Transfer learning is a deep learning (DL) method that allows the use of a pretrained model with a new dataset. The new data must be similar enough to the original data so the learned features in the model weights apply well.
Intel® Arc⢠A-Series discrete GPUs provide an easy way to run DL workloads quickly on your PC, working with both TensorFlow* and PyTorch* models.
This is because a "headless model" is used, where the bulk of the layers are left with pretrained static weights, while the final model layer is replaced with a fresh dense layer that is trained to the new dataset.
[
add
]
[
|
|
...
]