Capacity Bounds for Hyperbolic Neural Network Representations of Latent Tree Structures
Kratsios, Anastasis, Hong, Ruiyang, Borde, Haitz Sáez de Ocáriz
–arXiv.org Artificial Intelligence
We study the representation capacity of deep hyperbolic neural networks (HNNs) with a ReLU activation function. We establish the first proof that HNNs can $\varepsilon$-isometrically embed any finite weighted tree into a hyperbolic space of dimension $d$ at least equal to $2$ with prescribed sectional curvature $\kappa<0$, for any $\varepsilon> 1$ (where $\varepsilon=1$ being optimal). We establish rigorous upper bounds for the network complexity on an HNN implementing the embedding. We find that the network complexity of HNN implementing the graph representation is independent of the representation fidelity/distortion. We contrast this result against our lower bounds on distortion which any ReLU multi-layer perceptron (MLP) must exert when embedding a tree with $L>2^d$ leaves into a $d$-dimensional Euclidean space, which we show at least $\Omega(L^{1/d})$; independently of the depth, width, and (possibly discontinuous) activation function defining the MLP.
arXiv.org Artificial Intelligence
Aug-17-2023
- Country:
- North America
- United States
- New York (0.04)
- Maryland > Baltimore (0.04)
- Vermont > Chittenden County
- Burlington (0.04)
- Canada > Ontario
- Hamilton (0.04)
- United States
- Europe > United Kingdom
- England > Oxfordshire > Oxford (0.04)
- Asia
- Middle East > Israel (0.04)
- China > Hong Kong (0.04)
- North America
- Genre:
- Research Report (0.81)
- Technology: