A Unifying Tensor View for Lightweight CNNs
This work addresses the problem of designing efficient and interpretable lightweight CNNs for resource-constrained applications, offering incremental improvements in compression and accuracy.
The paper tackles the lack of geometric intuition in decomposing convolutional kernels for lightweight CNNs by introducing a unifying tensor view that connects tensor approximations to efficient modules, discovering a pointwise-depthwise-pointwise configuration and enabling shift layer pruning that achieves nearly 50% compression with less than 1% accuracy drop for ShiftResNet.
Despite the decomposition of convolutional kernels for lightweight CNNs being well studied, existing works that rely on tensor network diagrams or hyperdimensional abstraction lack geometry intuition. This work devises a new perspective by linking a 3D-reshaped kernel tensor to its various slice-wise and rank-1 decompositions, permitting a straightforward connection between various tensor approximations and efficient CNN modules. Specifically, it is discovered that a pointwise-depthwise-pointwise (PDP) configuration constitutes a viable construct for lightweight CNNs. Moreover, a novel link to the latest ShiftNet is established, inspiring a first-ever shift layer pruning that achieves nearly 50% compression with < 1% drop in accuracy for ShiftResNet.