Gradient descent on ReLU networks with square loss implicitly favors balanced weights.
problem Understanding implicit regularization in nonlinear neural networks with regression losses.
method Analyzing gradient descent dynamics on ReLU networks with square loss.
result It is impossible to characterize the implicit regularization of ReLU networks with square loss by any explicit function of model parameters.
Global implicit function theorem for Fréchet spaces, solving derivative loss problems.
problem Solving initial value problems with derivative loss in Fréchet spaces.
method Global implicit function theorems for Keller's Cc1-mappings in Fréchet spaces, applied through submersions and transversality. result Global existence and uniqueness of solutions to initial value problems with derivative loss.
This work extends implicit bias analysis to multiclass classification using a new loss framework.
problem The implicit bias of gradient descent on multiclass data without explicit regularization.
method Employing the PERM framework to introduce a multiclass extension of the exponential tail property.
result Extended implicit bias result to multiclass classification using a new loss framework.
New model leads to optimal test loss in sparse linear regression.
problem Sparse linear regression with low test loss despite interpolating training data.
method Developed a new parametrization of the model that combines benefits of ℓ1 and ℓ2 norms.
result Training via gradient descent leads to an interpolator with near-optimal test loss.
SGD with large learning rates can achieve better test accuracy than expected.
problem SGD with large learning rates often outperforms expected convergence bounds.
method Proved that SGD with small learning rates stays close to gradient flow path on modified loss.
result Explicitly adding an implicit regularizer to the loss improves test accuracy.
Unified framework approximates gradient descent's implicit bias in high dimensions.
problem Understanding gradient descent's behavior in overparameterized settings with convex losses.
method Unified framework for convex losses, including sensitivity analysis.
result Approximation of minimum-norm interpolation in high dimensions.
This paper analyzes implicit bias in Deep Linear Discriminant Analysis.
problem The implicit bias of Deep Linear Discriminant Analysis.
method Analyzing gradient flow on a L-layer diagonal linear network.
result Under balanced initialization, the network transforms additive updates into multiplicative updates, conserving the (2/L) quasi-norm.
New method for high-fidelity shape representations from raw data.
problem Creating accurate shape representations from raw data.
method A simple loss function encouraging neural network to vanish on input point cloud and have unit norm gradient.
result Our method produces high-fidelity, smooth, and natural zero level set surfaces.
Quantum models face barren plateaus, but specific losses can be trainable.
problem Barren plateaus and loss concentration in quantum generative models.
method Investigated explicit and implicit losses, and their interplay.
result Explicit losses lead to new barren plateaus, while implicit losses can be trainable.
The study analyzes implicit biases in neural networks using backward error analysis.
problem Analyzing implicit biases in multitask and continual learning settings.
method Backward error analysis to compute implicit training biases, deriving modified losses with three terms.
result The conflict term, measuring gradient alignment, is a new quantity in continual learning.
Gradient matching method estimates implicit regularization in complex deep learning systems.
problem Estimating implicit regularization in modern deep learning systems with complex modifications.
method Gradient matching methods to empirically estimate implicit regularization.
result Empirical estimation of implicit regularization in arbitrary networks, including dropout.
New algorithm expands FTRL framework with improved worst-case regret bounds.
problem Online learning with improved worst-case regret bounds.
method Generalized implicit Follow-The-Regularized-Leader (FTRL) algorithm.
result Unified framework for designing updates improving worst-case regret bounds.
Adam's hyperparameters implicitly regularize solutions, penalizing or impeding loss gradients' norms.
problem Implicit regularization in Adam's hyperparameters and training stage.
method Backward error analysis and ODE approximations to study Adam's behavior.
result Adam's implicit regularization depends on hyperparameters and training stage, involving different norms.
Study reveals how initialization scale affects training accuracy in linear networks.
problem Understanding implicit bias in linear classification models.
method Asymptotic analysis of gradient flow trajectories and training loss minimization.
result Implicit bias is more complex at reasonable initialization scales and training accuracies.
A new framework for structured prediction on non-vectorial spaces.
problem Structured prediction on non-vectorial output spaces.
method Defining a suitable geometry for implicit loss functions.
result Efficient algorithmic framework with sharp statistical analysis.
Gradient descent implicitly regularizes neural networks by penalizing large loss gradients.
problem How to optimize deep neural networks without explicit regularization.
method Backward error analysis to calculate implicit gradient regularization and demonstrate its effectiveness empirically.
result Implicit gradient regularization biases gradient descent toward flat minima, improving model robustness and test errors.
