Similar to other optimization algorithms, the Adam optimizer can be implemented concisely using the high-level APIs provided by modern deep learning frameworks. Instead of manually initializing the state variables for momentum and the second moment, applying bias correction, and writing the update equations from scratch, practitioners can directly instantiate built-in optimizer classes. These built-in implementations handle the internal state tracking and numerical stability adjustments automatically. For example, in PyTorch, this is accomplished by instantiating torch.optim.Adam; in TensorFlow, by using tf.keras.optimizers.Adam; and in MXNet's Gluon API, by specifying the algorithm as 'adam'. The only requirement is to pass the appropriate configuration hyperparameters, such as the learning rate ($\eta$), to the built-in function. In PyTorch, this can be written concisely as:

trainer = torch.optim.Adam
d2l.train_concise_ch11(trainer, {'lr': 0.01}, data_iter)