Passing in multiple losses in tape.gradient

General Discussion
, , ,
#1

In the code below, there’s a line tape.gradient([reg_loss, cat_loss], model.trainable_variables). For trainable weights that affects both reg_loss and cat_loss, are the gradients for those weights just averaged or summed with respect to the two losses?

import tensorflow as tf
from tensorflow.keras.layers import Dense
from tensorflow.keras import Model
from sklearn.datasets import load_iris
tf.keras.backend.set_floatx('float64')
iris, target = load_iris(return_X_y=True)

X = iris[:, :3]
y = iris[:, 3]
z = target

ds = tf.data.Dataset.from_tensor_slices((X, y, z)).shuffle(150).batch(8)

class MyModel(Model):
    def __init__(self):
        super(MyModel, self).__init__()
        self.d0 = Dense(16, activation='relu')
        self.d1 = Dense(32, activation='relu')
        self.d2 = Dense(1)
        self.d3 = Dense(3, activation='softmax')

    def call(self, x, training=None, **kwargs):
        x = self.d0(x)
        x = self.d1(x)
        a = self.d2(x)
        b = self.d3(x)
        return a, b

model = MyModel()

loss_obj_reg = tf.keras.losses.MeanAbsoluteError()
loss_obj_cat = tf.keras.losses.SparseCategoricalCrossentropy()

optimizer = tf.keras.optimizers.Adam(learning_rate=1e-3)

loss_reg = tf.keras.metrics.Mean(name='regression loss')
loss_cat = tf.keras.metrics.Mean(name='categorical loss')

error_reg = tf.keras.metrics.MeanAbsoluteError()
error_cat = tf.keras.metrics.SparseCategoricalAccuracy()

@tf.function
def train_step(inputs, y_reg, y_cat):
    with tf.GradientTape() as tape:
        pred_reg, pred_cat = model(inputs)
        reg_loss = loss_obj_reg(y_reg, pred_reg)
        cat_loss = loss_obj_cat(y_cat, pred_cat)

    gradients = tape.gradient([reg_loss, cat_loss], model.trainable_variables)
    optimizer.apply_gradients(zip(gradients, model.trainable_variables))
    loss_reg(reg_loss)
    loss_cat(cat_loss)

    error_reg(y_reg, pred_reg)
    error_cat(y_cat, pred_cat)
#3

I have the same question too! Please answer if anyone is aware.
I also have an add on question. How would it work if I had multiple outputs ( say 4) and I wanted a total_loss to be applied on the first three outputs that was calculated by summing up the individual losses from each of these outputs and for the final fourth output alone I’d want a separate loss to be applied.In addition what if I wanted one of the outputs to have no specific loss applied? Passing None while doing tape.gradient gives errors Please explain how the custom training loop would look for this