Neural Embedding Compression For Efficient Multi-Task Earth Observation Modelling
This addresses resource efficiency for earth observation data consumers, though it is incremental as it builds on existing foundation models and compression techniques.
The paper tackles the problem of high transfer and storage costs for earth observation data by introducing Neural Embedding Compression (NEC), which compresses embeddings from foundation models for multi-task use, achieving similar accuracy with a 75% to 90% reduction in data and only a 5% performance drop at 99.7% compression.
As repositories of large scale data in earth observation (EO) have grown, so have transfer and storage costs for model training and inference, expending significant resources. We introduce Neural Embedding Compression (NEC), based on the transfer of compressed embeddings to data consumers instead of raw data. We adapt foundation models (FM) through learned neural compression to generate multi-task embeddings while navigating the tradeoff between compression rate and embedding utility. We update only a small fraction of the FM parameters (10%) for a short training period (1% of the iterations of pre-training). We evaluate NEC on two EO tasks: scene classification and semantic segmentation. Compared with applying traditional compression to the raw data, NEC achieves similar accuracy with a 75% to 90% reduction in data. Even at 99.7% compression, performance drops by only 5% on the scene classification task. Overall, NEC is a data-efficient yet performant approach for multi-task EO modelling.