Marchenko--Pastur
PulseAugur coverage of Marchenko--Pastur — every cluster mentioning Marchenko--Pastur across labs, papers, and developer communities, ranked by signal.
4 day(s) with sentiment data
-
Spectral outliers in Transformer attention reveal dominant learned structures
Researchers have applied Marchenko-Pastur random matrix theory to analyze pre-trained transformer attention weights, identifying spectral outliers that represent dominant learned structures. By zeroing these identified …
-
New spectral regularization method improves linear regression risk performance
Researchers have developed a new method called negative-shifted gradient descent for overparameterized linear regression. This technique aims to overcome the limitations of traditional negative-ridge regularization by a…
-
LLM sampling variation doesn't reveal model ignorance, study finds
A new research paper published on arXiv explores the limitations of stochastic sampling in large language models (LLMs). The study, titled "Stochastic Sampling is Epistemically Shallow: The Dimensionality Gap Between Te…
-
New De-floored Principal Component Regression Method Enhances Prediction Accuracy
Researchers have introduced De-floored Principal Component Regression (dPCR), a novel method designed to improve prediction accuracy in high-dimensional data. Unlike traditional Principal Component Regression (PCR), whi…
-
New 'Muse' optimizers explore representation geometry for LLMs
Researchers have introduced "Muse," a novel family of optimizers designed for large language models that explores the geometric properties of parameter representations. Unlike standard Muon-style optimizers, Muse's upda…
-
New method improves RBMs for out-of-distribution data rejection
Researchers have developed a new method to improve the performance of Restricted Boltzmann Machines (RBMs) when dealing with out-of-distribution (OOD) inputs. By training RBMs with auxiliary random binary images assigne…
-
New criterion predicts effectiveness of time-lagged spectral embeddings
Researchers have developed a new criterion to determine the applicability of training-free time-lagged spectral embeddings for multivariate time series. This criterion, based on stationarity and temporal coupling, predi…