Lin, Hongzhou, Jegelka, Stefanie

We demonstrate that a very deep ResNet with stacked modules that have one neuron per hidden layer and ReLU activation functions can uniformly approximate any Lebesgue integrable function in d dimensions, i.e. \ell_1(R^d). Due to the identity mapping inherent to ResNets, our network has alternating layers of dimension one and d. This stands in sharp contrast to fully connected networks, which are not universal approximators if their width is the input dimension d [21,11]. Hence, our result implies an increase in representational power for narrow deep networks by the ResNet architecture.

Lin, Hongzhou, Jegelka, Stefanie

Recently, deep learning has been playing a central role in machine learning research and applications. Since AlexNet, increasingly more advanced networks have achieved state-of-the-art performance in computer vision, speech recognition, language processing, game playing, medical imaging, and so on. In our previous studies, we proposed quadratic/second-order neurons and deep quadratic neural networks. In a quadratic neuron, the inner product of a vector of data and the corresponding weights in a conventional neuron is replaced with a quadratic function. The resultant second-order neuron enjoys an enhanced expressive capability over the conventional neuron. However, how quadratic neurons improve the expressing capability of a deep quadratic network has not been studied up to now, preferably in relation to that of a conventional neural network. In this paper, we ask three basic questions regarding the expressive capability of a quadratic network: (1) for the one-hidden-layer network structure, is there any function that a quadratic network can approximate much more efficiently than a conventional network? (2) for the same multi-layer network structure, is there any function that can be expressed by a quadratic network but cannot be expressed with conventional neurons in the same structure? (3) Does a quadratic network give a new insight into universal approximation? Our main contributions are the three theorems shedding light upon these three questions and demonstrating the merits of a quadratic network in terms of expressive efficiency, unique capability, and compact architecture respectively.

Lin, Hongzhou, Jegelka, Stefanie

We demonstrate that a very deep ResNet with stacked modules with one neuron per hidden layer and ReLU activation functions can uniformly approximate any Lebesgue integrable function in $d$ dimensions, i.e. $\ell_1(\mathbb{R}^d)$. Because of the identity mapping inherent to ResNets, our network has alternating layers of dimension one and $d$. This stands in sharp contrast to fully connected networks, which are not universal approximators if their width is the input dimension $d$ [Lu et al, 2017; Hanin and Sellke, 2017]. Hence, our result implies an increase in representational power for narrow deep networks by the ResNet architecture.

Sannai, Akiyoshi, Takai, Yuuki, Cordonnier, Matthieu

In this paper,we develop a theory of the relationship between permutation ($S_n$-) invariant/equivariant functions and deep neural networks. As a result, we prove an permutation invariant/equivariant version of the universal approximation theorem, i.e $S_n$-invariant/equivariant deep neural networks. The equivariant models are consist of stacking standard single-layer neural networks $Z_i:X \to Y$ for which every $Z_i$ is $S_n$-equivariant with respect to the actions of $S_n$ . The invariant models are consist of stacking equivariant models and standard single-layer neural networks $Z_i:X \to Y$ for which every $Z_i$ is $S_n$-invariant with respect to the actions of $S_n$ . These are universal approximators to $S_n$-invariant/equivariant functions. The above notation is mathematically natural generalization of the models in \cite{deepsets}. We also calculate the number of free parameters appeared in these models. As a result, the number of free parameters appeared in these models is much smaller than the one of the usual models. Hence, we conclude that although the free parameters of the invariant/equivarint models are exponentially fewer than the one of the usual models, the invariant/equivariant models can approximate the invariant/equivariant functions to arbitrary accuracy. This gives us an understanding of why the invariant/equivariant models designed in [Zaheer et al. 2018] work well.