GD at EoS edge minimizes logistic loss without monotonic convergence.
problem Understanding GD's implicit bias at the edge of stability.
method Theoretical analysis of logistic regression with constant stepsize GD.
result GD with any constant stepsize minimizes logistic loss over long time scales.
The paper analyzes the implicit bias of SGD near loss manifold and provides new insights.
problem Understanding the implicit bias of SGD near loss manifolds in overparametrized models.
method Adapting ideas from Katzenberger (1991) to analyze SGD dynamics using a stochastic differential equation (SDE).
result SGD with label noise locally decreases the sharpness of loss, leading to a global analysis of implicit bias.
New method trains generative models without discriminators, improving stability and accuracy.
problem Training implicit generative models with adversarial discriminators leads to instability and mode-dropping.
method Invariant statistical loss function, avoiding discriminators.
result Successfully trains generative models for various complex distributions without mode-dropping.
This work uses PAC-Bayes for structured prediction with ILE, yielding insights and algorithms.
problem Structured prediction with interdependent outputs and implicit loss embeddings.
method PAC-Bayes perspective applied to ILE framework, deriving generalization bounds and learning algorithms.
result Two learning algorithms derived from PAC-Bayes bounds, analyzed and implemented.
Geometric framework explains and controls implicit bias in machine learning.
problem Understanding and controlling the selection of solutions in overparameterized models.
method Developed a theoretical and constructive framework based on geometric corrections induced by gradient noise and continuous symmetries of the loss.
result Computed the induced bias across various architectures and enabled inverse design to shape the bias.
AuxiLearn combines auxiliary tasks into a single loss function.
problem Improving neural network performance on a main task using auxiliary tasks.
method Implicit differentiation to learn a network that combines auxiliary tasks into a single coherent objective function.
result AuxiLearn consistently outperforms competing methods in various tasks and domains.
We examine gradient descent on unregularized logistic regression problems, with homogeneous linear predictors on linearly separable datasets. We show the predictor converges to the direction of the max-margin (hard margin SVM) solution. The result also generalizes to other monotone decreasing loss functions with an inf…
Proposes a semi-implicit back propagation method for neural networks.
problem Challenges in training neural networks, especially gradient vanishing and small step sizes.
method Proposes a semi-implicit back propagation method using error back propagation and proximal methods.
result The proposed method leads to better performance in terms of loss decreasing and training/validation accuracy compared to SGD and ProxBP.
JoVA combines two VAEs to learn user and item representations for better recommendation.
problem Collaborative filtering with implicit feedback.
method Joint Variational Autoencoders (JoVA) with a hinge-based pairwise loss function (JoVA-Hinge).
result JoVA-Hinge outperforms state-of-the-art methods in top-k recommendation.
Study shows deep linear networks can converge to flatter minima at large learning rates.
problem Understanding the implicit bias of deep linear networks at large learning rates.
method Characterization of deep linear networks for binary classification using logistic loss in the large learning rate regime.
result Gradient descent iterates converge to a flatter minimum in the catapult phase for certain data separation conditions.
This paper proves SGD converges to global minimum for over-parameterized ReLU networks.
problem Theoretical understanding of implicit neural networks is limited.
method Gradient flow analysis of ReLU activated implicit neural networks.
result Randomly initialized gradient descent converges to global minimum at a linear rate for square loss function in over-parameterized ReLU networks.
Recent works propose using the discriminator of a GAN to filter out unrealistic samples of the generator. We generalize these ideas by introducing the implicit Metropolis-Hastings algorithm. For any implicit probabilistic model and a target distribution represented by a set of samples, implicit Metropolis-Hastings oper…
The paper studies implicit regularization in over-parameterized models for high-dimensional data.
problem Understanding implicit regularization in over-parameterized models for high-dimensional data.
method The paper designs regularization-free algorithms for the high-dimensional single index model and provides theoretical guarantees for the induced implicit regularization phenomenon.
result The proposed methods achieve minimax optimal statistical rates of convergence and outperform classical methods with explicit regularization.
We reparametrize ReLU NNs as splines to understand their learning dynamics.
problem Understanding the learning dynamics and inductive bias of neural networks.
method Reparametrize ReLU NNs as continuous piecewise linear splines to study learning dynamics.
result Standard weight initializations yield very flat functions, leading to strength and type of implicit regularization.
