This work explores the relation between trainability and dequantization in variational QML models.
problem Understanding the interplay between trainability and dequantization in variational QML models.
method Provide precise definitions of trainability and dequantization, study their relation, and introduce recipes for building PQC-based QML models.
result Identify conditions under which trainability and non-dequantization are not mutually exclusive.
Flow++ improves flow-based models by dequantizing with variational methods and using expressive architectures.
problem Flow-based models have poor density estimation compared to autoregressive models.
method Variational dequantization, expressive affine flows, and improved architecture design.
result Flow++ is now the state-of-the-art non-autoregressive model for unconditional density estimation.
This paper introduces methods to handle discrete data by dequantization.
problem Handling discrete data in deep learning models.
method Dequantization framework, including importance-weighted and Rényi dequantization objectives, and autoregressive dequantization.
result Improved performance on uniform dequantization distributions and state-of-the-art negative log-likelihood on CIFAR10.
MSD removes dequantization bottleneck in LLM inference by approximating high-precision activations.
problem Dequantization bottleneck in LLM inference on modern AI accelerators.
method MSD decomposes high-precision activations into multiple low-precision components for direct multiplication with quantized weights.
result MSD avoids INT8-to-BF16 weight conversion, reducing dequantization cycles and HBM traffic.
Method estimates densities on manifolds using dequantization.
problem Estimating densities on non-Euclidean manifolds.
method Inspired by dequantization, coordinate transformation, and normalizing flows.
result Successfully models densities on spheres, tori, and orthogonal groups.
PixelCNN models can achieve state-of-the-art results on CIFAR-10 with exact likelihood computation.
problem Dequantization gap in modeling discrete data like images.
method Introducing subset flows to allow exact computation of likelihoods for discrete data.
result PixelCNN models trained with exact likelihood computation achieve state-of-the-art results on CIFAR-10.
Develops a new method for learning discrete distributions without embedding them in a continuous space.
problem Challenges in learning discrete distributions using current methodologies.
method Introduces a MAD invertible map and a mixed variational flow (MAD Mix) for discrete distributions.
result MAD Mix produces more reliable approximations than continuous-embedding flows.
SurVAE Flows combine VAEs and flows using surjective transformations.
problem Combining the strengths of VAEs and flows to model complex densities.
method Modular framework of composable deterministic and stochastic transformations.
result Exact likelihood computation and lower bound on likelihood.
Quantum circuits reveal pathways to dequantization in machine learning models.
problem Navigating the complex landscape of quantum machine learning models and algorithms.
method Introducing a framework connecting quantum circuit structure to function representability.
result Fundamental properties of quantum circuits determine classical simulability of models.
Paper introduces Categorical Normalizing Flows for better handling of categorical data.
problem Limited application of normalizing flows on categorical data due to lack of intrinsic order.
method Categorical Normalizing Flows use continuous transformations to model latent relations in categorical data, optimizing both continuous representation and model likelihood.
result GraphCNF, a permutation-invariant generative model, outperforms state-of-the-art on molecule generation.
In this work, we propose a deep neural network architecture motivated by primal-dual splitting methods from convex optimization. We show theoretically that there exists a close relation between the derived architecture and residual networks, and further investigate this connection in numerical experiments. Moreover, we…
We consider the reconstruction problem in compressed sensing in which the observations are recorded in a finite number of bits. They may thus contain quantization errors (from being rounded to the nearest representable value) and saturation errors (from being outside the range of representable values). Our formulation …
Paper proposes Sp-GD for sparse max-affine regression with theoretical guarantees.
problem Sparse max-affine regression model selection and estimation.
method Sparse Gradient Descent (Sp-GD) initialization using sparse PCA and covering search.
result Sp-GD provides ε-accurate estimates with optimal number of observations.
Paper introduces new methods for modeling categorical data.
problem Training generative models on categorical data like text and segmentation.
method Argmax Flows and Multinomial Diffusion models.
result Models outperform existing methods in log-likelihood.
Novel flows generate molecules without post-processing.
problem Generating new molecules efficiently and without post-processing issues.
method Continuous normalizing E(3)-equivariant flows based on node ODEs coupled as a graph PDE.
result Generated samples achieve state-of-the-art performance on QM9 and ZINC250K benchmarks.
