arXiv research

PipeDream-2BW accelerates large model training by 20x with minimal memory usage.

The paper explores coalescent contractions in contractible spaces, providing criteria and examples.

We propose a new algorithm to do posterior sampling of Kingman's coalescent, based upon the Particle Markov Chain Monte Carlo methodology. Specifically, the algorithm is an instantiation of the Particle Gibbs Sampling method, which alternately samples coalescent times conditioned on coalescent tree structures, and tree…

Develops a variational method for ultrametric phylogenetic trees.

We introduce a new Bayesian model for hierarchical clustering based on a prior over trees called Kingman's coalescent. We develop novel greedy and sequential Monte Carlo inferences which operate in a bottom-up agglomerative fashion. We show experimentally the superiority of our algorithms over others, and demonstrate o…

New algorithms learn simple staged trees from data, improving model fit.

Paper connects Painlevé VI equation to irregular systems, solving monodromy data.

Linear-cost unbiased estimates for complex models via couplings.

Two oppositely charged droplets of (say) water in e.g. oil or air will tend to drift together under the influence of their charges. As they make contact, one might expect them to coalesce and form one large droplet, and this indeed happens when the charge difference is sufficiently small. However, Ristenpart et al disc…

We extend the analytic theory of Frobenius manifolds to semisimple points with coalescing eigenvalues of the operator of multiplication by the Euler vector field. We clarify which freedoms, ambiguities and mutual constraints are allowed in the definition of monodromy data, in view of their importance for conjectural re…

New definition of angular momentum avoids supertranslation ambiguity.

Kolmogorov-Arnold network improves GW catalog posterior construction.

New algorithm learns mixtures of any constant number of Gaussians robustly.

Study on kinetic Langevin diffusions and their couplings, showing subtle TV bounds and new non-Markovian couplings.

Convex clustering is a recent stable alternative to hierarchical clustering. It formulates the recovery of progressively coalescing clusters as a regularized convex problem. While convex clustering was originally designed for handling Euclidean distances between data points, in a growing number of applications, the dat…

Bayesian Neural Networks detect gravitational wave events with high accuracy and real-time potential.

We propose a nonparametric Bayesian factor regression model that accounts for uncertainty in the number of factors, and the relationship between factors. To accomplish this, we propose a sparse variant of the Indian Buffet Process and couple this with a hierarchical model over factors, based on Kingman's coalescent. We…

Study models Indian stock market using hyperbolic geometry for market stability and volatility analysis.

Gradient descent on normalized networks reveals sparsity preferences.

We give a complete description of finite braid group orbits in Aff(C)-character varieties of the punctured Riemann sphere. This is performed thanks to a coalescence procedure and to the theory of finite complex reflection groups. We then derive consequences in the theory of differential equations. These concern algebra…

Extends ML fairness to handle minority groups over time.

New RL approach builds short ancestral recombination graphs.

In distributed function computation, each node has an initial value and the goal is to compute a function of these values in a distributed manner. In this paper, we propose a novel token-based approach to compute a wide class of target functions to which we refer as "Token-based function Computation with Memory" (TCM) …

Paper extends Aronson-Bénilan estimates for porous medium equations on manifolds with negative curvature.

The paper estimates gradients for a weighted parabolic equation under geometric flow.

A new accelerated method with simpler momentum update rules.

Gradient flow on softmax attention minimizes nuclear norm of weight matrices.

Gradient descent with random weights in linear regression analyzed for various noise types.

Proposes a method to balance imbalanced image datasets using capsule-GAN.

The paper analyzes neural network dynamics after weights escape the origin.

Enhanced visual feature attribution via adaptive baseline weighting.

Quantized deep neural networks (QDNNs) are attractive due to their much lower memory storage and faster inference speed than their regular full precision counterparts. To maintain the same performance level especially at low bit-widths, QDNNs must be retrained. Their training involves piecewise constant activation func…

Article provides Bernstein gradient estimates for heat equations with potential terms.

This paper compares gradient estimators in importance-weighted VI and justifies the superiority of DREP over REP.

Gradient descent converges to perfect classification in neural nets for non-separable data.

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

AGBoost uses attention weights to improve GBM for regression problems.

Gradient flow with weight decay shows grokking effect in deep learning.

In this paper, we prove that a noncompact complete hypersurface with finite weighted volume, weighted mean curvature vector bounded in norm, and isometrically immersed in a complete weighted manifold is proper. In addition, we obtain an estimate for $f$-stability index of a constant weighted mean curvature hypersurface…

Algorithm learns which weights to share in deep multi-task learning.

Much of the focus in machine learning research is placed in creating new architectures and optimization methods, but the overall loss function is seldom questioned. This paper interprets machine learning from a multi-objective optimization perspective, showing the limitations of the default linear combination of loss f…

U-statistics improve gradient estimation in importance-weighted variational inference.

Derives gradient estimation for a specific heat equation on evolving manifolds.

Stochastic gradient descent on manifolds improves low-rank approximation.

Paper improves REINFORCE for VI without restrictive assumptions.

Paper explores weighted averaging schemes for SGD, achieving asymptotic normality and optimality.

Paper proves multiplicative weight updates can train neural networks without learning rate tuning.

This paper analyzes convergence of large-scale Transformers with weight decay.