We study the generalization properties of stochastic gradient methods for learning with convex loss functions and linearly parameterized functions. We show that, in the absence of penalizations or constraints, the stability and approximation properties of the algorithm can be controlled by tuning either the step-size o…
We study minimax convergence rates of nonparametric density estimation under a large class of loss functions called "adversarial losses", which, besides classical Lp losses, includes maximum mean discrepancy (MMD), Wasserstein distance, and total variation distance. These losses are closely related to the …
Stochastic gradient descent procedures have gained popularity for parameter estimation from large data sets. However, their statistical properties are not well understood, in theory. And in practice, avoiding numerical instability requires careful tuning of key parameters. Here, we introduce implicit stochastic gradien…
The paper explores how the depth of neural networks affects their ability to represent data accurately.
problem Understanding the implicit bias and rank of neural networks with large depth.
method Analyzing the convergence of representation cost to a notion of rank as network depth increases, and investigating conditions for recovering the true rank of data.
result There is a range of network depths where the true rank of data is recovered, and this affects the topology of class boundaries.
Novel approach finds implicit regularisation in two-player games using BEA.
problem Understanding implicit regularisation in two-player games.
method Using backward error analysis to construct continuous-time flows with gradient-eligible vector fields.
result Identifies new implicit regularisation effects in two-player games.
Gradient descent implicitly favors group sparsity in neural networks.
problem Understanding implicit regularization in neural networks for structured sparsity.
method Novel neural reparameterization for diagonally grouped linear networks.
result Gradient descent without explicit regularization biases towards group sparsity.
This paper shows that the implicit bias of gradient descent on linearly separable data is exactly characterized by the optimal solution of a dual optimization problem given by a smoothed margin, even for general losses. This is in contrast to prior results, which are often tailored to exponentially-tailed losses. For t…
New algorithms for efficient learning with partial information, reducing regret.
problem Online learning with partial observability and semi-bandit feedback.
method Implicit exploration strategy for near-optimal regret guarantees.
result First algorithms with near-optimal regret guarantees without knowing the observation system.
In this paper, we propose a novel ranking framework for collaborative filtering with the overall aim of learning user preferences over items by minimizing a pairwise ranking loss. We show the minimization problem involves dependent random variables and provide a theoretical analysis by proving the consistency of the em…
We consider networks, trained via stochastic gradient descent to minimize ℓ2 loss, with the training labels perturbed by independent noise at each iteration. We characterize the behavior of the training dynamics near any parameter vector that achieves zero training error, in terms of an implicit regularization te…
Study accelerates gradient methods in machine learning, revealing risk and stability connections.
problem Understanding the statistical risk of accelerated gradient methods in machine learning.
method Continuous-time analysis of Nesterov's accelerated gradient method and Polyak's heavy ball method for least squares regression.
result Connections between early stopping, stability, and curvature of loss function are revealed.
Dropout introduces both explicit and implicit regularization effects.
problem Understanding the full impact of dropout regularization.
method Disentangled explicit and implicit regularization effects through experiments and analytic simplifications.
result Explicit and implicit regularization effects of dropout are distinct and can be characterized analytically.
The paper analyzes optimal implicit bias in linear regression for over-parameterized models.
problem Finding the best generalization performance in over-parameterized linear regression.
method Asymptotic analysis of generalization performance for convex functions/potentials.
result Optimal implicit bias that achieves the best generalization error under certain conditions.
New study shows deep networks generalize well due to loss surface geometry.
problem Why deep networks generalize well despite many parameters.
method Analyzed local geometry of loss surface and its effect on SGD.
result SGD stays close to low-dimensional subspace, leading to better generalization bounds.
We improve generative models for heavy-tailed multivariate data using an invariant statistical loss.
problem Traditional generative models struggle with heavy-tailed and multivariate data, leading to unstable training and mode dropping.
method We extend the invariant statistical loss method to handle heavy-tailed and multivariate data using a Pareto-ISL generator trained with input noise from a generalised Pareto distribution.
result Pareto-ISL accurately models the tails of heavy-tailed distributions while capturing central characteristics.
Our paper characterizes how ReLU affects GD's implicit bias in high-dimensional neural networks.
problem Understanding the implicit bias of gradient descent on neural networks.
method Novel primal-dual analysis tracking predictions and coefficients.
result The implicit bias approximates the minimum-ℓ2-norm solution with high probability. Generalization error defines the discriminability and the representation power of a deep model. In this work, we claim that feature space design using deep compositional function plays a significant role in generalization along with explicit and implicit regularizations. Our claims are being established with several im…
Efficient implicit differentiation for Lasso hyperparameter optimization.
problem Difficult hyperparameter optimization for Lasso-type models.
method Implicit differentiation algorithm tailored for Lasso-type problems, avoiding matrix inversion and solving linear systems.
result Outperforms standard methods in optimizing error on held-out data or Stein Unbiased Risk Estimator (SURE).