Speaker
Description
TorchSig, an open-source signal processing machine learning (ML) library, is expanding into a reproducible Radio Frequency Machine Learning (RFML) framework with its most recent v2.x releases. We present an interactive workshop that covers: v2.x updates, new libraries torchsig-models and torchsig-gui, and new geolocation capabilities.
TorchSig v2.x updates include a complete restructure of the signal generation process, hierarchical metadata relationships, improved dataset writing, new dataset utilities, and structured signal generation. A significant new feature of TorchSig is its new geolocation tools–leveraging TorchSig's data generation for geospatial emitter simulation. We also showcase two new libraries, torchsig-models and torchsig-gui. TorchSig Models lets users create a full RFML pipeline, from creating synthetic datasets, augmenting and transforming data, to training a PyTorch-based detector. Additionally, we will demonstrate the TorchSig GUI, which provides an easy-to-use interface for creating TorchSig datasets.
Overall, the workshop aims to showcase ways RF engineers, ML practitioners, and researchers can use TorchSig for reproducible data generation, model development, and evaluation.
| Talk Length | N/A |
|---|---|
| Link to Open Source Code | https://github.com/TorchDSP/torchsig |
| Acknowledge | Acknowledge In-Person |