On-device research index

arXiv research

A locally-built, LLM-digested index of recent arXiv papers in quant finance, geometry/topology, and statistical ML — keyword search served straight from SQLite on this machine.

168,657 papers · 148 categories

Trend · papers per month

count % of corpus

20 results for “early-time”

The Surprising Simplicity of the Early-Time Learning Dynamics of Neural Networkscs.LG

Early neural networks can be simplified to linear models, revealing surprising simplicity.

problem Complexity of neural network learning dynamics.

method Formal proof and empirical verification of early-time learning dynamics of neural networks.

result Early learning dynamics of neural networks can be approximated by simple linear models.

neural networks learning dynamics early-time simplification linear models

2020-06-25abs ↗pdf ↗

Early Time Classification with Accumulated Accuracy Gap Controlcs.LG

Paper introduces a method to control early classification accuracy gaps.

problem Maintaining accuracy in early classification without full input processing.

method Statistical framework for a calibrated stopping rule.

result Reduces up to 94% of timesteps while controlling accuracy gaps.

early classification accuracy gap sequential classifier stopping rule Learn-then-Test

2024-02-01abs ↗pdf ↗

Early Time-Series Classification Algorithms: An Empirical Comparisoncs.LG

Evaluates six ETSC algorithms on various datasets.

problem Comparing ETSC algorithms for different data characteristics.

method Empirical comparison of six ETSC algorithms on public and new datasets.

result Provides insights into algorithm performance on real-life applications.

ETSC algorithms time-series classification data characteristics benchmarking life sciences

2022-03-03abs ↗pdf ↗

Deep Learning Framework for History Matching CO2 Storage with 4D Seismic and Monitoring Well Datacs.LG

Deep learning calibrates CO2 storage formations from seismic and well data.

problem Uncertainty in CO2 storage formation properties.

method Two deep learning models for well and seismic data, integrated into MCMC history matching.

result Significant uncertainty reduction in key parameters and accurate CO2 plume predictions.

carbon storage history matching deep learning 4D seismic well data

2024-08-02abs ↗pdf ↗

Exact Hurst exponent and crossover behavior in a limit order market modelcond-mat.stat-mech

An exclusion particle model is considered as a highly simplified model of a limit order market. Its price behavior reproduces the well known crossover from over-diffusion (Hurst exponent H>1/2) to diffusion (H=1/2) when the time horizon is increased, provided that orders are allowed to be canceled. For early times a ma…

2002-06-24abs ↗pdf ↗

A Tale of Two (and More) Altruistsphysics.soc-ph

Model compares altruism and individualism in wealth dynamics.

problem Comparing altruism and individualism in wealth dynamics.

method Minimalist dynamical model of wealth evolution and sharing among N agents.

result Altruism leads to more global median wealth at early times but individualists accumulate most wealth in the long run.

altruism individualism wealth dynamics dynamical model wealth sharing

2021-09-09abs ↗pdf ↗

Catapult Dynamics and Phase Transitions in Quadratic Netscs.LG

Catapult phase in neural nets shows exponential loss growth before quick decrease.

problem Understanding phase transitions in neural networks during training.

method Analyzing weight norm and loss behavior for super-critical learning rates.

result Proven existence of catapult phase in quadratic models and two-layer nets.

neural networks phase transitions learning rates weight norm ReLU activation

2023-01-18abs ↗pdf ↗

Semi-Supervised Risk Control via Prediction-Powered Inferencecs.LG

Improves risk control in predictions using semi-supervised calibration.

problem Noisy hyper-parameter tuning from limited labeled data.

method Semi-supervised calibration using unlabeled data to tune hyper-parameters rigorously.

result Improves prediction accuracy without sacrificing statistical validity.

risk control semi-supervised learning calibration hyper-parameter tuning machine learning

2024-12-15abs ↗pdf ↗

Predicting the Age of Astronomical Transients from Real-Time Multivariate Time Seriesastro-ph.IM

Predicts the age of astronomical transients from real-time data.

problem Improving understanding of transients and their progenitor systems.

method Bayesian probabilistic recurrent neural network.

result Accurately predicts the age of transients with robust uncertainties.

astronomical transients time series analysis Bayesian neural networks real-time prediction stellar explosions

2023-11-28abs ↗pdf ↗

Universal Sharpness Dynamics in Neural Network Training: Fixed Point Analysis, Edge of Stability, and Route to Chaoscs.LG

The study analyzes sharpness dynamics in neural networks, revealing mechanisms and conditions.

problem Understanding sharpness in neural network training.

method Fixed point analysis and edge of stability analysis in a simplified 2-layer linear network.

result Reveals mechanisms behind sharpness trends, conditions for edge of stability, and a period-doubling route to chaos.

sharpness dynamics neural networks fixed points edge of stability period-doubling

2023-11-03abs ↗pdf ↗

An Improved Long-Time Bishop-Gromov Theorem Using Shearmath.DG

Enhances the Bishop-Gromov theorem for curved spaces, especially at late times.

problem The Bishop-Gromov theorem's volume growth upperbound is often too loose, especially at late times.

method Identified and quantified the effect of shear, using higher curvature invariants to improve the upperbound.

result Tighter upper bounds on late-time growth rates of geodesic balls in homogeneous spaces with non-positive sectional curvature.

