Inference objectives

Inference objectives.

apg_loss(q_trace, p_trace, incoming_log_weight, incremental_log_weight, aggregate=True)

RWS objective that exploits conditional dependency.

avo_loss(q_trace, p_trace, incoming_log_weight, incremental_log_weight, aggregate=True)

Annealed Variational Objective.

elbo_loss(q_trace, p_trace, incoming_log_weight, incremental_log_weight, aggregate=True)

Evidence Lower Bound objective.

fkl_loss(q_trace, p_trace, incoming_log_weight, incremental_log_weight, aggregate=True)

Forward KL objective. Here we do not optimize p.

iwae_loss(q_trace, p_trace, incoming_log_weight, incremental_log_weight, aggregate=True)

Importance Weighted Autoencoder objective.

rkl_loss(q_trace, p_trace, incoming_log_weight, incremental_log_weight, aggregate=True)

Reverse KL objective.

rws_loss(q_trace, p_trace, incoming_log_weight, incremental_log_weight, aggregate=True)

Reweighted Wake-Sleep objective.