【速報】IEEE SSCI 2014

IEEE SSCIという学会でフロリダに来ています。
1) Endoscope Image Analysis Method for Evaluating the Extent of Early Gastric Cancer Tomoyuki Hiroyasu, Katsutoshi Hayashinuma, Hiroshi Ichikawa, Nobuyuki Yagi and Utako Yamoto
2) Gender classification of subjects from cerebral blood flow changes using Deep Learning Tomoyuki Hiroyasu, Kenya Hanawa and Utako Yamoto
3) A feature transformation method using genetic programming for two-class classification Tomoyuki Hiroyasu, Toshihide Shiraishi, Tomoya Yoshida and Utako Yamamoto
4) Electroencephalographic Method Using Fast Fourier Transform Overlap Processing for Recognition of Right- or Left-handed Elbow Flexion Motor Imagery Tomoyuki Hiroyasu, Yuuki Ohkubo and Utako Yamamoto


 報告者氏名 林沼勝利
発表論文タイトル 内視鏡画像における早期胃癌の進展範囲評価のための解析手法の検討
発表論文英タイトル Endoscope Image Analysis Method for Evaluating the Extent of Early Gastric Cancer
著者 廣安知之, 林沼勝利, 市川寛, 八木信明, 山本詩子
主催 IEEE Computational Intelligence Society
講演会名 IEEE SSCI 2014
会場 Caribe Royale All-Suite Hotel and Convention Center
開催日程 2014/12/09-2014/12/12


  1. 講演会の詳細

2014/12/09から2014/12/12にかけて,アメリカ・フロリダ州のCaribe Royale All-Suite Hotel and Convention Centerにて開催されました2014 IEEE Symposium Series on Computational Intelligence (IEEE SSCI 2014)に参加いたしました.このSSCI 2014は,IEEE Computational Intelligence Societyによって主催された学会で,Computational Intelligenceに関する様々なジャンルの29のシンポジウムが1つの会場で開催されました.

  1. 研究発表
    • 発表概要

私は10日の午前のセッション「CIMSIVP’14 Session 2: Application」に参加いたしました.発表の形式は口頭発表で,15分の講演時間と5分の質疑応答時間となっておりました.

In this study, a system is proposed to help physicians perform processing on images taken with a magnifying endoscopy with narrow band imaging. In our proposed system, the transition from lesion to normal zone is quantitatively analyzed and presented by texture analysis. Eleven feature values are calculated, i.e., six from a co-occurrence matrix and five from a run length matrix with a scanning window. Integrating these feature values formulates an effective and representative feature value, which is used to draw a color map, so the transition from lesion to normal zone can be visibly illustrated. In this paper, the proposed method is applied to images, and the efficacy is considered. This method is also applied to some rotated images to examine whether it could work effectively on such images.


  • 質疑応答

蔚山大学のMyeongsu Kangさんからの質問です.こちらの質問は,特徴量は画像が回転しても変化しないのかというものでした.回転した画像で実験を行ってみたところ,大きな差は見られなかったと回答したかったのですが,うまく答えることができませんでした.

  • 感想


  1. 聴講


発表タイトル       : Counting, Detecting and Tracking of People in Crowded Scenes著者                  : Mubarak Shahセッション名       : Keynote Talk: CIMSIVP’14Abstract            : In this talk, first I will present a new approach for counting people in extremely dense crowds. Our approach relies on multiple sources of information such as low confidence head detections, repetition of texture elements (using SIFT), and frequency-domain analysis to estimate counts, along with confidence associated with observing individuals, in an image region. In addition, we employ a global consistency constraint on counts using Markov Random Field. This caters for disparity in counts in local neighborhoods and across scales.
Next, I will discuss how we explore context for human detection in dense crowds in the form of locally-consistent scale prior which captures the similarity in scale in local neighborhoods and its smooth variation over the image. Using the scale and confidence of detections obtained from an underlying human detector, we infer scale and confidence priors using Markov Random Field. In an iterative mechanism, the confidences of detections are modified to reflect consistency with the inferred priors, and the priors are updated based on the new detections. The final set of detections obtained are then reasoned for occlusion using Binary Integer Programming where overlaps and relations between parts of individuals are encoded as linear constraints.
Finally, I will present a method for tracking in dense crowds using prominence and neighborhood motion concurrence. Our method begins with the automatic identification of prominent individuals from the crowd that are easy to track. Then, we use Neighborhood Motion Concurrence to model the behavior of individuals in a dense crowd, this predicts the position of an individual based on the motion of its neighbors.


