arXiv research

Regularization leads to balancedness in deep linear networks.

This paper optimizes revenue and resource balance in network revenue management.

Characterizes metrics on Lie groups, proving non-simultaneous existence of balanced and pluriclosed metrics.

The paper proposes a neural network method to estimate treatment effects by balancing treated and control distributions.

MCNet improves uncertainty calibration in online advertising by modeling complex relations and balancing performance.

Flashback Learning balances model stability and plasticity in continual learning.

Study on $p$-Kähler structures on fibrations and Lie groups.

We address the problem of {\it adaptivity} in the framework of reproducing kernel Hilbert space (RKHS) regression. More precisely, we analyze estimators arising from a linear regularization scheme $g_\lam$. In practical applications, an important task is to choose the regularization parameter $\lam$ appropriately, i.e.…

TempBalance improves neural network training by balancing learning rates across layers.

Dynamic information balancing reduces catastrophic forgetting in modular neural networks.

This research improves binary classification by balancing overfitting and generalization with a novel Bayesian approach.

Classifies complex structures on SL(2,C), finding new non-regular ones.

Latent variable models are a fundamental modeling tool in machine learning applications, but they present significant computational and analytical challenges. The popular EM algorithm and its variants, is a much used algorithmic tool; yet our rigorous understanding of its performance is highly incomplete. Recently, wor…

We study the implicit regularization imposed by gradient descent for learning multi-layer homogeneous functions including feed-forward fully connected and convolutional deep neural networks with linear, ReLU or Leaky ReLU activation. We rigorously prove that gradient flow (i.e. gradient descent with infinitesimal step …

The study constructs balanced and rigid curves on specific types of hypersurfaces and complete intersections.

AER dynamically adjusts entropy regularization for better LLM reinforcement learning.

This paper balances bias and variance in adaptive importance sampling using mirror descent.

Kernel balancing weights are generalized as KRRR, providing better confidence intervals for treatment effects.

State-of-the-art subspace clustering methods are based on expressing each data point as a linear combination of other data points while regularizing the matrix of coefficients with $\ell_1$, $\ell_2$ or nuclear norms. $\ell_1$ regularization is guaranteed to give a subspace-preserving affinity (i.e., there are no conne…

Paper introduces MinDiff framework for balancing classifier performance and fairness.

Large learning rates lead to various implicit biases in nonconvex optimization.

Cross-regularization adapts model complexity during training.

Large neural networks learn low-dimensional representations that balance complexity and regularity.

SVM used for estimating treatment effects without confounding.

TempBalance boosts model performance with low data.

Unified model for signed networks separates balance and anomaly effects.

The thesis tackles overconfident approximations in simulation-based inference.

We prove that the balanced Chekhov-Fock algebra of a punctured triangulated surface is isomorphic to a skein algebra which is a deformation of the algebra of regular functions of some abelian character variety. We first deduce from this observation a classification of the irreducible representations of the balanced Che…

The paper shows regularization can't always find all optimal solutions in constrained ML.

sgboost reduces variable selection bias in boosting with balanced group selection.

Regularizes sparse Gaussian processes for better model performance.

Noise balance theory explains SGD's behavior in neural networks.

QMME balances cost and speed in convex optimization.

Tensor decomposition methods allow us to learn the parameters of latent variable models through decomposition of low-order moments of data. A significant limitation of these algorithms is that there exists no general method to regularize them, and in the past regularization has mostly been performed using bespoke modif…

Novel characterization of augmented balancing weights combining outcome and weighting models.

Self-balancing sampler improves sampling efficiency and unpredictability.

Proves regularity of geodesic equation on Hermitian manifolds.

Hidden Markov Neural Networks balance adaptation and forgetting in time-series data.

Jigsaw-VAE tackles feature imbalance in VAE latent variables, improving generalization across environments.

Although deep learning has produced dazzling successes for applications of image, speech, and video processing in the past few years, most trainings are with suboptimal hyper-parameters, requiring unnecessarily long training times. Setting the hyper-parameters remains a black art that requires years of experience to ac…

HEBAE improves VAEs by adaptively balancing reconstruction and regularization.

Study shows how to balance memory and learning efficiency in continual learning.

Gradient descent solves asymmetric low-rank matrix sensing without balancing.

New method uses random convex polytopes to measure representation quality.

A RL framework selects features to balance bias and accuracy dynamically.

Proposes a method to adapt models in nonstationary environments using ℓ1 regularization.

This paper is concerned with the factorization form of the rank regularized loss minimization problem. To cater for the scenario in which only a coarse estimation is available for the rank of the true matrix, an $\ell_{2,0}$-norm regularized term is added to the factored loss function to reduce the rank adaptively; and…

Gradient descent on ReLU networks with square loss implicitly favors balanced weights.