Foundation models like V-JEPA are increasingly deployed across a heterogeneous edge-cloud continuum spanning smart glasses, phones, robots, and servers. However, the paper notes that these models are typically trained as rigid, fixed-capacity networks bound to a single deployment context.
The new work, titled "EV-JEPA: Elastic V-JEPA for Any Device and Cross-Device Inference," addresses this limitation. While the abstract is truncated, the title and opening sentences indicate a focus on making V-JEPA elastic, enabling it to operate across devices and support cross-device inference.
Because the source is a single arXiv abstract, details of the proposed method are not yet available. The significance lies in the shift from static, device-specific models to flexible ones that can span the edge-cloud continuum.