Authors: Randall Balestriero , Yann LeCun - arXiv 2025 cited by 79
Chimère Ω is a blueprint for a local-first LLM runtime that allocates computation across four modes (fast prediction, block-diffusion refinement, symbolic verification, latent-space planning) via a scheduler driven by token-level residual entropy rather than fixed architectural choices. The blueprint integrates verified Q4 2025–Q2 2026 state of the art: LeJEPA/SIGReg (Balestriero & LeCun, arXiv:2511.08544), hybrid Gated-DeltaNet + sparse-attention backbones (Qwen3-Next, Kimi-Linear), test-time memory optimisation. The thesis is that token-level residual entropy is a sufficient signal to schedule across modes and yield aggregate quality gains within a fixed local compute budget. Companion design document to the ECI framework (DOI 10.5281/zenodo.19708889).
✨ Checking sign-in… PDF Cited by View BibTeX Hide BibTeX View BibTeX Cite
Representation Learning with Contrastive… Representation Learning with Contrastive Predictive Coding Learning deep representations by… Learning deep representations by mutual information estimation and maximization VICReg: Variance-Invariance-Cov… VICReg: Variance-Invariance-Covariance Regularization for Self-Supervised Learning Understanding Dimensional Collapse in… Understanding Dimensional Collapse in Contrastive Self-supervised Learning The Hidden Uniform Cluster Prior in… The Hidden Uniform Cluster Prior in Self-Supervised Learning A Cookbook of Self-Supervised Learning A Cookbook of Self-Supervised Learning Vision Transformers Need Registers Vision Transformers Need Registers DINOv2: Learning Robust Visual Features without… DINOv2: Learning Robust Visual Features without Supervision Learning by Reconstruction Produces… Learning by Reconstruction Produces Uninformative Features For Perception Gaussian Embeddings: How JEPAs Secretly Learn… Gaussian Embeddings: How JEPAs Secretly Learn Your Data Density Joint Embedding vs Reconstruction: Provabl… Joint Embedding vs Reconstruction: Provable Benefits of Latent Space Prediction for Self Supervised Learning DINOv3 DINOv3 VL-JEPA: Joint Embedding Predictive Architecture… VL-JEPA: Joint Embedding Predictive Architecture for Vision-language KerJEPA: Kernel Discrepancies for… KerJEPA: Kernel Discrepancies for Euclidean Self-Supervised Learning LeWorldModel: Stable End-to-End… LeWorldModel: Stable End-to-End Joint-Embedding Predictive Architecture from Pixels V-JEPA 2.1: Unlocking Dense Features in Video… V-JEPA 2.1: Unlocking Dense Features in Video Self-Supervised Learning Temporal Straightening for Latent Planning Temporal Straightening for Latent Planning IConE: Batch Independent Collapse Prevention for… IConE: Batch Independent Collapse Prevention for Self-Supervised Representation Learning HQ-JEPA: Hybrid Quantum Joint-Embedding… HQ-JEPA: Hybrid Quantum Joint-Embedding Predictive Architecture for Cross-Modal Remote Sensing Representation Learning Video Generation Models in Robotics -… Video Generation Models in Robotics - Applications, Research Challenges, Future Directions Laya: A LeJEPA Approach to EEG via Latent… Laya: A LeJEPA Approach to EEG via Latent Prediction over Reconstruction Predictive but Not Plannable: RC-aux for… Predictive but Not Plannable: RC-aux for Latent World Models JEPA-DNA: Grounding Genomic Foundation… JEPA-DNA: Grounding Genomic Foundation Models through Joint-Embedding Predictive Architectures Pretext Matters: An Empirical Study of SSL… Pretext Matters: An Empirical Study of SSL Methods in Medical Imaging LeJEPA: Provable and Scalable Self-Supervise… LeJEPA: Provable and Scalable Self-Supervised Learning Without the Heuristics Earlier references Focus paper Citing papers Older Newer Click a node to pin it, click the empty canvas to go back to this paper, or hover to preview. Open a node’s page from its title.
