A Review of the Gumbel-max Trick and its Extensions for Discrete Stochasticity in Machine Learning (2021-10-04T00:00:00.000000Z)

TL;DR

The goal of this survey article is to present background about the Gumbel-max trick, and to provide a structured overview of its extensions to ease algorithm selection, and presents a comprehensive outline of (machine learning) literature in which Gumbal-based algorithms have been leveraged.

Abstract

The Gumbel-max trick is a method to draw a sample from a categorical distribution, given by its unnormalized (log-)probabilities. Over the past years, the machine learning community has proposed several extensions of this trick to facilitate, e.g., drawing multiple samples, sampling from structured domains, or gradient estimation for error backpropagation in neural network optimization. The goal of this survey article is to present background about the Gumbel-max trick, and to provide a structured overview of its extensions to ease algorithm selection. Moreover, it presents a comprehensive outline of (machine learning) literature in which Gumbel-based algorithms have been leveraged, reviews commonly-made design choices, and sketches a future perspective.

Authors

Iris A. M. Huijben

1 papers

W. Kool

1 papers

Max B. Paulus

1 papers

TL;DR

Abstract

Authors

References171 items

Dynamic Probabilistic Pruning: A General Framework for Hardware-Constrained Pruning at Different Granularities

Zero-Shot Text-to-Image Generation

A compact review of molecular property prediction with graph neural networks.

Channel Selection Using Gumbel Softmax

More to diverse: Generating diversified responses in a task oriented multimodal dialog system

Event Representation with Sequential, Semi-Supervised Discrete Variables

Rao-Blackwellizing the Straight-Through Gumbel-Softmax Gradient Estimator

Interpreting Graph Neural Networks for NLP With Differentiable Edge Masking

Deep Learning Based Antenna Selection for Channel Extrapolation in FDD Massive MIMO

On Diverse Asynchronous Activity Anticipation

Neural machine translation with Gumbel Tree-LSTM based encoder

RetrieveGAN: Image Synthesis via Differentiable Patch Retrieval

Operation-Aware Soft Channel Pruning using Differentiable Masks

Adaptive Compression of Word Embeddings

Gradient Estimation with Stochastic Softmax Tricks

Improving Inference for Neural Image Compression

Learning to Branch for Multi-Task Learning

Active Learning for ML Enhanced Database Systems

A Relaxed Quantization Training Method for Hardware Limitations of Resistive Random Access Memory (ReRAM)-Based Computing-in-Memory

Community Detection Clustering via Gumbel Softmax

Which Evaluations Uncover Sense Representations that Actually Make Sense?

Deep probabilistic subsampling for task-adaptive compressed sensing

Learning Sampling and Model-Based Signal Recovery for Compressed Sensing MRI

Learning to Select Important Context Words for Event Detection

Drug-drug interaction prediction with Wasserstein Adversarial Autoencoder-based knowledge graph embeddings

Gumbel-softmax-based optimization: a simple general framework for optimization problems on graphs

FBNetV2: Differentiable Neural Architecture Search for Spatial and Channel Dimensions

Dynamic Network Pruning with Interpretable Layerwise Channel Selection

Predictive Sampling with Forecasting Autoregressive Models

Incremental Sampling Without Replacement for Sequence Models

Learning with Differentiable Perturbed Optimizers

Estimating Gradients for Discrete Random Variables by Sampling without Replacement

Fast Generating A Large Number of Gumbel-Max Variables

Paraphrase Generation with Latent Bag of Words

Invertible Gaussian Reparameterization: Revisiting the Gumbel-Softmax

Dynamic Convolutions: Exploiting Spatial Sparsity for Faster Inference

PyTorch: An Imperative Style, High-Performance Deep Learning Library

SAG-VAE: End-to-end Joint Inference of Data Representations and Feature Relations

Binary Autoencoder for Text Modeling

Stochastic Fusion for Multi-stream Neural Network in Video Classification

Feature Selection and Extraction for Graph Neural Networks

Deep Clusteringwith Concrete K-Means

Towards Hierarchical Discrete Variational Autoencoders

vq-wav2vec: Self-Supervised Learning of Discrete Speech Representations

Learning Dynamic Graph Embedding for Traffic Flow Forecasting: A Graph Self-Attentive Method

Adversarial Distillation for Learning with Privileged Provisions

Deep Declarative Networks

Stochastic Filter Groups for Multi-Task CNNs: Learning Specialist and Generalist Convolution Kernels

Learning Sub-Sampling and Signal Recovery With Applications in Ultrasound Imaging

Batch-shaping for learning conditional channel gated networks

Bi-Directional Message Passing Based Scanet for Human Pose Estimation

Latent multi-criteria ratings for recommendations

Monte Carlo Gradient Estimation in Machine Learning

Doc2hash: Learning Discrete Latent variables for Documents Retrieval

Concrete Autoencoders: Differentiable Feature Selection and Reconstruction

Network Pruning via Transformable Architecture Search

Seeker: Real-Time Interactive Search

Counterfactual Off-Policy Evaluation with Gumbel-Max Structural Causal Models

Dynamic Routing Networks

ARSM: Augment-REINFORCE-Swap-Merge Estimator for Gradient Backpropagation Through Categorical Variables

The Curious Case of Neural Text Degeneration

Modeling Point Clouds With Self-Attention and Gumbel Subset Sampling

Buy 4 REINFORCE Samples, Get a Baseline for Free!

Stochastic Beams and Where to Find Them: The Gumbel-Top-k Trick for Sampling Sequences Without Replacement

Stochastic Optimization of Sorting Networks via Continuous Relaxations

Reparameterizable Subset Sampling via Continuous Relaxations

Concrete Autoencoders for Differentiable Feature Selection and Reconstruction

A general deep learning framework for network reconstruction and dynamics learning

FBNet: Hardware-Aware Efficient ConvNet Design via Differentiable Neural Architecture Search

Learning to Assemble Neural Module Tree Networks for Visual Grounding

Recursive Visual Attention in Visual Dialog

End-to-end Feedback Loss in Speech Chain Framework via Straight-through Estimator

Learning to Screen for Fast Softmax Inference on Large Vocabulary Neural Networks

SNAS: Stochastic Neural Architecture Search

Differentiable Perturb-and-Parse: Semi-Supervised Parsing with a Structured Variational Autoencoder

Stochastic Layer-Wise Precision in Deep Neural Networks