HeteroSense-FL: A Multimodal Simulation Testbed for Modality-Heterogeneous Federated Learning

## HeteroSense-FL v1.0.0 This release provides the first stable version of HeteroSense-FL, a multimodal simulation testbed for modality-heterogeneous federated learning research. ### Key features - Structured multimodal sensor simulation: - 3D LiDAR point clouds - 2D bed pressure maps - Configurable N-client modality heterogeneity (ClientFactory) - Temporal window sampling interface for plug-in encoder development (TemporalWindowSampler) - Automated observation integrity checks (V1–V4) - Reproducible reference benchmarks via: `heterosense-benchmark` ### Reproducibility The results reported in the SoftwareX paper (Table 3) can be reproduced with a single command: `heterosense-benchmark` Conditions: - seeds: {42, 123, 7} - n_steps: 3000 - FL rounds: 3 - encoder: numpy-based TinyMLP ### Installation `pip install heterosense-fl` ### Platforms - Python 3.9–3.12 - Linux, macOS, Windows ### License MIT License ### Citation A Zenodo DOI will be assigned to this release for citation. --- This release is archived for reproducibility of the SoftwareX publication.