Evaluation Of Parameter-Scaling For Efficient Deep Learning On Small Satellites