The paper investigates how memorizing long‑tail data can aid generalization when combined with a model’s ability to compose features. It demonstrates theoretically that, in a linear setting, memorization plus composition enables correct predictions on rare test examples that involve unseen combinations of long‑tail features. Experiments with neural networks on simple datasets confirm that this effect extends beyond linear models and that a model’s compositional capability depends on its architecture.
arXiv:2607. 11541v1 Announce Type: new Abstract: We introduce a straightforward yet effective method to empirically study memorization in deep neural networks for classification tasks.
arXiv:2610.08577v1 Announce Type: cross Abstract: While unlearning seeks to negate undesired capabilities acquired through learning, little research has examined how the way models learn shapes their...
While unlearning seeks to negate undesired capabilities acquired through learning, little research has examined how the way models learn shapes their subsequent unlearning. In this paper, we investiga...
The paper introduces a pointwise generalization theory for fully connected deep neural networks, using a pointwise Riemannian Dimension derived from eigenvalues of learned feature representations across layers. This framework provides hypothesis-dependent, representation-aware generalization bounds that are significantly tighter than traditional size- or norm-based approaches, both theoretically and experimentally. The authors analytically identify structural properties that explain deep networks’ tractability and empirically show that the pointwise Riemannian Dimension captures feature compression, over‑parameterization effects, and optimizer bias.
arXiv:2602.08302v3 Announce Type: replace-cross Abstract: Delayed generalization, where training accuracy becomes nearly perfect substantially before test accuracy, has often been studied in the cont...
Over the past decade, deep neural networks (DNNs) have achieved remarkable success on complex machine-learning tasks, yet the theoretical foundations of their performance remain incomplete. From a statistical viewpoint, a natural question is: can DNNs attain feature-learning and prediction consistency comparable to that of classical models?
The paper studies memorization and generalization in conditional diffusion models using a random-feature conditional score model in a high-dimensional setting. It derives asymptotic expressions for training and test losses, showing that in overparameterized regimes wider models better predict the condition-dependent mean but also exhibit reduced within-condition variance, a behavior termed "malign generalization." The analysis also reveals that more informative conditions cause training samples to be memorized at smaller widths, with experimental validation on U‑Net architectures.
The paper introduces TPR-Attention, an attention mechanism that operates over tensor‑product representations to embed structured inductive bias into deep learning models. Experiments on compositional tasks demonstrate that TPR‑Attention outperforms existing architectural components in achieving combinatorial generalization. The results suggest that incorporating explicit compositional structure into neural attention can improve systematic generalization.
arXiv:2505.09716v4 Announce Type: replace Abstract: Out-of-distribution (OOD) generalisation through composition requires a system to discover invariant properties from input-output associations and...
arXiv:2603.25579v2 Announce Type: replace-cross Abstract: A key capability of modern neural networks is their capacity to simultaneously learn underlying rules and memorize specific facts or exceptio...
arXiv:2503. 07325v2 Announce Type: replace Abstract: Understanding and certifying the behavior of modern deep neural networks remains a fundamental challenge in reliable machine learning.
The paper introduces Neural Expansion, a geometric framework that explains how deep neural networks generalize by expanding directionally-selective ray coverage in input space. It defines Generalization Intervals (GIs) to measure stable prediction regions around training samples and shows through experiments that training progressively enlarges this coverage, especially along directions with low Jacobian singular values. This expansion accounts for correct test predictions, adversarial robustness, and handling of mislabeled or out-of-distribution data, offering a mechanistic complement to existing global theories.