**6. Conclusion**

In this paper, we proposed a deep multi-modal approach to fuse data through recovering the underlying subspaces of data observations from data corrupted by noise to scale to complex data scenarios. DRoGSuRe provides a natural way to fuse multi-modal data by employing the self-representation matrix as an embedding for each data modality. Experimental results show a significant improvement for DRoGSuRe over DMSC under different types of potential limitations and provides robustness with limited sensing modalities. We also proposed the concatenated CNNs model, which can work better for different multi-modal data structures.
