AirMoE: Statistic-Augmented Over-the-Air MoE for Collaborative Intelligence
2026-07-21T08:52:23Z•f9448b19e46f5886bc7345817b5bb817b573f3b452a2fbc27df4f2a4870ca020
GPU inferenceI/O resource managementJensen–Shannon divergenceKubernetesMixture-of-ExpertsOCI imagesOracle Exadatacold-startcolocationdigest pinningdistributed scheduleredge inferencefeature-prototypeshash verificationkernel interferencemodel deliverymodel signingmodel supply-chain integrityorchestrationover-the-air aggregationrate limitingserverless LLMssession-aware servingstate management/runtime controlwireless ML
What happened
Collection of recent distributed-systems and ML-serving research. Highlights: AirMoE — a statistic-augmented, over-the-air Mixture-of-Experts design that reduces uplink bandwidth via prototype statistics and aggregates expert outputs via waveform superposition with channel-aware power control; a Kubernetes model-delivery study that implements OCI-based image volumes and an initializer path (oci+native:// and oci+fetch://), measures large-model delivery (2–140 GB) showing node-cached OCI is far faster than object-store downloads, and recommends digest-pinning and OpenSSF model-signing with in‑t
Why it matters
A reviewed impact interpretation has not been published for this record.
Evidence and limitations
- Source ID
- arxiv_cs_dc
- Record identifier
- f9448b19e46f5886bc7345817b5bb817b573f3b452a2fbc27df4f2a4870ca020
- Enrichment time
- 2026-07-21T08:52:23Z
- AI-assisted enrichment
- Yes
This record may overlap with other records. Its enrichment can be incomplete or wrong, and machine assistance was used. Validate consequential decisions against the linked source and your own environment.