Kullback Leibler

The Kullback-Leibler (KL) divergence is a measure of the difference between two probability distributions, frequently used as a regularization term or objective function in various machine learning applications. Current research focuses on leveraging KL divergence for improved variational inference, particularly in complex models like Gaussian processes and deep learning architectures, often employing techniques like sequential Monte Carlo or advantage learning to overcome computational challenges and enhance accuracy. This work is significant because it enables more robust and efficient training of probabilistic models, impacting fields ranging from time series forecasting to data augmentation and reinforcement learning.

Papers

November 9, 2024

Learning Mixtures of Experts with EM
Quentin Fruytier, Aryan Mokhtari, Sujay Sanghavi
Language Model Machine Learning Model Gradient Descent Expert Knowledge Expectation Maximization Learning Mixture Kullback Leibler

November 7, 2024

Sharp Analysis for KL-Regularized Contextual Bandits and RLHF
Heyang Zhao, Chenlu Ye, Quanquan Gu, Tong Zhang
Reinforcement Learning Contextual Bandit Reinforcement Learning From Human Feedback Kullback Leibler

March 15, 2024

Sequential Monte Carlo for Inclusive KL Minimization in Amortized Variational Inference
Declan McNamara, Jackson Loper, Jeffrey Regier
Kullback Leibler Divergence Sequential Monte Carlo Variational Distribution Amortized Variational Inference Kullback Leibler

November 21, 2023

Probabilistic Forecast Reconciliation with Kullback-Leibler Divergence Regularization
Guanyu Zhang, Feng Li, Yanfei Kang
Kullback Leibler Forecast Reconciliation

June 16, 2023

SLACK: Stable Learning of Augmentations with Cold-start and KL regularization
Juliette Marrie, Michael Arbel, Diane Larlus, Julien Mairal
Data Augmentation Soft Augmentation Cold Start Stable Learning Augmentation Policy Kullback Leibler Data Augmentation Algorithm

January 30, 2023

Variational sparse inverse Cholesky approximation for latent Gaussian processes via double Kullback-Leibler minimization
Jian Cao, Myeongjong Kang, Felix Jimenez, Huiyan Sang, Florian Schafer, Matthias Katzfuss
Gaussian Process Posterior Approximation Variational Approximation Sparse Variational Kullback Leibler Cholesky Factorization

October 31, 2022

Lipschitz-regularized gradient flows and generative particle algorithms for high-dimensional scarce data
Hyemin Gu, Panagiota Birmpa, Yannis Pantazis, Luc Rey-Bellet, Markos A. Katsoulakis
Gradient Flow Gene Expression Data Target Distribution Generative Algorithm Particle Based Algorithm Lipschitz Gradient Kullback Leibler

July 18, 2022

Kullback-Leibler and Renyi divergences in reproducing kernel Hilbert space and Gaussian process settings
Minh Ha Quang
Gaussian Process Kernel Hilbert Space Divergence Regularization R\'enyi Divergence Kullback Leibler

May 23, 2022

RL with KL penalties is better viewed as Bayesian inference
Tomasz Korbak, Ethan Perez, Christopher L Buckley
Language Model Reinforcement Learning Bayesian Inference LLM Fine Tuning Regularized Reinforcement Learning Kullback Leibler

May 16, 2022

Enforcing KL Regularization in General Tsallis Entropy Reinforcement Learning via Advantage Learning
Lingwei Zhu, Zheng Chen, Eiji Uchibe, Takamitsu Matsubara
Maximum Entropy DQN Agent Entropy Regularized Advantage Learning Kullback Leibler Tsallis Entropy

Kullback Leibler

Papers

Learning Mixtures of Experts with EM

Sharp Analysis for KL-Regularized Contextual Bandits and RLHF

Sequential Monte Carlo for Inclusive KL Minimization in Amortized Variational Inference

Probabilistic Forecast Reconciliation with Kullback-Leibler Divergence Regularization

SLACK: Stable Learning of Augmentations with Cold-start and KL regularization

Variational sparse inverse Cholesky approximation for latent Gaussian processes via double Kullback-Leibler minimization

Lipschitz-regularized gradient flows and generative particle algorithms for high-dimensional scarce data

Kullback-Leibler and Renyi divergences in reproducing kernel Hilbert space and Gaussian process settings

RL with KL penalties is better viewed as Bayesian inference

Enforcing KL Regularization in General Tsallis Entropy Reinforcement Learning via Advantage Learning