An arXiv preprint presents DS-Net, a dynamic split network for edge-cloud collaborative inference. The abstract frames collaborative inference as a critical methodology in edge intelligence and notes that existing methods operate by splitting a deep neural network (DNN) between edge devices and cloud servers.
The abstract then begins to describe a limitation of these methods with "However, these…" but the text cuts off. As a result, the specific shortcomings addressed by DS-Net, and the details of its dynamic splitting mechanism, are not available from the source.
The paper's title and the phrase "dynamic split network" suggest that the contribution lies in making the split adaptive rather than fixed, but this is an inference from the name, not a claim made in the abstract. Readers interested in the full details would need to consult the complete paper.