paper
How abundant are good interpolators?
paperactiveprovisional
how-abundant-are-good-interpolators--ac658894·1 events·first seen 12d agoAliases: How abundant are good interpolators?
More like this (12)
When Does Mixing Help? Analyzing Query Embedding Interpolation in Multilingual Dense RetrievalSecond-Order Path Kernel Interpolation Formulas in Machine LearningExtrapolative Weight AveragingProbabilistic Smoothing with Ratio-Monotone TransformsModel Merging / Weight InterpolationInflucoder: Distilling Decoders' Gradient Influence Rankings into an Encoder for Data AttributionIntegrated GradientsLanguage Models Compare Quantities Using Number-specific and Unit-specific HeuristicsBeyond Uniform Tokens: Adaptive Compression for Time Series Language ModelsFixed-Point Reasoners: Stable and Adaptive Deep Looped Transformersintra-frame token sparsificationOEIS Conjectures
Recent events (1)
Large deviation analysis shows most interpolating classifiers share the same generalization performance
A new arXiv preprint establishes a large deviation principle characterizing the generalization performance of interpolating linear classifiers in the overparameterized regime (n/d → α, small α). The key result is a concentration phenomenon: all but an exponentially small fraction of interpolators achieve approximately the same generalization error, determined by a unique rate-function maximizer. Empirically, gradient descent and a natural linear program both outperform this typical interpolator, providing theoretical grounding for benign overfitting in overparameterized models.