The expressive power of neural networks is important for understanding deep learning. Most existing works consider this problem from the view of the depth of a network. In this paper, we study how width affects the expressiveness of neural networks. Classical results state that depth-bounded (e.g.

Critical in this context is the prior knowledge accumulated from related tasks.

However, such an approach is highly empirical and provides an indirect evaluation affected by external factors.

Geometric deep learning enables the encoding of physical symmetries in modeling 3D objects.

Previous works re-design a small network for IoT devices, and then compress the network size by mixed-precision quantization.