発表タイトル       : Single Frame Super Resolution: Gaussian Mixture Regression and Fuzzy Rule-Based Approaches著者                  : Nikhil R. Palセッション名       : Plenary TalkAbstract            : High quality image zooming is an important problem. There are many methods that use multiple low resolution (LR) frames of the same scene with different sub-pixel shifts as input to generate the high resolution (HR) images. Now a days single frame super resolution (SR) methods that use just one LR image to obtain the HR image has become popular. In this talk we shall discuss a novel fuzzy rule based single frame super resolution method. This is a patch based method, where each LR patch is replaced by a HR patch generated by a Takagi-Sugeno type fuzzy rule-based system. We shall discuss in details the generations of the training data, the initial generation of the fuzzy rules, their refinement and how to use the rules for generation of SR images. In this context we shall also develop a Gaussian Mixture Regression (GMR) model for the same problem. Both the fuzzy rule based system and GMR are found to be quite effective. Comparison of performance of the fuzzy rule-based system with five existing methods as well as with the GMR method in terms of the several quality criteria demonstrates the superior performance of the fuzzy rule-based system.

この発表は,ファジィ推論やガウス混合回帰を用いたsingle imageからの超解像技術に関する内容でした.近年,超解像技術はテレビや顕微鏡,医用画像など様々な方面への応用が進んでおり,画像処理の分野では注目の技術の一つであるため,とても勉強になる発表でした.

発表タイトル       : Finding Optimal Transformation Function for Image Thresholding Using Genetic Programming著者                  : Shaho Shahbazpanahi, Shahryar Rahnamayanセッション名       : CIMSIVP’14 Session 4: Algorithms IAbstract            : In this paper, Genetic Programming (GP) is employed to obtain an optimum transformation function for bi-level image thresholding. The GP utilizes a user prepared gold sample to learn from. A magnificent feature of this method is that it does not require neither a prior knowledge about the modality of the image nor a large training set to learn from. The performance of the proposed approach has been examined on 147 X-ray lung images. The transformed images are thresholded using Otsu’s method and the results are highly promising. It performs successfully on 99% of the tested images. The proposed method can be utilized for other image processing tasks, such as, image enhancement or segmentation.


発表タイトル       : Disguised face detection and recognition under the complex background著者                  : Jing Li, Bin Li, Yong Xu, Kaixuan Lu, Ke Yan, Lunke Feiセッション名       : CIBIM’14 Session 3: Face Detection and RecognitionAbstract            : In this paper, we propose an effective method for disguised face detection and recognition under the complex background. This method consists of two stages. The first stage determines whether the object is a person. In this stage, we propose the first-dynamic-then-static foreground object detection strategy. This strategy exploits the updated learning-based codebook model for moving object detection and uses the Local Binary Patterns (LBP) + Histogram of Oriented Gradients (HOG) feature-based head-shoulder detection for static target detection. The second stage determines whether the face is disguised and the classes of disguises. Experiments show that our method can detect disguised faces in real time under the complex background and achieve acceptable disguised face recognition rate.


