Research
On-device research index

arXiv research

A locally-built, LLM-digested index of recent arXiv papers in quant finance, geometry/topology, and statistical ML — keyword search served straight from SQLite on this machine.

169,181 papers · 148 categories

Trend · papers per month

5101520 · Jun 202019922001200920182026
48 results for Gumbel trick

The Gumbel-max trick and its extensions simplify sampling from categorical distributions in machine learning.

problem Sampling from categorical distributions with unnormalized probabilities.
method Extensions of the Gumbel-max trick for various applications.
result Simplified and efficient methods for sampling and gradient estimation.

The Gumbel trick is a method to sample from a discrete probability distribution, or to estimate its normalizing partition function. The method relies on repeatedly applying a random perturbation to the distribution in a particular way, each time solving for the most likely configuration. We derive an entire family of r…

2017-06-13abs ↗pdf ↗

A new algorithm FastGM speeds up generating Gumbel-Max variables.

problem Efficiently generating multiple Gumbel-Max variables from high-dimensional vectors.
method FastGM reduces time complexity from O(kn+)O(kn^+) to O(klnk+n+)O(k \ln k + n^+) by generating variables in descending order.
result Significantly reduces computation time for generating kk Gumbel-Max variables.

Paper proposes a new estimator for generic discrete distributions.

problem Estimating gradients for stochastic nodes in deep generative models.
method Generalized Gumbel-Softmax estimator using truncation, Gumbel-Softmax trick, and linear transformation.
result Efficacy and practical value demonstrated in synthetic examples and topic models.

Improves performance in various machine learning tasks by reparameterizing subset sampling.

problem Stochastic optimization involving subset sampling is not reparameterizable.
method Continuous relaxation of subset sampling to provide reparameterization gradients.
result Improves performance in instance-wise feature selection, deep stochastic k-nearest neighbors, and parametric t-SNE.

A new method reparameterizes Gaussian noise for better flexibility and performance.

problem Improving the Gumbel-Softmax for better flexibility and performance.
method Invertible Gaussian Reparameterization (IGR) using modified softmax and transformations.
result IGR outperforms Gumbel-Softmax in various experiments.

Direct optimization of discrete variational auto-encoders using arg max.

problem Optimizing discrete latent variables in variational auto-encoders.
method Direct optimization through arg max without softmax relaxations.
result Empirical effectiveness of direct loss minimization in discrete latent variables.

This paper introduces Gumbel-Sinkhorn networks for learning latent matchings.

problem Learning in latent variable models with permutations is difficult due to combinatorial intractability.
method Approximates maximum-weight matching using the Sinkhorn operator, extending Gumbel-Softmax.
result Demonstrates effectiveness on sorting, jigsaw puzzles, and neural signal identification tasks.

Proposes a new method for estimating counterfactual treatment effects.

problem Uncertainty in identifying causal mechanisms from observational data.
method Introduces a parameterized family of causal mechanisms that generalize Gumbel-max, trained to minimize counterfactual effect variance.
result Trained mechanisms yield lower variance estimates of counterfactual treatment effects.

GSO framework optimizes COPs on graphs using Gumbel-softmax.

problem Finding optimal configurations or network structures in combinatorial optimization problems.
method Introducing Gumbel-softmax technique for direct optimization of objective functions.
result High-quality solutions obtained with less time compared to traditional methods.

Categorical variables are a natural choice for representing discrete structure in the world. However, stochastic neural networks rarely use categorical latent variables due to the inability to backpropagate through samples. In this work, we present an efficient gradient estimator that replaces the non-differentiable sa…

2016-11-03abs ↗pdf ↗

The problem of drawing samples from a discrete distribution can be converted into a discrete optimization problem. In this work, we show how sampling from a continuous distribution can be converted into an optimization problem over continuous space. Central to the method is a stochastic process recently described in ma…

2014-10-31abs ↗pdf ↗

Investigates statistical properties of perturb-softmax and perturb-argmax distributions.

problem Underexplored statistical properties of Gumbel-Softmax and Gumbel-Argmax distributions.
method Investigates convexity and differentiability to determine completeness and minimality of these distributions.
result Identifies parameters that admit complete and minimal representation of probability distributions.

Differentiable NAS method optimizes network architecture and parameters efficiently.

problem Challenging to simultaneously guarantee effectiveness and efficiency in network architecture search.
method Differentiable architecture search with ensemble Gumbel-Softmax estimator.
result End-to-end mechanism for searching network architectures, discovering high-performance architectures efficiently.

Unified approach to DP problems using Gumbel distribution and variational Bayesian inference.

problem Solving classical optimal path problems in a probabilistic framework.
method Gumbel distribution and variational Bayesian inference for latent optimal paths.
result Unified approach transforms DP problems into directed acyclic graphs with Gibbs distribution.

GDM models time series with smoother transitions and interpretable states.

problem Capturing smooth, variable-speed transitions and stochastic mixtures of states.
method Introduces a continuous relaxation of discrete states and a Gumbel noise model.
result Models real-world datasets more faithfully with smoother dynamics and interpretable states.

Study on estimating Gumbel--Max watermark proportions in edited documents.

problem Estimating the proportion of a document generated from a watermarked LLM.
method Comparison of full observation and pivotal reduction observation regimes; development of estimators and information-theoretic lower bounds.
result Full observation yields a substantially smaller sample complexity compared to pivotal reduction.

LLM-as-a-service prices vary arbitrarily due to tokenization multiplicity.

problem Arbitrary price variation in LLM-as-a-service due to multiple tokenizations of the same output.
method Introduce canonical generation to restrict LLMs to unique tokenizations and develop an efficient sampling algorithm.
result Our sampling algorithm for canonical generation solves tokenization multiplicity and maintains comparable performance and runtime to standard sampling.

Improved learning of probabilistic box embeddings by modeling parameters with Gumbel distributions.

problem Local identifiability issues in geometric embeddings.
method Modeling box parameters with min and max Gumbel distributions, calculating expected intersection volume.
result Improves the ability of probabilistic box embeddings to learn.

Improved CAEs reduce training time and enhance generalization.

problem Stability issues in Concrete Autoencoders (CAEs) for feature selection.
method Indirectly Parameterized Concrete Autoencoders (IP-CAEs) learn parameters of Gumbel-Softmax distributions.
result IP-CAEs achieve significant improvements in generalization and training time.

New framework maximizes perturbed samples for inverse classification with budget constraints.

problem Maximizing perturbed samples for desired classification outcomes under budget constraints.
method Gradient methods, stochastic processes, Lagrangian relaxations, Gumbel trick.
result Stochastic process-based algorithms outperform in different budget settings.

A new gradient estimator for categorical distributions reduces bias and variance.

problem Intractability of gradients for categorical distributions in discrete latent variable models.
method CatLog-Derivative trick and IndeCateR gradient estimator.
result IndeCateR reduces bias and variance of gradients for categorical distributions.

Doubling tricks help improve multi-armed bandit algorithms, but their effectiveness depends on the horizon length.

problem Improving the performance of multi-armed bandit algorithms using doubling tricks.
method Analyzed geometric and exponential doubling tricks for different horizon lengths.
result Geometric doubling tricks can conserve regret bounds in O(T)O(\sqrt{T}), but not in O(logT)O(\log T). Exponential doubling tricks can conserve bounds in O(logT)O(\log T).