New shuffling methods improve convergence without Lipschitz smoothness.
problem Lack of convergence guarantees for shuffling methods under non-Lipschitz conditions.
method Revisit shuffling methods, prove convergence under general bounded variance condition.
result Matched current best-known convergence rates without Lipschitz smoothness.
Improved neural network convergence with causal Bayesian modeling in retail performance.
problem Improving neural network convergence in retail performance models.
method Causal Bayesian neural network implementation, removal of weakest SEM path, Flipout layers, Vadam optimizer.
result Neural network convergence improved with removal of the weakest SEM path.
SIBRE boosts reinforcement learning convergence by rewarding improvement over past performance.
problem Improving the rate of convergence in reinforcement learning.
method SIBRE is a reward shaping approach that rewards improvement over the agent's own past performance.
result SIBRE converges faster and more stably to the optimal policy compared to baseline RL algorithms.
This thesis investigates belief propagation's performance in graphical models with loops.
problem Belief propagation's performance and convergence guarantees in models with loops are uncertain.
method Investigates how model parameters affect belief propagation's performance, convergence, and approximation quality.
result Model parameters influence the number of fixed points, convergence properties, and approximation quality of belief propagation.
Value aggregation is a general framework for solving imitation learning problems. Based on the idea of data aggregation, it generates a policy sequence by iteratively interleaving policy optimization and evaluation in an online learning setting. While the existence of a good policy in the policy sequence can be guarant…
New HMC method uses asymmetrical momentum distributions and improves performance.
problem Rigorous convergence guarantees for HMC with Gaussian momentum distributions.
method New convergence analysis for HMC with general asymmetrical momentum distributions, proposing AD-HMC.
result AD-HMC exhibits geometric convergence in Wasserstein distance under certain conditions.
This paper improves inverse problem solving with weakly convex regularisers and proves convergence.
problem Improving solution methods for inverse problems.
method Generalised formulation of convergent regularisation using weakly convex regularisers, and proof of convergence for primal-dual hybrid gradient method.
result Proves convergence of primal-dual hybrid gradient method for variational problems and shows improved performance with IWCNNs.
Expands MLM by masking token positions, improving performance and convergence.
problem Improving language model performance and convergence.
method Masking token positions along with [MASK] tokens, using a fully connected classifier stage.
result Shows .3% improvement and 50% faster convergence for BERT Base with position masking.
Gradient boosting improved with lassoed trees achieves faster convergence.
problem Improving gradient boosting convergence in large nonparametric spaces.
method Lassoed gradient boosted trees with early stopping.
result Achieves faster than n−1/4 L2 convergence rate. The paper analyzes the training dynamics of a transformer for next-token prediction.
problem Understanding the non-asymptotic performance of transformers in next-token prediction.
method Characterizes training dataset properties, designs a two-stage training algorithm, and analyzes attention gradient properties.
result Trained transformers converge sub-linearly to max-margin solutions and exhibit linear convergence in cross-entropy loss.
RVI accelerates encoderless VI for faster convergence.
problem Slow convergence in encoderless VI methods.
method Introduces Relay Variational Inference (RVI) for faster learning.
result RVI outperforms existing methods in convergence speed and performance.
We propose a new analytical approximation to the χ2 kernel that converges geometrically. The analytical approximation is derived with elementary methods and adapts to the input distribution for optimal convergence rate. Experiments show the new approximation leads to improved performance in image classification and …
This research explores how different discrete diffusion kernels affect graph generation quality.
problem The impact of different discrete diffusion kernels on graph generation quality.
method Developed a family of discrete diffusion kernels that converge to different Bernoulli priors.
result The quality of generated graphs is sensitive to the prior used, challenging previous intuitions.
Develops robust MDPs for unknown disturbances with performance guarantees.
problem Unknown disturbance distribution in MDPs.
method Empirical distribution, sublevel set of distance function, weak convergence, concentration inequality.
result Robust optimal value function converges to true optimal value function with increasing sample sizes.
Local Bayesian optimization shows strong performance and converges well, contrary to folklore.
problem Understanding the behavior and convergence of local Bayesian optimization methods.
method Studied the behavior of local optimization strategies and rigorously analyzed a specific algorithm.
result Local Bayesian optimization algorithms converge well and perform strongly, contrary to the folklore.
We extend contrastive learning theory for multiway classification and prove convergence guarantees.
problem Efficient self-supervised training for multiway classification tasks.
method Contrastive representation learning with multiple negative samples and convergence guarantees for gradient descent.
result Convergence guarantees for contrastive learning with gradient descent of an overparametrized encoder.
