Knowledge boosting: Model collaboration during low-latency inference
Knowledge boosting is a novel technique that allows a large model running remotely to operate on time-delayed input during inference, while boosting small model performance running locally. This technique can benefit real-time applications across various domains such as robotics, self-driving vehicles, and audio and video processing.
Paper: Knowledge boosting during low-latency inference, Interspeech 2024
Project page: https://knowledgeboosting.cs.washingt...