PROPOSEDPOD 06: Systems, HPC & MLOps
Heterogeneous Edge-to-Server Model Offloading Benchmarks
Objective & Abstract
Profiling network latency thresholds for splitting neural pipeline layers between mobile nodes and local GPU servers over varying bandwidth constraints.
Investigated Methodologies
- Layer cut-point optimization
- gRPC tensor serialization
- Network jitter profiling
Datasets & Benchmarking Corpora
Simulated Edge Network Latency Profiles
All training and evaluation datasets undergo ethics audit and licensing compliance per AIAC Research Council standards.