2014.7.12 'Multilayer Networks'* 輪読会資料
* M. Kivela, A. Arenas, M. Barthelemy, J. P. Gleeson,
Y. Moreno, and M. A. Porter. Multilayer Networks.
arXiv:1309.7233.
As a part of the curriculum, I read related papers on machine learning and arranged the outline.
This time, I focused on "audio signal processing" and dealt with the latest survey paper and related papers about algorithm referenced in the paper.
2014.7.12 'Multilayer Networks'* 輪読会資料
* M. Kivela, A. Arenas, M. Barthelemy, J. P. Gleeson,
Y. Moreno, and M. A. Porter. Multilayer Networks.
arXiv:1309.7233.
As a part of the curriculum, I read related papers on machine learning and arranged the outline.
This time, I focused on "audio signal processing" and dealt with the latest survey paper and related papers about algorithm referenced in the paper.
Casa Ubicada dentro de un fraccionamiento privado, 2 recámaras con closet panorámico cada una, 2 y medio baños, patio trasero con piso azulejo, sala comedor, sala de tv, cocina integral, estacionamiento, vigilancia las 24 hrs. En Venta/Renta
e-mail et réseaux sociaux pour la communication institutionnelleStefano Amekoudi
Formation sensibilisation du personnel de la CONFEJES sur : De l'intérêt et de l'importance d'avoir et d'utiliser une messagerie institutionnelle et l'usage professionnelle des réseaux sociaux (Facebook et Twitter)
Blog link here >> http://goo.gl/Z8P6Q
(context/introduction of the presentation)
This is the presentation I used for my talk at the IDCAMP.
I tried to put together two things:
- an analysis of the new practices we need to create enduring and impacting enterprise in a time of radical change
- a practical 10 rules guide to be adopted.
All the material produced on my own is CC-BY-NC.
Data Pioneers - Roland Haeve (Atos Nederland) - Big data in organisatiesMultiscope
Roland Haeve is cross competence manager Big Data voor Atos Nederland. Roland heeft ruim 18 jaar ICT-ervaring in het aanbieden van complete oplossingen binnen onder andere Business Intelligence (BI) en Big Data (Analytics). Big Data is voor veel bedrijven nog pionieren en uitzoeken wat de mogelijkheden zijn. In zijn presentatie zal Roland ingaan op succesvolle Big Data cases. Hij zal hierbij niet enkel inzoomen op Nederland, maar ook bredere, Europese voorbeelden meenemen.
Vortrag cyborg marketing andreas wagener_hof_221014_slideAndreas Wagener
Science Fiction becomes true. Due to Google Glass, Wearables & (digital) Implants. Human beings turn into semi-digital objects - great chances for Marketing and creepy vision at the same time.
Structural data analysis based on multilayer networkstm1966
Introduction on data analysis based on multilayer networks (in Japanese). Some references of tools, datasets, conferences and Web sites are also mentioned.
