【DL輪読会】Mastering Diverse Domains through World ModelsDeep Learning JP
The document summarizes Mastering Diverse Domains through World Models, which introduces Dreamer V3. Dreamer V3 improves on previous Dreamer models through the use of symlog prediction networks and actor critics trained with temporal difference learning. It achieves better performance than ablation models in the Atari domain.
【DL輪読会】Mastering Diverse Domains through World ModelsDeep Learning JP
The document summarizes Mastering Diverse Domains through World Models, which introduces Dreamer V3. Dreamer V3 improves on previous Dreamer models through the use of symlog prediction networks and actor critics trained with temporal difference learning. It achieves better performance than ablation models in the Atari domain.
1. The document discusses implicit behavioral cloning, which was presented in a 2021 Conference on Robot Learning (CoRL) paper.
2. Implicit behavioral cloning uses an implicit model rather than an explicit model to map observations to actions. The implicit model is trained using an InfoNCE loss function to discriminate positive observation-action pairs from negatively sampled pairs.
3. Experiments showed that the implicit model outperformed explicit models on several manipulation tasks like bi-manual sweeping, insertion, and sorting. The implicit approach was able to generalize better than explicit behavioral cloning.
1. The document discusses implicit behavioral cloning, which was presented in a 2021 Conference on Robot Learning (CoRL) paper.
2. Implicit behavioral cloning uses an implicit model rather than an explicit model to map observations to actions. The implicit model is trained using an InfoNCE loss function to discriminate positive observation-action pairs from negatively sampled pairs.
3. Experiments showed that the implicit model outperformed explicit models on several manipulation tasks like bi-manual sweeping, insertion, and sorting. The implicit approach was able to generalize better than explicit behavioral cloning.
16. 実験:画像とテキストのデータセット
• CUB
– 200種類のカテゴリ
– 11788の鳥画像
• Oxford-102
– 102のカテゴリ
– 8189の花画像
各画像に対して5つの説明テキスト
(著者らが付けた?)
this bird has wings
that are black and
has a yellow crown
013.Bobolink
16
19. 課題:文章に含まれない画像情報(スタイル)
• 文章に含まれる画像情報
– 黒い羽で黄色い頭頂の鳥
• 文章に含まれない画像情報
– 背景が緑、左を向いている、など
– 著者らはスタイルと呼んでいる
• 入力ベクトルのうち、
– 文章情報はテキストembedding
– スタイル情報はzが獲得する
this bird has wings
that are black and
has a yellow crown
19