# The Predictability Bias of Joint-Embedding Predictive Learning in Medical Imaging

arXiv: https://arxiv.org/abs/2610.13237 · HTML: https://mlsanity.com/paper/2610.13237/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 29 of its 41 references that are in the map.
- First posted on arXiv 9 Oct 2026.

## Builds on (most cited first, 29 of 29)

- [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://mlsanity.com/paper/2010.11929/) (2020)
- [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://mlsanity.com/paper/2104.09864/) (2021)
- [UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction](https://mlsanity.com/paper/1802.03426/) (2018)
- [Attention-based Deep Multiple Instance Learning](https://mlsanity.com/paper/1802.04712/) (2018)
- [Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecture](https://mlsanity.com/paper/2301.08243/) (2023)
- [V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning](https://mlsanity.com/paper/2506.09985/) (2025)
- [Going deeper with Image Transformers](https://mlsanity.com/paper/2103.17239/) (2021)
- [data2vec: A General Framework for Self-supervised Learning in Speech, Vision and Language](https://mlsanity.com/paper/2202.03555/) (2022)
- [TotalSegmentator: robust segmentation of 104 anatomical structures in CT images](https://mlsanity.com/paper/2208.05868/) (2022)
- [Revisiting Feature Prediction for Learning Visual Representations from Video](https://mlsanity.com/paper/2404.08471/) (2024)
- [Generalist Foundation Models from a Multimodal Dataset for 3D Computed Tomography](https://mlsanity.com/paper/2403.17834/) (2024)
- [Merlin: A Computed Tomography Vision-Language Foundation Model and Dataset](https://mlsanity.com/paper/2406.06512/) (2024)
- [Machine-Learning-Based Multiple Abnormality Prediction with Large-Scale Chest Computed Tomography Volumes](https://mlsanity.com/paper/2002.04752/) (2020)
- [RadGPT: Constructing 3D Image-Text Tumor Datasets](https://mlsanity.com/paper/2501.04678/) (2025)
- [Joint Embedding vs Reconstruction: Provable Benefits of Latent Space Prediction for Self Supervised Learning](https://mlsanity.com/paper/2505.12477/) (2025)
- [How JEPA Avoids Noisy Features: The Implicit Bias of Deep Linear Self Distillation Networks](https://mlsanity.com/paper/2407.03475/) (2024)
- [Joint Embedding Predictive Architectures Focus on Slow Features](https://mlsanity.com/paper/2211.10831/) (2022)
- [Comprehensive language-image pre-training for 3D medical image understanding](https://mlsanity.com/paper/2510.15042/) (2025)
- [CheXWorld: Exploring Image World Modeling for Radiograph Representation Learning](https://mlsanity.com/paper/2504.13820/) (2025)
- [ReXGroundingCT: A 3D Chest CT Dataset for Segmentation of Findings from Free-Text Reports](https://mlsanity.com/paper/2507.22030/) (2025)
- [Learning Sparse Latent Predictive Foundation Model for Multimodal Neuroimaging](https://mlsanity.com/paper/2606.14957/) (2026)
- [Pretext Matters: An Empirical Study of SSL Methods in Medical Imaging](https://mlsanity.com/paper/2603.22649/) (2026)
- [Learning Robust Visual Features in Computed Tomography Enables Efficient Transfer Learning for Clinical Tasks](https://mlsanity.com/paper/2604.04133/) (2026)
- [US-JEPA: A Joint Embedding Predictive Architecture for Ultrasound](https://mlsanity.com/paper/2602.19322/) (2026)
- [Predictive Objectives Discard Exogenous Control-Relevant Features: A Controlled Mechanistic Study](https://mlsanity.com/paper/2606.30068/) (2026)
- [The JEPA Paradox in Language: The Geometry of Linguistic Alternatives](https://mlsanity.com/paper/2607.23531/) (2026)
- [RadJEPA: Radiology Encoder for Chest X-Rays via Joint Embedding Predictive Architecture](https://mlsanity.com/paper/2601.15891/) (2026)
- [Rad-JEPA 3D: Radiology Joint-Embedding Predictive Model for 3D Computed Tomography](https://mlsanity.com/paper/2607.26196/) (2026)
- [Big, Bright, or Invisible: A Frozen-Feature Benchmark of 3D CT Foundation Models](https://mlsanity.com/paper/2608.05960/) (2026)

## Abstract (arXiv)

Joint-embedding predictive architectures (JEPAs) learn by predicting teacher representations of masked regions from visible context. Point prediction summarizes the conditional target distribution and cannot reproduce variation that the context does not resolve. We call this constraint predictability bias. In computed tomography (CT), recurrent anatomy is predictable from context, whereas the sporadic, localized pathology that determines clinical interpretation often is not. We introduce TAP-JEPA, a ViT-B/3D trained on 113,611 CT volumes that is competitive with state-of-the-art encoders (COLIPRI, Merlin, VoxelFM) in scan-level classification. We then audit the learned predictive pathway by replaying the exact masking, teacher, and predictor computations used during training. Prediction lowers the mean Skill (prevalence-normalized average precision) from 0.388 to 0.271 for pathology, but only from 0.921 to 0.873 for anatomy. Along the teacher direction, separating pathology from matched control regions in the same organ, prediction moves pathology 34.7% toward the controls and controls 13.6% toward pathology, with larger shifts farther from visible context. Replacing visible matched-control representations increases surrounding-target prediction loss more than fully replacing pathology representations across four pathology categories. Downstream, removing all final-layer pathology tokens reduces the decision margin by at most 0.45% in four of seven scan-level probes. Together, these results show that contextual predictability is not a sufficient proxy for clinical relevance in medical representation learning. Training objectives should therefore retain diagnostic features regardless of their predictability from surrounding anatomy.