Joint-Embedding Predictive Architectures
PulseAugur coverage of Joint-Embedding Predictive Architectures — every cluster mentioning Joint-Embedding Predictive Architectures across labs, papers, and developer communities, ranked by signal.
7 day(s) with sentiment data
-
New research questions JEPA effectiveness for EHR representation learning
A new research paper explores the effectiveness of Joint-Embedding Predictive Architectures (JEPAs) for learning representations from longitudinal electronic health records (EHRs). Despite stable training, the JEPA form…
-
New G-JEPA model enhances latent world models with temporal context
Researchers have introduced G-JEPA, a novel approach to latent world models that enhances physical grounding by incorporating temporal context. This method addresses limitations in existing JEPA-based models, specifical…
-
Slot-JEPA introduces object-structured prediction for latent world models
Researchers have introduced Slot-JEPA, a novel latent world model that decomposes dense representations into stable semantic slots and time-varying spatial masks. This approach aims to overcome challenges in existing Jo…
-
Metro-WM framework improves AI long-horizon planning with realisable sub-goals
Researchers have introduced Metro-WM, a novel hierarchical framework designed to improve long-horizon latent planning in AI systems. Unlike previous methods that generate ungrounded latent vectors for sub-goals, Metro-W…
-
New research explores latent world models for AI planning and prediction · 2 sources tracked
Two new research papers explore methods for improving latent world models, which are crucial for AI agents to understand cause and effect. The first paper, "Learning Intrinsic Dynamics in Latent World Models," introduce…
-
New research highlights predictability bias in medical imaging AI
A new research paper introduces the concept of "predictability bias" in joint-embedding predictive architectures (JEPAs) used for medical imaging. The study found that while JEPAs can accurately predict anatomical struc…
-
DSReg enables provable recovery of world latents without reconstruction
Researchers have developed DSReg (Dependency-Sparsity Regularization), a novel method that enables the provable recovery of individual latent variables from world models without relying on reconstruction, auxiliary supe…
-
New research advances robot world models for efficient planning and control · 10 sources tracked
Recent research explores advancements in world action models (WAMs) for robotics and agent planning. WAM-Cache and WAMJET focus on accelerating inference by optimizing key-value reuse and employing agentic harnesses for…
-
JEPA-Anything framework enables domain-agnostic world modeling
Researchers have introduced JEPA-Anything, a new framework designed for domain-agnostic world modeling. This approach extends joint-embedding predictive architectures by employing orthogonal predictive factorization (OP…
-
New Med-D-JEPA model advances 3D brain MRI synthesis and classification
Researchers have adapted Joint-Embedding Predictive Architectures (JEPAs), a framework primarily used for self-supervised representation learning, for 3D brain MRI synthesis. The proposed Med-D-JEPA model combines maske…
-
Orthogonal JEPA framework enhances latent world models with factorized states
Researchers have introduced Orthogonal JEPA, a novel latent world-modeling framework designed to improve prediction and reasoning capabilities. This method addresses limitations in standard Joint-Embedding Predictive Ar…
-
New JEPA method uses contrastive inverse dynamics to improve world models
Researchers have developed a new method called Action-Contrastive Masked Transition Modeling (AC-MTM) for Joint-Embedding Predictive Architectures (JEPAs) that addresses the issue of trivial solutions in world models. U…
-
New diagnostic tool assesses AI world models for visual perturbation resilience
Researchers have developed a new diagnostic tool called Action-Conditioned Predictive Consistency (ACPC) to evaluate world models within Joint-embedding predictive architectures (JEPAs). ACPC measures how much a world m…
-
UniJEPA unifies image and video visual world modeling
Researchers have introduced UniJEPA, a novel unified architecture for self-supervised visual world modeling. This new framework integrates both image-level photometric prediction and video-level temporal prediction into…
-
JEPA models face challenges with language's conditional structure
A new paper explores the challenges of applying Joint-Embedding Predictive Architectures (JEPAs) to language processing, contrasting their effectiveness in image and audio domains with their limitations in text. The res…
-
Temporal-Distance JEPA enhances world model predictive control
Researchers have introduced Temporal-Distance JEPA (TD-JEPA), a novel approach to representation learning for latent world model predictive control. This method enhances Joint-Embedding Predictive Architectures (JEPAs) …
-
New regularization method tackles bias in self-supervised learning
Researchers have introduced Unbiased Open World Regularization (UOWReg), a novel framework designed to mitigate biases in self-supervised learning (SSL) and Joint-Embedding Predictive Architectures (JEPAs). Unlike previ…
-
Yann LeCun proposes JEPA to enhance LLMs' physical world understanding
Yann LeCun, in a recent interview, discussed the limitations of current Large Language Models (LLMs) in truly understanding the physical world, contrasting their ability to answer questions with their inability to perfo…
-
MorphologyFM model learns from ECG and pulse oximetry waveforms
Researchers have developed MorphologyFM, a novel foundation model designed to learn representations from electrocardiogram (ECG) and pulse oximetry (SpO2) waveforms. Unlike previous methods that focus on reconstruction …
-
Qantara JEPA model enables multi-paradigm control from single checkpoint
Researchers have introduced Qantara, a novel Joint-Embedding Predictive Architecture (JEPA) that enables a single model checkpoint to support multiple inference paradigms for control from raw pixels. Unlike previous JEP…