  • 2014 IEEE Symposium Series on Computational Intelligence Proceedings


 報告者氏名 大久保祐希
発表論文タイトル 肘関節屈曲運動イメージのFFTオーバーラップ処理を用いた脳波の左右識別手法
発表論文英タイトル Electroencephalographic Method Using Fast Fourier Transform Overlap Processing for Recognition of Right- or Left-handed Elbow Flexion Motor Imagery
著者 廣安知之, 大久保祐希, 山本詩子,
開催日程 2014/12/09 – 2014/12/12


  1. 講演会の詳細

2014/12/09から2014/12/12にかけて,アメリカ・フロリダ・オーランドにて開催されました2014 IEEE SYMPOSIUM SERIES ON COMPUTATIONAL INTELLIGENCE (SSCI2014)1) に参加しました.このSSCI2014は,IEEEによって主催された学会で,多くの学生や教員,企業が参加しました.この学会では,ビッグデータや画像処理,機械学習,さらに認知機構のモデル化,ブレイン・コンピュータ・インターフェイスなどの知的コンピューティングに関する研究が口頭やポスターにより発表されていました.

  1. 研究発表
    • 発表概要

私は12月12日の「CIBCI’14 session2」に参加いたしました.発表の形式は15分の口頭発表,4分の質疑応答でした.

Recently, systems using motor imagery (MI) have been developed as practical examples of brain-computer interface (BCI). Electroencephalography (EEG) was used to generate an electroencephalogram of elbow flexion. In addition, a method was proposed to extract the feature values that would enable the recognition right- or left-handed elbow flexion MI. In the proposed method, fast Fourier transform overlap processing was used to determine the time period required to extract feature values. In this study, the following two experiments were performed. 1) the recognition of right- or left-handed elbow flexion by analyzing only the MI time period and 2) recognition of the right- or left-handed when the MI time period was presumed. In the first experiment, right- or left-handed elbow flexion MI was processed for 20 subjects using support vector machine and the proposed method was used to extract the feature values. In the second experiment, the presumed MI time was determined using the channels in which the highest accuracy was obtained in the first experiment, and then, right- or left-handed recognition was processed for the time period presumed. In the first experiment, the recognition accuracy of the proposed method was superior to that of the previous method in 15 of 20 the subjects. In the second experiment, the mean accuracy was 7.2%. Therefore, the recognition accuracy can be improved by improving the MI detection method.


  • 質疑応答


  • 感想

私の発表はBCI専門のセッションで行われました.質疑応答の際にはこの研究の方向性や改善点をBCIのプロフェッショナルの方々に教授して頂いたため,今後の研究に対するモチベーションの向上につながりました.また,私自身初めての英語での発表であり私は英語が苦手であるため,論文作成から発表まで非常に苦労しました.また,発表時の質疑応答やBreak Time,Banquetなどで多くの研究者の方々と話す機会がありましたが,消極的になり上記のような方々とあまり話すことが出来なかったことを後悔しています.

  1. 聴講


発表タイトル       :Development of an Autonomous BCI Wheelchair著者                  :Danny Wee-Kiat Ng, Ying-Wei Soh and Sing-Yau Goh, UTAR, Malaysiaセッション名       :CIBCI’14 Session 1
Abstruct            :Restoration of mobility for the movement impaired is one of the important goals for numerous Brain Computer Interface (BCI) systems. In this study, subjects used a steady state visual evoked potential (SSVEP) based BCI to select a desired destination. The selected destination was communicated to the wheelchair navigation system that controlled the wheelchair autonomously avoiding obstacles on the way to the destination. By transferring the responsibility of controlling the wheel chair from the subject to the navigation software, the number of BCI decisions needed to be completed by the subject to move to the desired destination is greatly reduced.


発表タイトル       :Across-subject estimation of 3-back task performance using EEG signals著者                  :Jinsoo Kim, Min-Ki Kim, Christian Wallraven and Sung-Phil Kim, Department of Brain and Cognitive Engineering, Korea University, Korea (South); Department of Human and Systems Engineering, Ulsan National Institute of Science and Technology, Korea (South)セッション名       :CIBCI’14 Session 1
Abstruct            :This study was aimed at estimating subjects’ 3-back working memory task error rate using electroencephalogram (EEG) signals. Firstly, spatio-temporal band power features were selected based on statistical significance of across- subject correlation with the task error rate. Method-wise, ensemble network model was adopted where multiple artificial neural networks were trained independently and produced separate estimates to be later on aggregated to form a single estimated value. The task error rate of all subjects were estimated in a leave-one-out cross-validation scheme. While a simple linear method underperformed, the proposed model successfully obtained highly accurate estimates despite being restrained by very small sample size.