Extends ONNX for quantized neural networks with new formats and operators.
problem Handling arbitrary-precision quantization in neural networks.
method Introduces new formats and operators in ONNX to represent quantized neural networks.
result Enabled representation of uniform quantization in neural networks.
Unified framework for robust, stable, and efficient density ratio estimation.
problem Density-chasm and support-chasm problems in density ratio estimation.
method Dequantified diffusion-Schrödinger bridge (D3RE) framework with DDBI and DSBI.
result Offers uniform approximation and bounded time scores in theory and empirical performance.
A new method learns quantization boundaries in continuous space using tessellation.
problem Mapping between discrete and continuous distributions is difficult.
method Constructs normalizing flows on convex polytopes with exact likelihood evaluations.
result Improves likelihood evaluation and quantization learning across various data modalities.
Let K⊂S3 be a knot, X:=S3∖K its complement, and T the circle group identified with R/Z. To any oriented long knot diagram of K, we associate a quadratic polynomial in variables bijectively associated with the bridges of the diagram such that, when the variables pr…
FleXOR trains fractional quantization for neural networks, improving accuracy and size.
problem Quantization limits to integer bits restricts compression and accuracy.
method Encryption algorithm with XOR gates for fractional bits during inference.
result FleXOR achieves high accuracy with fractional sub-1-bit weights.
Improved EXACT strategy reduces GNN memory consumption and runtime.
problem Efficiently training large-scale GNNs with reduced memory usage.
method Block-wise quantization of intermediate activation maps with improved variance minimization.
result Further reduction in memory consumption (>15%) and runtime speedup (5%) with similar performance trade-offs.
New classical algorithm outperforms quantum in neural network subnetwork selection.
problem Selecting sparse subnetworks from large neural networks efficiently.
method Quantum-inspired classical algorithm using ridgelet transform sampling.
result Runs in polynomial time, outperforming naive classical methods.
Develops c-GNF for personalized social science policy analysis.
problem Challenges in estimating causal effects and counterfactual inference in social sciences.
method causal-Graphical Normalizing Flow (c-GNF) method.
result c-GNF performs well in estimating causal effects and counterfactual inference.
New variational principle found for non-variational differential equations.
problem Non-variational differential equations without variational multipliers.
method Connecting functional forms with antiexact differential forms to identify obstructions.
result Formulation of variational problem for non-variational equations.
Derives Lagrangian for minimal surfaces, proving tangential variations vanish.
problem Variational calculus for minimal surfaces.
method Lagrangian formulation, pullback covariant derivative, geometric argument.
result Tangential variations vanish for minimal surfaces.
Improved Bayesian uncertainty quantification using variational bagging.
problem Inefficient and underestimating uncertainty in mean-field variational Bayes.
method Integrates bagging with variational Bayes for improved inference.
result Bagged variational posterior provides proper uncertainty quantification.
This work improves variational inference by reducing gradient variance.
problem Hard optimization of flexible variational distributions.
method Control variate based on quadratic approximation of the model's mean and covariance.
result Significant improvement in gradient variance and optimization convergence.
A new natural gradient accounts for correlated variational parameters in variational inference.
problem Traditional natural gradients fail to correct for correlations in variational inference.
method Construct a new natural gradient called the Variational Predictive Natural Gradient (VPNG).
result VPNG accounts for the relationship between model parameters and variational parameters.
Adaptive variational Bayes framework improves inference adaptively.
problem Lack of general and computationally tractable variational Bayes method for adaptive inference.
method Proposes a novel adaptive variational Bayes framework combining variational posteriors over individual models.
result Adaptive variational Bayes achieves optimal contraction rates adaptively under general conditions.
A new EVI framework improves ParVI methods by maintaining variational structure and reducing KL-divergence.
problem Improving variational inference methods for better approximation of target distributions.
method EVI framework that minimizes the VI objective function based on an energy-dissipation law, including a new 'Approximation-then-Variation' scheme.
result The new scheme significantly decreases KL-divergence and outperforms existing ParVI methods in fidelity.
We propose a family of variational approximations to Bayesian posterior distributions, called α-VB, with provable statistical guarantees. The standard variational approximation is a special case of α-VB with α=1. When α∈(0,1], a novel class of variational inequalities are developed for linking the Bayes risk …
New examples of variational bivectors found that are not Poissonian.
problem Identifying variational bivectors that are not Poissonian.
method Constructing examples of variational bivectors.
result Found examples of variational bivectors that are not Poissonian.
