An attempt to generate new bridge types from latent space of energy-based model

Hongjun Zhang

2024-01-31

Abstract:Use energy-based model for bridge-type innovation. The loss function is explained by the game theory, the logic is clear and the formula is simple and clear. Thus avoid the use of maximum likelihood estimation to explain the loss function and eliminate the need for Monte Carlo methods to solve the normalized denominator. Assuming that the bridge-type population follows a Boltzmann distribution, a neural network is constructed to represent the energy function. Use Langevin dynamics technology to generate a new sample with low energy value, thus a generative model of bridge-type based on energy is established. Train energy function on symmetric structured image dataset of three span beam bridge, arch bridge, cable-stayed bridge, and suspension bridge to accurately calculate the energy values of real and fake samples. Sampling from latent space, using gradient descent algorithm, the energy function transforms the sampling points into low energy score samples, thereby generating new bridge types different from the dataset. Due to unstable and slow training in this attempt, the possibility of generating new bridge types is rare and the image definition of generated images is low.

Machine Learning,Artificial Intelligence

What problem does this paper attempt to address?

This paper attempts to use the Energy - Based Model (EBM) to generate new bridge types from the latent space. Specifically, the paper aims to represent the energy function by constructing a neural network and use Langevin dynamics technology to generate new samples with low energy values, thereby establishing an energy - based bridge - type generation model. During the training process, this model is trained on a symmetric - structure - image data set, including 3 - span beam bridges, arch bridges, cable - stayed bridges and suspension bridges, in order to accurately calculate the energy values of real and fake samples. By sampling from the latent space and using the gradient - descent algorithm, the energy function converts sampling points into samples with low - energy scores, thus generating new bridge types different from those in the data set.

However, due to the instability and slowness during the training process, the probability of generating new bridge types is low, and the clarity of the generated images is also low. Despite these problems, the author still obtained three technically - feasible new bridge types from the sampling buffer through manual screening, which are completely different from the bridge types in the data set. This shows the innovative ability of the model, but also points out the necessity of further optimization, especially when the hardware conditions are limited.