Kullback--Leibler divergence
PulseAugur coverage of Kullback--Leibler divergence — every cluster mentioning Kullback--Leibler divergence across labs, papers, and developer communities, ranked by signal.
- used by Gotit.pub 70%
- used by ScienceCast 70%
- used by CatalyzeX 70%
- used by alphaXiv 70%
- instance of Gotit.pub 70%
- used by DagsHub 70%
- instance of Kullback-Leibler information as a basis for strong inference in ecological studies 70%
- instance of Wasserstein 70%
- used by Kullback-Leibler information as a basis for strong inference in ecological studies 70%
- used by Wasserstein 70%
12 day(s) with sentiment data
-
Knowledge Distillation: Shrinking LLMs for Efficient Deployment
Knowledge distillation is a technique used to compress large language models (LLMs) by transferring the learned behaviors of a large "teacher" model into a smaller "student" model. This process is crucial for deploying …
-
New research explores advanced speech enhancement using neural audio codecs · 2 sources tracked
Two new research papers explore advanced techniques for speech enhancement, focusing on methods that improve audio quality under challenging acoustic conditions. The first paper introduces a test-time adaptation approac…
-
Quantum algorithms promise speedups for sampling and optimization
Researchers have developed new quantum algorithms that offer speedups for sampling from complex probability distributions and for non-convex optimization tasks. These algorithms enhance classical methods like Langevin M…
-
New NeVI-Cut method enables uncertainty propagation without upstream data
Researchers have developed NeVI-Cut, a novel method for neural variational inference in cut-Bayes problems. This approach allows for the propagation of parameter uncertainty in downstream analyses without requiring acce…
-
New TTM method enhances machine learning knowledge distillation
Researchers have developed a new method called Temperature-Adaptive Transformed Teacher Matching (TTM) to improve knowledge distillation in machine learning. This approach addresses the limitations of fixed temperature …
-
LLMs show limited cross-lingual knowledge transfer, new distillation method favors reasoning · 2 sources tracked
Two new research papers explore how large language models acquire and retain knowledge. The first paper investigates factual knowledge transfer across languages, finding that models exhibit limited transfer from English…
-
Bayesian Experimental Design: KL Divergence vs. Wasserstein Distance
A new paper published on arXiv explores the use of Bayesian experimental design (BED) for calibrating model discrepancies. The research compares Kullback-Leibler (KL) divergence and Wasserstein distance as utility funct…
-
New theory quantifies parallel sampling cost in diffusion models
Researchers have developed a new theoretical framework for adaptive parallel sampling of discrete vectors, motivated by parallel decoding in masked diffusion models. The core finding is an exact identity linking approxi…
-
New theory combines offline and online learning for AI systems
Researchers have developed a novel theoretical framework for combining offline and online learning methods in artificial intelligence systems. This two-stage approach aims to improve prediction performance for non-stati…
-
AI research tackles GPS-spoofed drone separation
Researchers have developed a new method for ensuring separation between small Unmanned Aircraft Systems (sUAS) even when GPS signals are degraded or spoofed. This approach uses Multi-Agent Reinforcement Learning (MARL) …
-
AI safety verification method for aviation collision avoidance systems proposed
A new paper proposes a method for verifying the representativeness of data distributions used in AI/ML systems for aviation safety. The approach addresses European Union Aviation Safety Agency (EASA) requirements for de…
-
New algorithm precisely computes learning coefficients for singular AI models
Researchers have developed a new deterministic algorithm for precisely calculating learning coefficients in two-dimensional singular models. This method addresses limitations of traditional information criteria like BIC…
-
New Bethe free energy formulation enhances active inference capabilities
Researchers have proposed a new formulation for active inference using a Bethe free energy functional, which supports inference by message passing. This approach addresses limitations of existing methods where the free …
-
New distributional view of knowledge distillation for language models unveiled
Researchers have introduced a new distributional perspective on knowledge distillation (KD) for language models. This approach moves beyond pointwise comparisons of token distributions to consider a family of multi-temp…
-
New algorithm enhances generative models for extreme event prediction
Researchers have introduced the CVaR-penalized Generative Particle Algorithm (CVaR-GPA), a novel method for fine-tuning generative models to better capture extreme events and heavy-tailed distributions. This algorithm u…
-
New research explores online learning for score-driven filters
A new research paper published on arXiv details advancements in online learning for score-driven filters. The study focuses on optimizing the gain parameter, which controls the update magnitude in these filters, by trea…
-
New attacks target federated GANs with label flipping and oversampling
Researchers have detailed new adversarial attacks targeting federated learning setups for Generative Adversarial Networks (GANs). These attacks involve malicious clients manipulating data by flipping labels or oversampl…
-
New framework improves LLM alignment with heavy-tailed rewards
A new research paper introduces a tail-aware information-theoretic framework designed to improve the alignment of large language models (LLMs), particularly in scenarios involving heavy-tailed rewards. The framework uti…
-
New framework unifies analysis of generative diffusion models
A new research paper introduces a unified framework for analyzing generative diffusion models by examining the entropy production rate of the forward-reverse diffusion process. This approach allows for a precise decompo…
-
New method enhances Variational Autoencoder latent space optimization
Researchers have developed a new method for training Variational Autoencoders (VAEs) by treating the process as a soft-constrained optimization problem. This approach aims to improve both the encoding capacity of indivi…