Generative Distribution Embedding
A method for representing probability distributions with a parameterized generative model. It learns a mapping from a low-dimensional latent variable to the state space and uses an encoder to map observation histories to a distribution over latent variables, allowing an arbitrary number of candidate states to be generated and their densities computed.
Timeline (last 90 days)
BETS uses a generative distribution embedding (GDE) to represent the belief at each node and updates it through the POMDP dynamics.
A generative distribution embedding (GDE) is a method for representing probability distributions with a parameterized generative model, learning a mapping from low-dimensional latent variables to the state space, while using an encoder to map observation histories to a distribution over latent variables.
BETS jointly optimizes the belief model (GDE), policy network, and value network during training, and the belief model's loss minimizes the discrepancy between the generated distribution and the true hidden state, commonly using negative log-likelihood or maximum mean discrepancy.
All Facts (3)
BETS uses a generative distribution embedding (GDE) to represent the belief at each node and updates it through the POMDP dynamics.
50%UnverifiedA generative distribution embedding (GDE) is a method for representing probability distributions with a parameterized generative model, learning a mapping from low-dimensional latent variables to the state space, while using an encoder to map observation histories to a distribution over latent variables.
50%UnverifiedBETS jointly optimizes the belief model (GDE), policy network, and value network during training, and the belief model's loss minimizes the discrepancy between the generated distribution and the true hidden state, commonly using negative log-likelihood or maximum mean discrepancy.
50%