During the award ceremony of the prestigious 18th IEEE/ACM International Conference on Utility and Cloud Computing (UCC 2025), our paper entitled “FakeInf: Selective Deep Neural Network Inference for Latency and Energy-Aware Model Serving Pipelines” won the Best Paper Award!
In a nutshell, our work introduces FakeInf, a lightweight decision-making framework designed for EdgeAI video analytics. FakeInf intelligently tracks data volatility and uses probabilistic reasoning to decide whether to run a full Deep Learning model inference or “fake it” by relying on low-cost statistical estimations. With FakeInf, the pipeline selectively bypasses unnecessary computations during periods of low volatility, significantly reducing application latency, network traffic, and energy consumption by up to 72% while maintaining user-desired accuracy and Quality-of-Service.