' to control model disagreement between machine learning models trained independently. The method provides bounds for reducing disagreement to zero across four common ML algorithms including stacked aggregation, gradient boosting, neural networks, and regression trees.
AIBullisharXiv – CS AI · Mar 34/106
🧠Researchers developed a unified machine learning framework that predicts both pass/fail outcomes and continuous grades for secondary school students with up to 96% accuracy. The study of 4424 students demonstrates how AI can enable early identification of at-risk students and optimize educational resource allocation through data-driven predictions.
