The generalized cross entropy (GCE) loss offers robustness to noisy labels by
interpolating between categorical cross entropy (q -> 0) and mean absolute
error (q -> 1). For a true-class probability p and noise parameter q,
the loss is loss = (1 - p^q) / q.
loss_categorical_generalized_cross_entropy(
y_true,
y_pred,
q = 0.5,
...,
reduction = "sum_over_batch_size",
name = "categorical_generalized_cross_entropy",
dtype = NULL
)Generalized cross entropy loss value(s).
Integer class indices with shape (batch_size) or (batch_size, 1).
Predicted class probabilities with shape (batch_size, num_classes).
Float in (0, 1). Controls the transition between cross entropy and mean
absolute error. Defaults to 0.5.
As q approaches 0: behaves like categorical cross entropy.
As q approaches 1: behaves like mean absolute error.
For forward/backward compatibility.
Type of reduction to apply to the loss. In almost all cases
this should be "sum_over_batch_size". Supported options are
"sum", "sum_over_batch_size", "mean",
"mean_with_sample_weight" or NULL. "sum" sums the loss,
"sum_over_batch_size" and "mean" sum the loss and divide by the
sample size, and "mean_with_sample_weight" sums the loss and
divides by the sum of the sample weights. "none" and NULL
perform no aggregation. Defaults to "sum_over_batch_size".
Optional name for the loss instance.
Dtype used for loss computations. Defaults to config_floatx() (the global
float type).
y_true <- c(0L, 1L, 0L, 1L)
y_pred <- rbind(
c(0.7, 0.3),
c(0.2, 0.8),
c(0.6, 0.4),
c(0.4, 0.6)
)
gce <- loss_categorical_generalized_cross_entropy(q = 0.7)
gce(y_true, y_pred)
## tf.Tensor(0.34529287, shape=(), dtype=float32)Zhang & Sabuncu (2018), "Generalized Cross Entropy Loss for Training Deep Neural Networks with Noisy Labels"
Other losses:
Loss()
loss_binary_crossentropy()
loss_binary_focal_crossentropy()
loss_categorical_crossentropy()
loss_categorical_focal_crossentropy()
loss_categorical_hinge()
loss_circle()
loss_cosine_similarity()
loss_ctc()
loss_dice()
loss_hinge()
loss_huber()
loss_kl_divergence()
loss_log_cosh()
loss_mean_absolute_error()
loss_mean_absolute_percentage_error()
loss_mean_squared_error()
loss_mean_squared_logarithmic_error()
loss_poisson()
loss_sparse_categorical_crossentropy()
loss_squared_hinge()
loss_tversky()
metric_binary_crossentropy()
metric_binary_focal_crossentropy()
metric_categorical_crossentropy()
metric_categorical_focal_crossentropy()
metric_categorical_hinge()
metric_hinge()
metric_huber()
metric_kl_divergence()
metric_log_cosh()
metric_mean_absolute_error()
metric_mean_absolute_percentage_error()
metric_mean_squared_error()
metric_mean_squared_logarithmic_error()
metric_poisson()
metric_sparse_categorical_crossentropy()
metric_squared_hinge()