Geometrization of deep networks for the interpretability of deep learning systems

Abstract

How to understand deep learning systems remains an open problem. In this paper we propose that the answer may lie in the geometrization of deep networks. Geometrization is a bridge to connect physics, geometry, deep network and quantum computation and this may result in a new scheme to reveal the rule of the physical world. By comparing the geometry of image matching and deep networks, we show that geometrization of deep networks can be used to understand existing deep learning systems and it may also help to solve the interpretability problem of deep learning systems.

arXiv comment9 pages, draft version

References in corpus (15)

• Understanding Black-box Predictions via Influence Functions
• Quantum Computation as Geometry
• Neural Ordinary Differential Equations
• Fisher-Rao Metric, Geometry, and Complexity of Neural Networks
• Exact holographic mapping and emergent space-time geometry
• Algorithms for tensor network renormalization
• The momentum map representation of images
• Holography as deep learning
• Frustration free gapless Hamiltonians for Matrix Product States
• Deep learning and the renormalization group
• Geometric formulation of quantum mechanics
• How deep learning works --The geometry of deep learning
• Geodesic Distance in Fisher Information Space and Holographic Entropy Formula
• Emergent General Relativity from Fisher Information Metric
• Adversarial Noise Layer: Regularize Neural Network By Adding Noise

Cited by in corpus (6)

• Turbulence forecasting via Neural ODE
• Understanding over-parameterized deep networks by geometrization
• Deep network as memory space: complexity, generalization, disentangled representation and interpretability
• Gauge theory and twins paradox of disentangled representations
• Self-regularizing restricted Boltzmann machines
• Problems of representation of electrocardiograms in convolutional neural networks