Randall Balestriero

VIScore reveals that the reachability and capacity of predictors are crucial for planning success, outperforming traditional metrics in predictive accuracy.

LeDXA reveals that self-supervised learning can unlock critical health insights from DXA scans, outperforming traditional metrics in predicting disease risk with far fewer resources.

Non-contrastive pretraining can yield dense semantic features that outperform traditional contrastive methods in vision-language tasks.

LeNEPA achieves faster representation learning without relying on data augmentation, outperforming traditional methods in both efficiency and adaptability across diverse datasets.

Aionoscope reveals that while many time-series models can identify basic signal components, they often fail to capture essential timing and amplitude details crucial for effective debugging.

Transition effects can be disentangled into reusable primitives, leading to superior policy learning even in complex, ambiguous environments.

OctoSense outperforms conventional image-only models in multimodal robot perception, achieving robust performance even under degraded sensory conditions.

Achieving robust zero-shot sim-to-real transfer for quadrotors, this work redefines the boundaries of long-horizon prediction in robotic control.

Action-aligned representations can be achieved without complex training methods, enabling robust planning in dynamic environments.

A new phase diagram reveals that cross-modal training can be actively harmful in certain contexts, guiding practitioners to choose the right approach before training.

VISReg not only stabilizes embedding training but also achieves state-of-the-art performance with a fraction of the data used by competing methods.

Straightening latent space trajectories with a simple curvature regularizer dramatically improves the stability and success of gradient-based planning in world models.