Variational autoencoders learn deep latent models.
problem Learning deep latent-variable models.
method Principled framework using variational inference.
result Introduction to variational autoencoders and extensions.
Improves VAE training by refining variational parameters with BSVI.
problem Amortized inference in VAEs leads to suboptimal variational parameters and the amortization gap.
method Proposes BSVI, a refinement procedure using SVI's importance weights.
result Training VAEs with BSVI yields improved performance compared to SVI.
New method improves variational inference for hierarchical models.
problem Limited expressivity of variational distributions in Bayesian models.
method Importance weighted hierarchical variational inference.
result Superior performance in experiments compared to existing methods.
Improved VAE estimation from incomplete data using variational mixtures.
problem Estimating VAEs from incomplete data increases posterior complexity.
method Introducing variational mixtures based on finite and imputation distributions.
result Variational mixtures improve VAE estimation accuracy from incomplete data.
Variational Prediction simplifies Bayesian inference without test time costs.
problem Bayesian inference's computational costs and posterior predictive distribution marginalization.
method Variational Prediction learns a variational approximation to the posterior predictive distribution using a variational bound.
result Directly learns a variational approximation to the posterior predictive distribution without test time marginalization costs.
In the paper "On Truncated Variation of Brownian Motion with Drift" (Bull. Pol. Acad. Sci. Math. 56 (2008), no.4, 267 - 281) we defined truncated variation of Brownian motion with drift, Wt=Bt+μt,t≥0, where (Bt) is a standard Brownian motion. Truncated variation differs from regular variation by neglect…
Semi-Implicit Variational Inference (SIVI) is improved with SIVI-SM using score matching.
problem Intractable densities in variational distributions hinder SIVI training.
method SIVI-SM uses score matching to handle intractable densities in a minimax formulation.
result SIVI-SM outperforms ELBO-based SIVI methods in Bayesian inference tasks.
This work proposes using zero-variance control variates to reduce variance in pathwise gradient estimators for variational inference.
problem Pathwise gradient estimators in variational inference have high variance, leading to inefficient optimization.
method Apply zero-variance control variates to pathwise gradient estimators.
result Zero-variance control variates can significantly reduce the variance of pathwise gradient estimators without requiring complex assumptions.
New classification of hypersurfaces with conformal variations.
problem Classifying hypersurfaces with conformal infinitesimal variations.
method Analyzing hypersurfaces in conformal geometry, extending previous work by Cartan and Sbrana.
result The class of hypersurfaces with conformal infinitesimal variations is larger than previously known.
The paper classifies and studies conformal variations of submanifolds.
problem Classifying and understanding conformal variations of submanifolds.
method Develops a Fundamental theorem and a rigidity theorem for Euclidean submanifolds.
result Fundamental theorem and rigidity theorem for Euclidean submanifolds.
This tutorial derives the VAE loss function under Gaussian assumptions.
problem Computational intractability of posterior distributions in Bayesian machine learning.
method Derives the variational lower bound loss function of a standard VAE.
result The Kullback-Leibler divergence has a closed form solution under Gaussian assumptions.
Variational inference is increasingly being addressed with stochastic optimization. In this setting, the gradient's variance plays a crucial role in the optimization procedure, since high variance gradients lead to poor convergence. A popular approach used to reduce gradient's variance involves the use of control varia…
Variational approach to basic manifold structures.
problem Understanding basic differential geometric structures.
method Variational description of geometric structures.
result Variational formulation of manifold structures.
The paper introduces structured variational families to improve scalability in black-box variational inference.
problem Scalability issues in black-box variational inference, especially for large datasets and hierarchical models.
method Developed structured variational families that achieve better iteration complexity of O(N) compared to full-rank families.
result Structured variational families can achieve better scaling with respect to dataset size N, improving iteration complexity from O(N^2) to O(N).
New method reduces inference variance for faster optimization.
problem High variance in black-box variational inference.
method Joint control variate addressing both data subsampling and Monte Carlo noise.
result Significantly reduced gradient variance, leading to faster optimization.
We derive a formula for the first variation of horizontal perimeter measure for C2 hypersurfaces of completely general sub-Riemannian manifolds, allowing for the existence of characteristic points. For C2 hypersurfaces in vertically rigid sub-Riemannian manifolds we also produce a second variation formula for var…