New method uses momentum to converge in DC optimization with small batches.
problem Lack of convergence properties for stochastic difference-of-convex optimization with small batch sizes.
method Introduces momentum to enable convergence under standard assumptions for any batch size.
result Proves convergence of the algorithm under smoothness and bounded variance assumptions.
Novel geometry-informed irreversible perturbation accelerates Langevin dynamics convergence.
problem Accelerating convergence of Langevin dynamics for Bayesian computation.
method Geometry-informed irreversible perturbation of Riemannian manifold Langevin dynamics.
result Improves estimation performance over irreversible perturbations that ignore geometry.
Direct proof shows adaptive gradient descent converges near-linearly for convex functions.
problem Proving near-linear convergence of adaptive gradient descent for convex functions.
method Direct Lyapunov-based argument for convex functions with unique minimizer.
result Direct proof of near-linear convergence for convex functions.
Stochastic Gradient Descent has been widely studied with classification accuracy as a performance measure. However, these stochastic algorithms cannot be directly used when non-decomposable pairwise performance measures are used such as Area under the ROC curve (AUC) which is a common performance metric when the classe…
Recent works have shown that stochastic gradient descent (SGD) achieves the fast convergence rates of full-batch gradient descent for over-parameterized models satisfying certain interpolation conditions. However, the step-size used in these works depends on unknown quantities and SGD's practical performance heavily re…
Uniform convergence of isotopies implies ambient isotopy, aiding knot equivalence.
problem Determining when uniform convergence of isotopies leads to ambient isotopies.
method Using a diagrammatic condition to offload uniform convergence, constructing examples of tame knots.
result Constructing tame knots with countably-many crossings, distinguishing them from wild curves.
LLMs predict SGD convergence without training on specific data.
problem Predicting SGD convergence without labeled data.
method Link between SGD and Markov chains, leveraging LLMs' understanding of dynamical systems.
result Zero-shot prediction of SGD convergence for new starting points.
New proof shows coupling-based flows converge linearly to diagonalize data covariance.
problem Understanding convergence of coupling-based normalizing flows to arbitrary data distributions.
method Proved linear convergence rate for whitening of data distribution.
result Coupling-based flows achieve linear convergence to diagonalize data covariance.
In this paper, we explore techniques centered around periodic sampling of model weights that provide convergence improvements on gradient update methods (vanilla \acs{SGD}, Momentum, Adam) for a variety of vision problems (classification, detection, segmentation). Importantly, our algorithms provide better, faster and …
We propose a new stochastic L-BFGS algorithm and prove a linear convergence rate for strongly convex and smooth functions. Our algorithm draws heavily from a recent stochastic variant of L-BFGS proposed in Byrd et al. (2014) as well as a recent approach to variance reduction for stochastic gradient descent from Johnson…
Proposes a new algorithm for efficient online model selection of LLMs considering the increasing-then-converging trend.
problem Balancing cost and performance in choosing the best LLM among a diverse set of models.
method Introduces a time-increasing bandit algorithm (TI-UCB) that predicts model performance increases and balances exploration and exploitation.
result Achieves a logarithmic regret upper bound, indicating a fast convergence rate in model selection.
The study assesses ML model robustness under worst-case subpopulations.
problem ML model performance degradation under non-training population.
method Two-stage estimation procedure for evaluating worst-case robustness over subpopulations.
result The method certifies model robustness and prevents unreliable deployments.
A new Universal Activation Function improves performance across various machine learning tasks.
problem Achieving near optimal performance in different machine learning tasks.
method Optimization algorithms evolve the UAF's parameters to match the optimal activation function for each task.
result The UAF converges to near optimal performance in classification, quantification, and reinforcement learning tasks.
Federated learning (FL) provides a communication-efficient approach to solve machine learning problems concerning distributed data, without sending raw data to a central server. However, existing works on FL only utilize first-order gradient descent (GD) and do not consider the preceding iterations to gradient update w…
pFedGame uses game theory for decentralized federated learning in dynamic networks.
problem Performance bottlenecks, data bias, model convergence issues, and model poisoning attacks in federated learning.
method pFedGame employs game theory to decentralize federated learning, avoiding a central aggregation server and addressing dynamic network challenges.
result pFedGame achieves higher accuracy (over 70%) in heterogeneous data compared to existing methods.
A new optimizer, Grad-Avg, converges faster than SGD and improves classification task performance.
problem Optimizing error functions in regression and classification tasks.
method Grad-Avg optimizer based on gradient averaging, with mathematical convergence proof and parameter scaling for classification.
result Grad-Avg converges faster than other optimizers for classification tasks on benchmark datasets.
