
日 時 
2014年4月24日 （木） 15:00〜 17:00 
会 場 
九州大学 伊都キャンパス ウエスト２号館３F システム情報科学府第７講義室 
講演タイトル 
Recent development of recurrent neural network and graph matching for pattern
recogniton 
講演１ 
タイトル：
Fast HighPerformance Recognition Systems with Recurrent Neural Networks
and LSTM
（時間依存性を拡張したリカレントネットワークによる高精度パターン認識）
講演者： Volkmar Frinken (九州大学大学院システム情報科学研究院・特任助教)
アブストラクト：
Recurrent Neural Networks (RNNs) and their application to Pattern Recognition
will be presented, with a special focus on Long ShortTerm Memory (LSTM)
Neural Networks. Several problems of simple RNNs are described and LSTM
Neural Networks are introduced as a solution for those problems. For a
better understanding of the network, its behavior on several toy problems.
Finally, a quick outlook on extended architectures, such as the bi and
multi directional LSTM is give. An overview of realworld PRapplications
such as speech, handwriting and other PRdomains will be given will conclude
the talk.
●講演者の略歴
Volkmar Frinken is Research Assistant Professor and works since October
2013 at Kyushu University, Fukuoka, Japan. After obtaining his PhD in 2011
from the University of Bern, Switzerland, he spent 2 years at the Computer
Vision Center in Barcelona, Spain. His research interests encompass all
areas related to sequential processing, with a focus on handwriting recognition,
recurrent neural networks, hidden Markov models, and machine learning techniques
such as semisupervised learning.

講演２ 
タイトル：
Approximation of Graph Edit Distance Based on Hausdorff Matching
（Hausdorffマッチングによる近似グラフ編集距離)
講演者： Andreas Fischer (Polytechnique Montreal, Canada，学術研究員)
アブストラクト：
Graph edit distance is a powerful and flexible method for errortolerant
graph matching. Yet it can only be calculated for small graphs in practice
due to its exponential time complexity when considering unconstrained graphs.
In this talk, a recently proposed quadratic time approximation of graph
edit distance based on Hausdorff matching is presented. In a series of
experiments the performance of the proposed "Hausdorff edit distance"
is evaluated in the context of graph classification and is compared with
a cubic time algorithm based on the assignment problem. Investigated applications
include nearest neighbor classification of graphs representing letter drawings,
fingerprints, and molecular compounds as well as hidden Markov model classification
of vector space embedded graphs representing handwriting. In many cases,
a substantial speedup is achieved with only a minor loss in accuracy or,
in one case, even with a gain in accuracy. Overall, the proposed Hausdorff
edit distan ce shows a promising potential in terms of flexibility, efficiency,
and accuracy.
●講演者の略歴
Andreas Fischer is a postdoctoral researcher at Polytechnique Montreal,
Canada. He received his M.S. and Ph.D. degrees in Computer Science in 2008
and 2012 from the University of Bern, Switzerland. He has more than 30
publications related to handwriting recognition, historical document analysis,
hidden Markov models, recurrent neural networks, and graph matching for
structural pattern recognition. 
参 加 
聴講無料 / 予約不要 
問合先 
内田誠一（九州大学）
EMail: uchida（ａ）ait.kyushuu.ac.jp 
主催・共催 
主催：電子情報通信学会九州支部 
≪講演会ＴＯＰへ≫
