Hugo Tabanelli, Pierre Mergny, Lenka Zdeborova, Florent Krzakala
The paper explores multi-modal learning using spiked matrix-tensor models, revealing that sequential learning strategies can overcome computational challenges and achieve optimal signal recovery.
This research investigates how to recover signals from two different but related data sources: a matrix and a tensor, both containing noise and sharing some underlying structure. Typically, working with tensors is computationally difficult, especially when the data is noisy. However, the study finds that by first processing the matrix data and then using it to help with the tensor data, we can effectively recover the signals. This approach outperforms traditional methods that try to process both data types simultaneously, which often fail due to the complexity of the tensor data.
