One-Shot refers to one-shot learning, a specific scenario within the broader field of few-shot learning in machine learning. In one-shot learning, a model is trained with only a single example per class. This means that for each category or class that the model needs to recognize, only one instance with the corresponding label is provided during the training phase.


SpaceBeam: LiDAR-Driven One-Shot mmWave Beam Management

SpaceBeam: LiDAR-Driven One-Shot mmWave Beam Management mmWave 5G networks promise to enable a new generation of networked applications requiring a combination of high throughput and ultra-low latency. However, in practice, mmWave performance scales poorly for large numbers of users due to the significant overhead required to manage the highly-directional beams. We find that we can substantially reduce or eliminate this overhead by using out-of-band infrared measurements of the surrounding environment generated by a LiDAR sensor. To accomplish this, we develop a ray-tracing system that is robust to noise and other artifacts from the infrared sensor, create a method to estimate the reflection strength from sensor data, and finally apply this information to the multiuser beam selection process. We demonstrate that this approach reduces beam-selection overhead by over 95% in indoor multi-user scenarios, reducing network latency by over 80% and increasing throughput by over 2× in mobile scenarios.