17. “Mathematical Formulation of
Multilayer Networks”
• Manlio De Domenico, Albert Sole-Ribalta, Emanuele
Cozzo, Mikko Kivela, Ytamir Moreno, Mason A. Porter,
Sergio Gomez and Alex Arenas
• Physical Review X, 3, 041022, 2013, 15pages
• ネットワーク特徴量(次数中心性、クラスタ係数、固有
ベクトル中心性、モジュラリティ、von Neumann
entropy, diffusion)をテンソル表現に拡張。特殊な場合
として単一レイヤネットワークでのテンソル表現は既
存の特徴量と同一になることを示す。
• multiplexに限定されず、一般的なmultilayerでの枠組。
18. “Diffusion Dynamics on Multiplex
Networks”
• S. Gomez, A. Diaz-Guilera, J. Gomez-Gardenes,
C. J. Perez-Vicente, Y. Moreno, A. Arenas
• Physical Review Letters, 110, 028701, 5pages,
2013
• 2層のmultiplex networkにおけるsupra-
Laplacianの定義 ((N1+N2)×(N1+N2)の行列で
表記)
• Layer間の係数が小さい場合と大きい場合に
分けて議論
21. “Navigability of interconnected
networks under random failures”
• Manlio De Domenico, Albert Sole-Ribalta, Sergio
Gomez, and Alex Arenas, PNAS, doi
10.1073/pnas.1318469111 (2014)
• ランダムウオークによるカバレッジ、ランダムな
故障に対するresilienceについて
• Navigationを(i)同じノードに留まる(ii)同じレイヤ
内i->jに移動(iii)同じノードでレイヤα->βに移動
(iv)異なるノードi->j異なるレイヤα->βに移動に分
けて定式化
• London地下鉄や航空ネットワークなどのシミュ
レーションと実データとの比較
23. “Ranking in interconnected multilayer
networks reveals versatile nodes”
• Manlio De Domenico, Albert Sole-Ribalta, Elisa
Omodei, Sergio Gomez, Alex Arenas
• Nature Communications 6, Article
number:6868, Published 23 April 2015
• doi:10.1038/ncomms7868
• Multilayer networkの中心性としてversatile
centralityを提案。Aggregateなものと比較して
予測精度が向上。航空会社ネットワークでの
渋滞シミュレーションなどで実験
24. temporal networkとしてのmultilayer
network
• 一定の間隔毎に切ってmultilayer network化
– 「一定の時間間隔」をどう決めるか
– layer間の辺の強さをどう決めるか
A
B C
D
EF
0<=t < 5
A
B C
D
EF
5<=t < 10
A
B C
D
EF
10<=t < 15
"Temporal Networks", Petter Holme, Jari Saramakid, Physics Reports, Vol.519, Issue 3,
pp.97–125, 2012
55. 1.DNNによるグラフ処理
1(a)CNNの畳み込みの一般化
• Bruna et al. :”Spectral Networks and Deep Locally Connected
Networks on Graphs” (ICLR13) https://arxiv.org/abs/1312.6203
• Henaff et al. :”Deep Convolutional Networks on Graph-Structured
Data”, http://arxiv.org/abs/1506.05163
• Niepert et al.:”Spectral Representations for Convolutional Neural
Networks” (ICML16) http://arxiv.org/abs/1506.03767
1(b)オートエンコーダのグラフクラスタリングへの適用
• Tian et al., “Learning Deep Representations for Graph Clustering”
(AAAI14)
http://www.aaai.org/ocs/index.php/AAAI/AAAI14/paper/view/852
7
• Shao et al., “Deep Linear Coding for Fast Graph Clustering” (IJCAI15)
http://ijcai.org/Proceedings/15/Papers/534.pdf
65. Deep Compression: Compressing Deep
Neural Networks with Pruning, Trained
Quantization and Huffman Coding
• Song Han, Huizi Mao, William J. Dally
• http://arxiv.org/abs/1510.00149
• ICLR 2016 のbest paper
• 学習済DNNが大きくモバイルデバイスに入らな
い->メモリサイズの削減
• 3段階の処理(pruning, trained quantization,
Huffman coding)で1/35~1/49に
71. “EIE: Efficient Inference Engine on
Compressed Deep Neural Networks”
• Song Han, Xingyu Liu, Huizi Mao, Jing Pu,
Ardavan Pedram, Mark A. Horowitz, William J.
Dally, arXiv:1602.01528v1
• 圧縮したDNNを用いた推論機構
72. SqueezeNet: AlexNet-level accuracy
with 50x fewer parameters and
<0.5MB model size
• Forrest N. Iandola, Song Han, Matthew W.
Moskewicz, Khalid Ashraf, William J. Dally, Kurt
Keutzer
• https://arxiv.org/abs/1602.07360
• 多くの研究は精度向上を目指しているが、精
度が同じなら小さいCNNの方が良い
• ImageNetでAlexNetレベルの精度でパラメー
タ数を1/50に。モデル圧縮もおこなうと0.5MB
以下に(AlexNetの1/510)