情報処理装置、情報処理方法、及びプログラム

Abstract

情報処理装置は、多数の候補オペレーションが配置された機械学習モデルネットワークを構築する、ネットワーク構築手段と、前記機械学習モデルネットワークを、異なる前記候補オペレーションが配置されたキャプテンネットワークとフォロワーネットワークとに分割する、ネットワーク分割手段と、訓練データを用いて、前記キャプテンネットワークに対して機械学習を実行し、更に、機械学習された前記キャプテンネットワークを用いた知識蒸留によって、前記フォロワーネットワークに対して機械学習を実行する、学習処理手段と、を備え、前記ネットワーク分割手段は、前記機械学習の後に、前記知識蒸留の結果を用いて、前記キャプテンネットワーク及び前記フォローネットワークにおける前記候補オペレーションの配置を更新する。

Claims

1. A network construction method for building a machine learning model network with numerous candidate operations, A network partitioning means for dividing the machine learning model network into a captain network and a follower network, each having different candidate operations, Using the training data, machine learning is performed on the captain network. Furthermore, a learning processing means performs machine learning on the follower network by knowledge distillation using the machine-learned captain network, Equipped with, The network partitioning means updates the arrangement of candidate operations in the captain network and the follow network using the results of the knowledge distillation after the machine learning. An information processing device characterized by the following:

2. The aforementioned machine learning model network is a supernet, The learning processing means outputs the captain network, in which the arrangement of the candidate operations has been updated, as a subnet of the supernet. The information processing apparatus according to claim 1.

3. The captain network has a multilayer structure in which a single candidate operation is placed in each layer. The follower network has a multilayer structure in which one or more candidate operations are arranged in each layer. The information processing apparatus according to claim 1.

4. The learning processing means, after updating the arrangement of the candidate operations, performs new machine learning on the updated captain network using new training data, and further performs new machine learning on the updated follower network by knowledge distillation using the newly trained captain network. The network partitioning means updates the arrangement of candidate operations in the captain network and the follow network using the results of knowledge distillation using the newly trained captain network after the new machine learning. The information processing apparatus according to claim 1.

5. We built a machine learning model network with numerous candidate operations, The machine learning model network is divided into a captain network and a follower network, each with different candidate operations. Using the training data, machine learning is performed on the captain network. Furthermore, by performing machine learning on the follower network through knowledge distillation using the machine-learned captain network, After the machine learning, the arrangement of the candidate operations in the captain network and the follow network is updated using the results of the knowledge distillation. An information processing method characterized by the following:

6. On the computer, A machine learning model network with numerous candidate operations is constructed. The machine learning model network is divided into a captain network and a follower network, each containing different candidate operations. Using the training data, machine learning is performed on the captain network. Furthermore, by performing knowledge distillation using the machine learning-trained captain network, machine learning is applied to the follower network. After the machine learning, the results of the knowledge distillation are used to update the arrangement of the candidate operations in the captain network and the follow network. A computer-readable recording medium that stores a program containing instructions.

Images