発表タイトル       :Abnormal Event Detection in EEG Imaging – Comparing Predictive and Model-based Approaches著者                  :Jayanta Dutta, Banerjee Bonny, Ilin Roman and Kozma Robert, U of Memphis, United States; Air Force Research Lab, United Statesセッション名       :CIBCI’14 Session 1
Abstruct            :The detection of abnormal/unusual events based on dynamically varying spatial data has been of great interest in many real world applications. It is a challenging task to detect abnormal events as they occur rarely and it is very difficult to predict or reconstruct them. Here we address the issue of the detection of propagating phase gradient in the sequence of brain images obtained by EEG arrays. We compare two alternative methods of abnormal event detection. One is based on prediction using a linear dynamical system, while the other is a model-based algorithm using expectation minimization approach. The comparison identifies the pros and cons of the different methods, moreover it helps to develop an integrated and robust algorithm for monitoring cognitive behaviors, with potential applications including brain-computer interfaces (BCI).

この研究ではうさぎの脳に電極を埋めてデータを取得していました.処理方法としてはフィルタリング(FIR filter)とヒルバート変換を使用し,うさぎにとって突然の出来事が起きた時の脳波に通常時とは異なる傾向の脳波が混在することが挙げられていました.

発表タイトル       :Sensitivity Analysis of Hilbert Transform with Band-Pass FIR Filters for Robust Brain Computer Interface著者                  :Jeffery Davis and Kozma Robert, CLION, U of Memphis, United States; U of Memphis, United Statesセッション名       :CIBCI’14 Session 2
Abstruct            :Transient cortical oscillations in the form of rapid synchronization-desynchronization transitions are key candidates of neural correlates of higher cognitive activity monitored by scalp EEG and intracranial ECoG arrays. The transition period is in the order of 20-30 ms, and standard signal processing methodologies such as Fourier analysis are inadequate for proper characterization of the phenomenon. Hilbert transform- based (HT) analysis has shown great promise in detecting rapid changes in the synchronization properties of the cortex measured by high-density EEG arrays. Therefore, HT is a primary candidate of operational principles of brain computer interfaces (BCI). Hilbert transform over narrow frequency bands has been applied successfully to develop robust BCI methods, but optimal filtering is a primary concern. Here we systematically evaluate the performance of FIR filters over various narrow frequency bands before applying Hilbert transforms. The conclusions are illustrated using rabbit ECoG data. The results are applicable for the analysis of scalp EEG data for advanced BCI devices.

上記の「Abnormal Event Detection in EEG Imaging – Comparing Predictive and Model-based Approaches」と同じ研究を行っていました.

発表タイトル       :Development of SSVEP-based BCI using Common Frequency Pattern to Enhance System Performance著者                  :Li-Wei Ko, Shih-Chuan Lin, Wei-Gang Liang, Oleksii Komarov and Meng-Shue Song, Institute of Bioinformatics and Systems Biology, NCTU, Taiwan; Department of Physics, NTHU, Taiwan; Institute of Molecular Medicine andBioengineering, NCTU,  aiwan; Brain Research Center, NCTU, Taiwan
セッション名       :CIBCI’14 Session 2
Abstruct            :Brain Computer Interface(BCI) systems provide an additional way for people to interact with external environment without using peripheral nerves or muscles. In a variety of BCI systems, a BCI system based on the steady-state visual evoked potentials (SSVEP) is one most common system known for application, because of its ease of use and good performance with little user training. In this study, we employed the common frequency pattern method (CFP) to improve the accuracy of our EEG-based SSVEP BCI system. We used four basic classifiers (SVM, KNNC, PARZENDC, LDC) to estimate the accuracy of our SSVEP system. Without using CFP, the highest accuracy of the EEG-based SSVEP system was 80%. By using CFP, the accuracy could be upgraded to 95%..


