Distributed Convolutional Sparse Coding via Message Passing Interface (MPI)

Convolutional sparse coding methods focus on building representations of time signals as sparse and linear combinations of shifted patterns. These techniques have proven to be useful when dealing with signals (such as ECG or images) which are composed of several characteristic patterns ([1, 2, 3]). For this type of signals, the shapes and positions of these templates are crucial for their study (segmentation, classification, etc...). Depending on the context, the dictionary can either be fixed analytically, or learned from the data, using an alternated minimization of the representation cost over the dictionary and the representation.