{
  "claim": 3,
  "literal_claim": "PGCM training uses an ELBO-style objective (Equation 2) combining a KL regularization term over prototype selection, task loss, concept loss, and image reconstruction loss (Equation 2).",
  "native_result": {
    "all_required_terms_nonzero": true,
    "all_required_terms_present": true,
    "all_terms_finite": true,
    "batch_entropy": 2.523866653442383,
    "concept_loss_real": 0.5122461318969727,
    "entropy": 0.03418504819273949,
    "one_hot_selector_control_kl": 3.4011973816621555,
    "primary_elbo_derivation_present": true,
    "prototype_emb_loss": 0.0,
    "returned_matches_logged_total": true,
    "returned_training_loss": 0.20000694692134857,
    "segmentation_loss": 1.7451768030696257e-08,
    "selector_kl_in_closed_interval": true,
    "selector_kl_to_uniform": 3.367012333469416,
    "total_loss": 0.20000694692134857,
    "train_c_loss": 0.04683137685060501,
    "train_recons_loss": 0.031370654702186584,
    "train_y_loss": 0.013922294601798058,
    "uniform_prior_categories": 30,
    "uniform_selector_control_kl": 0.0
  }
}