Bishop-Gromov theorem geodesic balls curvature invariants shear homogeneous spaces

2023-01-18abs ↗pdf ↗

Approaches and Applications of Early Classification of Time Series: A Reviewcs.LG

This paper reviews early time series classification methods.

problem Minimizing class prediction delay in time-sensitive applications.

method Divided into four categories: prefix based, shapelet based, model based, and miscellaneous approaches.

result Demonstrates reasonable performance in various applications.

time series classification healthcare finance human activity

2020-05-06abs ↗pdf ↗

Endogeneous Versus Exogeneous Shocks in Systems with Memorycond-mat.stat-mech

Systems with long-range persistence and memory are shown to exhibit different precursory as well as recovery patterns in response to shocks of exogeneous versus endogeneous origins. By endogeneous, we envision either fluctuations resulting from an underlying chaotic dynamics or from a stochastic forcing origin which ma…

2002-06-05abs ↗pdf ↗

Why Diffusion Models Don't Memorize: The Role of Implicit Dynamical Regularization in Trainingcs.LG

Diffusion models generalize well until a threshold is reached, preventing memorization.

problem Understanding why diffusion models don't memorize training data.

method Investigation of training dynamics and two timescales:

τ_\mathrm{gen}

and

τ_\mathrm{mem}

result The threshold

τ_\mathrm{mem}

increases linearly with training set size

n

, preventing memorization.

diffusion models generalization memorization training dynamics timescales

2025-05-23abs ↗pdf ↗

Neural Stein critics with staged

L^2

-regularizationstat.ML

The paper introduces a novel method for training neural network Stein critics with staged $L^2$ -regularization.

problem Learning to differentiate model distributions from observed data in high-dimensional settings.

method Developed a novel staging procedure for

L^2

regularization over training time, leveraging the advantages of highly-regularized training at early times.

result Theoretical guarantees and empirical validation show that the method improves the approximation of the training dynamic by the kernel optimization, leading to faster convergence and better performance.

Stein discrepancy Neural network Regularization Kernel optimization High-dimensional data

2022-07-07abs ↗pdf ↗

Feature learning as alignment: a structural property of gradient descent in non-linear neural networksstat.ML

Gradient descent aligns neural feature matrices with pre-activation tangent features.

problem Understanding neural feature learning mechanisms.

method Analytical proof of alignment between weight matrices and pre-activation tangent features.

result Derivative alignment occurs almost surely in high-dimensional settings.

neural networks feature learning gradient descent alignment singular vectors

2024-02-07abs ↗pdf ↗

TEASER: Early and Accurate Time Series Classificationcs.LG

Early time series classification (eTSC) is the problem of classifying a time series after as few measurements as possible with the highest possible accuracy. The most critical issue of any eTSC method is to decide when enough data of a time series has been seen to take a decision: Waiting for more data points usually m…

2019-08-09abs ↗pdf ↗

Closed geodesics in short intervals for random hyperbolic surfacesmath.GT

Study geodesics on random hyperbolic surfaces, finding variance similar to prime number theory.

problem Distribution of closed geodesics on random hyperbolic surfaces.

method Investigate random variable counting geodesics with norms in short intervals, comparing to prime number theory.

result Establishes variance of geodesic counting function is asymptotic to $2H \log X$.

closed geodesics random hyperbolic surfaces variance Chebyshev function Riemann zeta function

2026-05-21abs ↗pdf ↗

Robust Layerwise Scaling Rules by Proper Weight Decay Tuningcs.LG

We introduce a new weight-decay scaling rule to maintain sublayer gains across different widths in modern scale-invariant architectures.

problem In modern scale-invariant architectures, training quickly enters a steady state where normalization layers create backward scale sensitivity, degrading learning-rate transfer.

method We introduce a weight-decay scaling rule for AdamW that preserves sublayer gain across widths by equalizing the effective learning rate.

result Our empirical weight-decay scaling rule

λ_2\propto \sqrt{d}

approximately keeps sublayer gains width invariant, enabling zero-shot transfer of learning rate and weight decay.

weight decay sublayer gain learning rate scale-invariant architectures AdamW

2025-10-17abs ↗pdf ↗

The Case Against Smooth Null Infinity II: A Logarithmically Modified Price's Lawgr-qc

Logarithmic corrections to Price's law near black hole event horizon.

problem Failure of smooth null infinity in black hole spacetimes.

method Analyzing linear wave equation on Schwarzschild background with specific initial conditions.

result Leading-order asymptotics of solutions near future null infinity and event horizon are logarithmically modified.

black hole Price's law Schwarzschild wave equation null infinity

2021-05-17abs ↗pdf ↗