Truncated Inference for Latent Variable Optimization Problems: Application to Robust Estimation and Learning
This work addresses efficiency issues in computer vision and machine learning for tasks involving latent variables, but it appears incremental as it builds on existing optimization frameworks.
The paper tackles the computational expense of optimization problems with latent variables by proposing two methods that dynamically adapt inference accuracy, eliminating the need to maintain latent variables. This approach is applied to large-scale robust estimation and learning energy-based models, though no concrete performance numbers are provided.
Optimization problems with an auxiliary latent variable structure in addition to the main model parameters occur frequently in computer vision and machine learning. The additional latent variables make the underlying optimization task expensive, either in terms of memory (by maintaining the latent variables), or in terms of runtime (repeated exact inference of latent variables). We aim to remove the need to maintain the latent variables and propose two formally justified methods, that dynamically adapt the required accuracy of latent variable inference. These methods have applications in large scale robust estimation and in learning energy-based models from labeled data.