New method accelerates convergence for entropy-regularized reinforcement learning problems.
problem Slow convergence of standard first-order methods for entropy-regularized Markov decision processes.
method Introduce a quadratically convexified primal-dual formulation and a new interpolating metric to accelerate convergence.
result Global convergence and exponential convergence rate for the new method.
Efficient RNN algorithm guarantees convergence in online learning.
problem Online nonlinear regression with RNNs.
method First-order training algorithm with convergence guarantee.
result The algorithm converges to optimum network parameters.
The Expectation-Maximization algorithm is perhaps the most broadly used algorithm for inference of latent variable problems. A theoretical understanding of its performance, however, largely remains lacking. Recent results established that EM enjoys global convergence for Gaussian Mixture Models. For Mixed Linear Regres…
FedCluster accelerates federated learning convergence by cycling device groups.
problem Federated learning convergence issues with device-level data heterogeneity.
method FedCluster groups devices into clusters that cycle through learning rounds, boosting convergence with meta-updates.
result FedCluster achieves faster convergence in nonconvex optimization compared to FedAvg.
Gradient Langevin dynamics (GLD) and stochastic GLD (SGLD) have attracted considerable attention lately, as a way to provide convergence guarantees in a non-convex setting. However, the known rates grow exponentially with the dimension of the space. In this work, we provide a convergence analysis of GLD and SGLD when t…
In this paper, we study a popular method for inference of the Bradley-Terry model parameters, namely the MM algorithm, for maximum likelihood estimation and maximum a posteriori probability estimation. This class of models includes the Bradley-Terry model of paired comparisons, the Rao-Kupper model of paired comparison…
Adam-type optimizers show one-sided convergence in GAN training, not reaching critical points.
problem Theoretical understanding of Adam-type optimizers in non-convex non-concave min-max optimization.
method Empirical and theoretical analysis of Adam-type algorithms' convergence in GAN training.
result Adam-type algorithms converge to one-sided first order stationary points under the one-sided MVI condition.
A Support Vector Method for multivariate performance measures was recently introduced by Joachims (2005). The underlying optimization problem is currently solved using cutting plane methods such as SVM-Perf and BMRM. One can show that these algorithms converge to an eta accurate solution in O(1/Lambda*e) iterations, wh…
In this paper we analyze, evaluate, and improve the performance of training generalized linear models on modern CPUs. We start with a state-of-the-art asynchronous parallel training algorithm, identify system-level performance bottlenecks, and apply optimizations that improve data parallelism, cache line locality, and …
Many practitioners who use the EM algorithm complain that it is sometimes slow. When does this happen, and what can be done about it? In this paper, we study the general class of bound optimization algorithms - including Expectation-Maximization, Iterative Scaling and CCCP - and their relationship to direct optimizatio…
Muon fails to converge on convex Lipschitz functions.
problem Understanding the convergence of Muon on convex and Lipschitz functions.
method Analyzing Muon's performance on convex and Lipschitz functions without error feedback.
result Muon does not converge on convex and Lipschitz functions, regardless of learning rate schedule.
Kernel learning FBSDE filter improves nonlinear filtering efficiency.
problem Nonlinear filtering problem in high-dimensional systems.
method Iterative and adaptive meshfree approach using forward backward SDE and KDE.
result Rigorous convergence analysis provided, supporting empirical results.
This paper analyzes OCBA algorithms' convergence rates for DEDS optimization.
problem Optimizing discrete-event dynamic systems with limited computing resources.
method Characterizes convergence rates of two OCBA algorithms under different performance measures.
result OCBA algorithms achieve optimal convergence rates under probability of correct selection and expected opportunity cost measures.
AdaX improves Adam by exponentially accumulating past gradients, leading to better performance in machine learning tasks.
problem Adam's fast convergence can lead to local minimums in non-convex problems.
method AdaX exponentially accumulates past gradients to adaptively tune the learning rate.
result AdaX outperforms Adam in various machine learning tasks, including computer vision and natural language processing.
New step-size methods improve SHB convergence for stochastic optimization.
problem Tuning step-size and momentum parameters in SHB is challenging.
method Proposed MomSPSmax, MomDecSPS, and MomAdaSPS for SHB. result Convergence guarantees for SHB to solution neighborhoods and exact minimizers.
Analysis of SGD+M convergence rates in high dimensions with batch size considerations.
problem Understanding convergence rates of SGD+M in high-dimensional settings.
method Analyzing the dynamics of SGD+M on least squares problems with large batch sizes and dimensions.
result Identifies the implicit conditioning ratio (ICR) that regulates SGD+M's acceleration and convergence rates.