発表タイトル       :Identification of Three Mental States Using a Motor Imagery Based Brain Machine Interface Machine著者                  :Trongmun Jiralerspong, Chao Liu and Jun Ishikawa, Tokyo Denki University, Japanセッション名       :CIBCI’14 Session 3
Abstruct            :The realization of robotic systems that understands human intentions and produces accordingly complex behaviors is needed particularly for disabled persons, and would consequently benefit the aged. For this purpose, a control technique that recognizes human intentions from neural responses called brain machine interface (BMI) have been suggested. The unique ability to communicate with machines by brain signals opens a wide area of applications for BMI. Recently, combination of BMI capabilities with assistive technology has provided solutions that can benefit patients with disabilities and many others. This paper proposes a BMI system that uses a consumer grade electroencephalograph (EEG) acquisition device. The aim is to develop a low cost BMI system suitable for households and daily applications. As a preliminary study, an experimental system has been prototyped to classify user intentions of moving an object up or down, which are basic instructions needed for controlling most electronic devices by using only EEG signals. In this study, an EEG headset equipped with 14 electrodes is used to acquire EEG signals but only 8 electrodes are used to identify user intentions. The features of EEG signals are extracted based on power spectrum and artificial neural network are used as classifiers. To evaluate the system performance, online identification experiments for three subjects are conducted. Experiment results show that the proposed system has worked well and could achieve an overall correct identification rate of up to 72 % with 15 minutes of training time by a user with no prior experience in BMI.




報告者氏名 白石駿英
発表論文タイトル 2クラス分類の為の遺伝的プログラミングを用いた特徴量変換手法の提案
発表論文英タイトル A Feature Transformation Method using GeneticProgramming for Two-Class Classification
著者 廣安知之,白石駿英, 吉田倫也,山本詩子
講演会名 2014 IEEE Symposium onComputational Intelligence and Data Mining
会場 Florida,Orlando,U.S.A
開催日程 2014/12/09-2014/12/12


  1. 講演会の詳細

2012/12/09から2012/12/12にかけて, フロリダ州オーランドにて開催されましたSSCI 20141)に参加いたしました.このSSCIは,IEEEによって主催されており,Computational Intelligenceについての様々な分野を含んだ国際会議です.私は12/10~12/12の3日間に参加いたしました.本研究室からは他に廣安先生,大久保さん,塙君,林沼君が参加しました.

  1. 研究発表
    • 発表概要


本論文では,遺伝的プログラミング(GP)を使用して,2クラス分類のための特徴量変換法を提案する.本手法で用いるGPは,SVMの分類精度を向上させるための変換式を導出する.本論文の目的は,変換された特徴空間を評価するために重み関数を提案し,GPの評価関数として実装して評価した.提案した評価関数の機能は,サンプルの理想的な2クラス分布が仮定され,実際に得られる分布との間の距離を算出する. 重みはこれらの距離に課されることになる.提案手法の有効性を確認するため,数値実験を行った.その結果、提案手法の分類精度は従来の方法よりも良好であった.


  • 質疑応答


  • 感想


  1. 聴講


発表タイトル    :Two key properties of dimensionality reduction methods著者          :John A. Lee, Michel Verleysen
セッション名  :CIDM’14 Session 5: High Dimensional Data Analysis
Abstruct        :Dimensionality reduction aims at providing faithful low-dimensional representations of high-dimensional data. Its general principle is to attempt to reproduce in a low-dimensional space the salient characteristics of data, such as proximities. A large variety of methods exist in the literature, ranging from principal component analysis to deep neural networks with a bottleneck layer. In this cornucopia, it is rather difficult to find out why a few methods clearly outperform others. This paper identifies two important properties that enable some recent methods like stochastic neighborhood embedding and its variants to produce improved visualizations of high-dimensional data. The first property is a low sensitivity to the phenomenon of distance concentration. The second one is plasticity, that is, the capability to forget about some data characteristics to better reproduce the other ones. In a manifold learning perspective, breaking some proximities typically allow for a better unfolding of data. Theoretical developments as well as experiments support our
claim that both properties have a strong impact. In particular, we show that equipping classical methods with the missing properties significantly improves their results.


