3. 紹介する論文
1. Reconstruction-Error-Based Learning for Continuous Emotion
Recognition in Speech
J. Han, Z. Zhang, F. Ringeval, and B. Schuller
2. Disjunctive Normal Shape Boltzmann Machine
E. Erdil, F. Mesadi, T. Tasdizen, and M. Cetin
3. Inferring Latent States in a Network Influenced by Neighbor
Activities: an Undirected Generative Approach
B. L. Samarakoon, M. N. Murthi, and K. Premaratne
MLSP-L6 Deep Learning III
20. 実験結果
- Walking silhouette data set of 150 binary images, each of which has 170x170 pixels.
- 1000 units for h1, and 50 units for h2.
- DNSM was processed using 6 polytopes.
22. 3. Inferring Latent States in a Network Influenced by
Neighbor Activities: an Undirected Generative Approach
B. L. Samarakoon, M. N. Murthi, and K. Premaratne
24. ちょっと疑問
To the best of our knowledge, no undirected models
have been applied to modeling neighbor influence
and hidden variables in networks.
“
25. 定式化
• あるノード i のneighborsの定義:
全ノード(ユーザ)集合隣接行列
• 表現したい確率分布:
隠れ状態(世論)系列
観測されるユーザカウント系列
e.g., the number of postings of a certain category
or number of infected people in a contact network
users may change their political biases depending on their
neighbors’ postings but they may not wish to express
these changes explicitly
40. 紹介した論文
1. Reconstruction-Error-Based Learning for Continuous Emotion
Recognition in Speech
エラー抽出器と識別器の2つのRNNを直列につないで連続感情認識
2. Disjunctive Normal Shape Boltzmann Machine
セグメンテーションとSBMを組み合わせてシルエット画像モデリング
3. Inferring Latent States in a Network Influenced by Neighbor
Activities: an Undirected Generative Approach
ダイナミックに変化する隠れ状態とユーザ嗜好をモデリング