arXiv research

Distributed training of massive machine learning models, in particular deep neural networks, via Stochastic Gradient Descent (SGD) is becoming commonplace. Several families of communication-reduction methods, such as quantization, large-batch methods, and gradient sparsification, have been proposed. To date, gradient s…

A new energy-efficient pruning method for federated learning.

Pruning method removes less important features in linear models.

This work analyzes the statistical properties of adaptive gradient methods.

Analyze SGD with biased gradients, improving convergence rates and accuracy.

MTAdam optimizes multiple loss terms in neural models, balancing gradients dynamically.

Cosine similarity can force points to grow in magnitude, causing convergence issues.

A new framework explains why early pruning works well.

In this work, we study robust deep learning against abnormal training data from the perspective of example weighting built in empirical loss functions, i.e., gradient magnitude with respect to logits, an angle that is not thoroughly studied so far. Consequently, we have two key findings: (1) Mean Absolute Error (MAE) D…

The rise of deep learning in recent years has brought with it increasingly clever optimization methods to deal with complex, non-linear loss functions. These methods are often designed with convex optimization in mind, but have been shown to work well in practice even for the highly non-convex optimization associated w…

The ADAM optimizer is exceedingly popular in the deep learning community. Often it works very well, sometimes it doesn't. Why? We interpret ADAM as a combination of two aspects: for each weight, the update direction is determined by the sign of stochastic gradients, whereas the update magnitude is determined by an esti…

The complex wave representation (CWR) converts unsigned 2D distance transforms into their corresponding wave functions. Here, the distance transform S(X) appears as the phase of the wave function φ(X)---specifically, φ(X)=exp(iS(X)/τwhere τis a free parameter. In this work, we prove a novel result using the higher-orde…

Expressive quantum circuits are harder to train due to flatter cost landscapes.

Gradient descent performs well on weakly convex losses, offering generalization guarantees.

AGNES accelerates gradient descent with noisy gradients.

Optimization with noisy gradients has become ubiquitous in statistics and machine learning. Reparameterization gradients, or gradient estimates computed via the "reparameterization trick," represent a class of noisy gradients often used in Monte Carlo variational inference (MCVI). However, when these gradient estimator…

Improved DeepONets for PDE solution operators with adaptive re-weighting and new architecture.

Improved DP-SGD for variational inference reduces noise and variance.

RSO uses random weight perturbations to train deep networks without gradients.

Stochastic Gradient Descent introduces noise in training, affecting model decision boundaries.

Hadamard Wirtinger Flow recovers sparse signals from fewer measurements.

A new algorithm speeds up convex clustering.

Logit dynamics formula reveals self-regulation in softmax policy gradient methods.

We design a new algorithm for batch active learning with deep neural network models. Our algorithm, Batch Active learning by Diverse Gradient Embeddings (BADGE), samples groups of points that are disparate and high-magnitude when represented in a hallucinated gradient space, a strategy designed to incorporate both pred…

Study separates learning rate effects from adaptive gradient methods.

The paper provides convergence guarantees for multicalibration gradient boosting.

Policy gradient algorithms typically combine discounted future rewards with an estimated value function, to compute the direction and magnitude of parameter updates. However, for most Reinforcement Learning tasks, humans can provide additional insight to constrain the policy learning. We introduce a general method to i…

L-HNNs improve Bayesian inference by reducing gradient requirements and improving ESS.

We present a theoretical and empirical study of the gradient dynamics of overparameterized shallow ReLU networks with one-dimensional input, solving least-squares interpolation. We show that the gradient dynamics of such networks are determined by the gradient flow in a non-redundant parameterization of the network fun…

We analyze the variance of stochastic gradients along negative curvature directions in certain non-convex machine learning models and show that stochastic gradients exhibit a strong component along these directions. Furthermore, we show that - contrary to the case of isotropic noise - this variance is proportional to t…

AGF explains feature learning in neural networks through alternating steps.

DG separates successes and failures by gating updates with advantage and surprisal.

MGD improves ML training efficiency and accuracy.

A commonly used heuristic in non-convex optimization is Normalized Gradient Descent (NGD) - a variant of gradient descent in which only the direction of the gradient is taken into account and its magnitude ignored. We analyze this heuristic and show that with carefully chosen parameters and noise injection, this method…

Proposes a query-efficient blackbox attack method.

LASG improves communication efficiency in distributed learning.

We consider the task of low-multilinear-rank functional regression, i.e., learning a low-rank parametric representation of functions from scattered real-valued data. Our first contribution is the development and analysis of an efficient gradient computation that enables gradient-based optimization procedures, including…

GD converges faster to flatter minima than gradient flow in shallow networks.

This work addresses the instability in asynchronous data parallel optimization. It does so by introducing a novel distributed optimizer which is able to efficiently optimize a centralized model under communication constraints. The optimizer achieves this by pushing a normalized sequence of first-order gradients to a pa…

NS-GAN mode collapse due to sample weighting inversion, solved with MM-nsat.

Despite having various attractive qualities such as high prediction accuracy and the ability to quantify uncertainty and avoid over-fitting, Bayesian Matrix Factorization has not been widely adopted because of the prohibitive cost of inference. In this paper, we propose a scalable distributed Bayesian matrix factorizat…

Gradient sparsification enhances privacy-preserving machine learning models.

Gradient-based methods improve understanding of deep learning survival models.

Classically, the time complexity of a first-order method is estimated by its number of gradient computations. In this paper, we study a more refined complexity by taking into account the `lingering' of gradients: once a gradient is computed at $x_k$, the additional time to compute gradients at $x_{k+1},x_{k+2},\dots$ m…

The notion of the stationary equilibrium ensemble has played a central role in statistical mechanics. In machine learning as well, training serves as generalized equilibration that drives the probability distribution of model parameters toward stationarity. Here, we derive stationary fluctuation-dissipation relations t…

We show through theory and experiment that gradient-based explanations of a model quickly reveal the model itself. Our results speak to a tension between the desire to keep a proprietary model secret and the ability to offer model explanations. On the theoretical side, we give an algorithm that provably learns a two-la…

The gradient noise of SGD is considered to play a central role in the observed strong generalization abilities of deep learning. While past studies confirm that the magnitude and the covariance structure of gradient noise are critical for regularization, it remains unclear whether or not the class of noise distribution…

Algorithm selects public datasets for private machine learning.