発表タイトル    :Valid Interpretation of Feature Relevance for Linear Data Mappings著者          :Beno Fr´enay,Daniela Hofmann,Alexander Schulz, Michael Biehlz, and Barbara Hammer
セッション名  :CIDM’14 Session 5: High Dimensional Data Analysis
Abstruct       :Linear data transformations constitute essential operations in various machine learning algorithms, ranging from linear regression up to adaptive metric transformation. Often, linear scalings are not only used to improve the model accuracy, rather feature coefficients as provided by the mapping are interpreted as an indicator for the relevance of the feature for the task at hand. This principle, however, can be misleading in particular for high-dimensional or correlated features, since it easily marks irrelevant features as relevant or vice versa. In this contribution, we propose a mathematical formalisation of the minimum and maximum feature relevance for a given linear transformation which can efficiently be solved by means of linear programming. We evaluate the method in several benchmarks, where it becomes apparent that the minimum and maximum relevance closely resembles what is often referred to as weak and strong relevance of the features; hence unlike the mere scaling provided by the linear mapping, it ensures valid interpretability.


発表タイトル    :Review of Coevolutionary Developments of Evolutionary Multi-Objective and Many-Objective Algorithms and Test Problems著者          :Hisao Ishibuchi, Hiroyuki Masuda, Yuki Tanigaki and Yusuke Nojima
セッション名  :MCDM’14 Session 6: Evolutionary Multi-Objective Optimization
Abstruct :In the evolutionary multi-objective optimization (EMO) community, some well-known test problems have been frequently and repeatedly used to evaluate the performance of EMO algorithms. When a new EMO algorithm is proposed, its performance is evaluated on those test problems. Thus algorithm development can be viewed as being guided by test problems. A number of test problems have already been designed in the literature. Since the difficulty of designed test problems is usually evaluated by existing EMO algorithms through computational experiments, test problem design can be viewed as being guided by EMO algorithms. That is, EMO algorithms and test problems have been developed in a coevolutionary manner. The goal of this paper is to clearly illustrate such a coevolutionary development.
We categorize EMO algorithms into four classes: non-elitist, elitist, many-objective, and combinatorial algorithms. In each category of EMO algorithms, we examine the relation between developed EMO algorithms and used test problems. Our examinations of test problems suggest the necessity of strong diversification mechanisms in many-objective EMO algorithms such as SMS-EMOA, MOEA/D and NSGA-II


発表論文タイトル 脳血流変化量からDeep Learningを用いた
発表論文英タイトル Gender classification of subjects from
cerebral blood flow changes using Deep Learning
著者 廣安知之,塙賢哉,山本詩子
主催 IEEE Symposium Series on Computational Intelligence
講演会名 SSCI 2014
会場 Caribe Royale Hotel
開催日程 2014/12/10-2014/12/12


  1. 講演会の詳細

2014/12/10から2014/12/12にかけて,フロリダのオーランドにて開催されましたSSCI 2014に参加いたしました.この学会は,IEEE Symposium Series on Computational Intelligenceによって主催された学会で,計算知能(CI)に関するすべての側面を促進させることを目的に開催されています.

  1. 研究発表
    • 発表概要

私は11日の午後のセッション「SSCI’14 Poster Session」に参加いたしました.発表の形式はポスター発表で,1時間35分の講演時間となっておりました.

本稿ではfNIRSによって計測された脳血流変化量を用いて被験者の性別をDeep Learningを用いて分類することを考える.脳血流変化量は脳活動と関係していることが報告されている.それで,もしこの分類が高い識別が可能であるなら,性別の分類は脳活動に関係しているはずである.実験ではホワイトノイズの環境下で記憶課題を行った被験者からfNIRSデータが計測された.結果から,学習された分類器は高い識別率であったことが確認された.この結果の要因として脳活動と脳血流変化量の間に関係が存在していることが考えられる.

