LeJEPA
PulseAugur coverage of LeJEPA — every cluster mentioning LeJEPA across labs, papers, and developer communities, ranked by signal.
5 day(s) with sentiment data
-
New S-JEPA model introduces stochasticity for diverse image completions
Researchers have introduced S-JEPA, a novel stochastic generalization of Joint-Embedding Predictive Architectures (JEPAs) for masked image prediction. Unlike previous deterministic approaches that estimate a single comp…
-
Event-stream selection methods evaluated for sensitivity and performance
A new research paper published on arXiv details an evaluation of event-stream candidate selection methods, focusing on their sensitivity to readout and threshold calibration. The study tested various encoders, including…
-
Self-supervised learning monitoring fails to detect critical encoder degradation
Researchers have identified a critical flaw in how self-supervised learning models are monitored. Using LeJEPA encoders trained on synthetic images with known generating factors, they observed that in nine out of ten in…
-
New LocaLeJEPA framework improves self-supervised learning with geometric priors
Researchers have introduced LocaLeJEPA, a new framework for self-supervised learning that addresses representation collapse issues in Joint-Embedding Predictive Architectures (JEPA). Unlike previous methods tied to Eucl…
-
New generative approach improves latent tokenization for AI models
Researchers have developed a novel self-supervised representation learning method designed for scalable latent tokenization in generative modeling. This approach uses a representation loss that factorizes mutual informa…
-
New fMRI foundation model CoFlu-Brain shows strong transfer learning
Researchers have developed CoFlu-Brain, a new foundation model for functional magnetic resonance imaging (fMRI) data. Unlike traditional methods that use a single correlation matrix, CoFlu-Brain organizes framewise cofl…
-
New Shared Gaussianization method analyzes contrastive learning
Researchers have introduced Shared Gaussianization (SG), a novel method for analyzing contrastive learning techniques. SG acts as a Gaussianity test on normalized views of data, detecting both misalignment and non-unifo…
-
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 CARNIVAL model enhances protein annotation in cryo-ET using simulated data
Researchers have developed a new self-supervised learning framework called CARNIVAL, designed for protein annotation in cryo-electron tomography (cryo-ET) volumes. This model leverages simulated data and incorporates in…
-
LeJEPA architecture adapted for molecular graph encoders
Researchers have adapted LeJEPA, a self-supervised pretraining architecture, for molecular graph neural networks to assess its impact on molecular property prediction. While pretraining enhances learned representations,…
-
New Transformer Architecture Optimizes Self-Supervised Learning
Researchers have developed an attention-only white-box Transformer model by integrating the LeJEPA self-supervised learning framework with optimization algorithms. This approach optimizes the sparse rate reduction objec…
-
New techniques aim to improve LLM KV-cache efficiency and accuracy
Researchers are exploring novel methods to improve the efficiency of large language models by optimizing their KV cache, a component crucial for inference but known for its high memory and bandwidth demands. One approac…
-
AV-JEPA model advances audio-visual self-supervised learning
Researchers have introduced AV-JEPA, a new self-supervised learning model that extends LeJEPA to handle both audio and visual data. This model utilizes an early-fusion Vision Transformer and modality dropout for masking…
-
New Kepler-Encoder-v0.1 model fuses robot state with vision
Researchers have developed Kepler-Encoder-v0.1, a novel multimodal embedding model designed for robots. This model integrates visual data with proprioception and force/torque sensor information into a unified latent spa…
-
AI advances symbolic music generation and analysis with new frameworks · 6 sources tracked
Researchers are developing new AI frameworks for symbolic music generation and analysis. One approach, MIDI-RAE-JEPA, uses self-supervised learning with a Swin Transformer V2 encoder to capture hierarchical musical stru…
-
Object-centric LeJEPA improves image representation learning with SAM
Researchers have developed an object-centric version of LeJEPA, a self-supervised learning method for image encoders. By leveraging object masks generated by SAM, this new approach aims to improve data efficiency compar…
-
New statistical regularizers enhance self-supervised learning stability
Researchers have introduced a new family of statistical regularizers for Self-Supervised Learning (SSL) that aim to improve representation collapse prevention. The proposed methods analytically integrate random projecti…
-
New DALE-CT models achieve near SOTA in CT abnormality detection
Researchers have developed DALE-CT, a new family of 2D foundation models for processing computed tomography (CT) data. Built from scratch using a self-supervised learning approach called LeJEPA, DALE-CT incorporates a n…
-
Self-Soupervision enables model soups from unlabeled data
Researchers have developed a new method called Self-Soupervision, which allows for the creation of "model soups" using self-supervised learning (SSL) instead of traditional supervised learning. This technique enables th…
-
New JEPA Architectures Achieve Stable End-to-End Training from Pixels
Researchers have developed LeWorldModel (LeWM), a novel Joint Embedding Predictive Architecture (JEPA) that stably trains end-to-end from raw pixels. Unlike previous fragile JEPA methods, LeWM uses only two loss terms a…