Beta DVBF: Learning State-Space Models for Control from High Dimensional Observations
This work addresses a core challenge in sequential decision-making for applications across domains, but appears incremental as it builds on existing latent-variable methods.
The paper tackled the problem of learning state-space models from high-dimensional images for control, identifying limitations in existing latent-variable methods with low-resolution images, and proposed solutions to address dimensionality discrepancies, resulting in improved handling of high-resolution observations.
Learning a model of dynamics from high-dimensional images can be a core ingredient for success in many applications across different domains, especially in sequential decision making. However, currently prevailing methods based on latent-variable models are limited to working with low resolution images only. In this work, we show that some of the issues with using high-dimensional observations arise from the discrepancy between the dimensionality of the latent and observable space, and propose solutions to overcome them.