Amortized Probabilistic Conditioning for Optimization, Simulation and Inference
This work addresses a key limitation in probabilistic meta-learning for researchers and practitioners, offering a more flexible tool for tasks requiring latent variable manipulation, though it is incremental as it builds on existing transformer-based neural processes.
The paper tackles the problem of inflexible conditioning on latent variables in amortized meta-learning models, introducing the Amortized Conditioning Engine (ACE) to enable explicit representation and runtime conditioning, and demonstrates its performance in tasks like image completion and Bayesian optimization.
Amortized meta-learning methods based on pre-training have propelled fields like natural language processing and vision. Transformer-based neural processes and their variants are leading models for probabilistic meta-learning with a tractable objective. Often trained on synthetic data, these models implicitly capture essential latent information in the data-generation process. However, existing methods do not allow users to flexibly inject (condition on) and extract (predict) this probabilistic latent information at runtime, which is key to many tasks. We introduce the Amortized Conditioning Engine (ACE), a new transformer-based meta-learning model that explicitly represents latent variables of interest. ACE affords conditioning on both observed data and interpretable latent variables, the inclusion of priors at runtime, and outputs predictive distributions for discrete and continuous data and latents. We show ACE's modeling flexibility and performance in diverse tasks such as image completion and classification, Bayesian optimization, and simulation-based inference.