Energy Based Models
PulseAugur coverage of Energy Based Models — every cluster mentioning Energy Based Models across labs, papers, and developer communities, ranked by signal.
2 day(s) with sentiment data
-
New EBM-AE Framework Enhances Generative Modeling and Image Inpainting
Researchers have developed a novel cooperative framework that merges Energy-Based Models (EBMs) with Autoencoders (AEs) to enhance generative modeling. This EBM-AE framework employs an iterative process where an EBM ref…
-
New STEP framework improves human pose video anomaly detection
Researchers have developed a new framework called STEP (Score-Based Temporal Energy) for detecting anomalies in human pose videos. This method addresses a key challenge in existing approaches by using Principal Componen…
-
New method ReFP-AD enhances anomaly detection using foundation models
Researchers have developed ReFP-AD, a novel method for unified anomaly detection that leverages foundation models like DINOv2 for rich token representations. The technique addresses challenges in training Energy-Based M…
-
New Parallel Trajectory Tempering algorithm enhances Energy-Based Model training
Researchers have developed a new training algorithm called Parallel Trajectory Tempering (PTT) for Energy-Based Models (EBMs). This method addresses the issue of poor Markov Chain Monte Carlo mixing, which often hinders…
-
Normalizing Flows Prove Capable for Continuous Control in RL
Researchers have demonstrated that normalizing flows (NFs) are capable models for continuous control tasks in reinforcement learning (RL). Contrary to the prevailing belief that NFs lack sufficient expressivity, this pa…
-
New framework unifies statistical methods for energy-based models
Researchers have developed a unified framework that connects several statistical methods, including noise contrastive estimation (NCE), reverse logistic regression (RLR), multiple importance sampling (MIS), and bridge s…
-
ERAlign framework aligns GNN and LLM representations on text-attributed graphs
Researchers have developed ERAlign, a novel framework for aligning representations from Graph Neural Networks (GNNs) and Large Language Models (LLMs) on text-attributed graphs. This approach utilizes Energy-based Models…
-
Energy-Based Models Training Dynamics Analyzed
Researchers have analyzed the training dynamics of energy-based learning models, which are known for their non-convexity and potential for poor local optima. Their work introduces the concept of an "effective model" to …
-
MEFA framework enables memory-efficient full-gradient attacks for robust defense evaluation
Researchers have developed a new framework called MEFA (Memory Efficient Full-gradient Attacks) to improve the evaluation of adversarial defenses against machine learning models. This framework utilizes gradient checkpo…