質問者の氏名を控え損ねてしまいました. 質問内容はfNIRS以外にもfMRIでも同様のことを行っていないのかということでした.この質問に対する回答ですが,fNIRSのみで行っていますと答えました.
質問者の氏名を控え損ねてしまいました. 質問内容は違うdeep learningの手法でも行って精度の比較をしていないかということでした.この質問に対する回答ですが,前段階ではこの手法のみで行っていますと答えました.

  • 感想


  1. 聴講


発表タイトル       : Adaptive Particle Swarm Optimization Learning in a Time Delayed Recurrent Neural Network for Multi-Step Prediction
著者                  : Kostas Hatalis, Basel Alnajjab, Shalinee Kishore and Alberto Lamadrid
セッション名       : Neural Networks
Abstruct            : In this study we propose the development of an adaptive particle swarm optimization (APSO) learning algorithm to train a non-linear autoregressive (NAR) neural network, which we call PSONAR, for short term time series prediction of ocean wave elevations. We also introduce a new stochastic inertial weight to the APSO learning algorithm. Our work is motivated by the expected need for such predictions by wave energy farms. In particular, it has been shown that the phase resolved predictions provided in this paper could be used as inputs to novel control methods that hold promise to at least double the current efficiency of wave energy converter (WEC) devices. As such, we simulated noisy ocean wave heights for testing. We utilized our PSONAR to get results for 5, 10, 30, and 60 second multistep predictions. Results are compared to a standard backpropagation model. Results show APSO can outperform backpropagation in training a NAR neural network.

この発表はNeual Networkの学習アルゴリズムを提案していました.また,海の波の上昇の短期時系列予測を用いて評価していました.数式がわかる部分もあったのですが,ほとんど難しくわからなかったです.

発表タイトル       :Attractor Flow Analysis for Recurrent Neural Network with Back-to-Back Memristors
著者                  : Gang Bao and Zhigang Zeng
セッション名       : Neural Networks
Abstruct            : Memristor is a nonlinear resistor with the character of memory and is proved to be suitable for simulating synapse of neuron. This paper introduces two memristors in series with the same polarity (back-to-back) as simulator for neuron’s synapse and presents the model of recurrent neural networks with such back-to-back memristors. By analysis techniques and fixed point theory, some sufficient conditions are obtained for recurrent neural network having single attractor flow and multiple attractors flow. At last, simulation with numeric examples verify our results.


発表タイトル       : Fingerprint multilateration for automatically classifying evolved Prisoner’s Dilemma agents
著者                  : Jeffrey Tsang
セッション名       : Neural Networks
Abstruct            : We present a novel tool for automatically analyzing evolved Prisoner’s Dilemma agents, based on combining two existing techniques: fingerprinting, which turns a strategy into a representation-independent functional summary of its behaviour, and multilateration, which finds the location of a point in space using measured distances to a known set of anchor points. We take as our anchor points the space of 2-state deterministic transducers; using this, we can emplace an arbitrary strategy into 7-dimensional real space by computing numerical integrals and solving a set of linear equations, which is sufficiently fast to be doable online. Several new aspects of evolutionary behaviour, such as the velocity of evolution and population diversity, can now be directly quantified.

この発表はとても難しい内容でした.途中で確率論と決定論の話をしているときに統計学の知識が不足していたためついていけませんでした.今後は統計学の勉強を少しずつして確率論的なNeural Networkのアルゴリズムを理解していく必要があると思いました.

発表タイトル       : Visual Analytics for Neuroscience-Inspired Dynamic Architectures
著者                  : Margaret Drouhard, Catherine Schuman, J. Douglas Birdwell and Mark Dean
セッション名       : Neural Networks
Abstruct            : We introduce a visual analytics tool for neuroscience-inspired dynamic architectures (NIDA), a network type that has been previously shown to perform well on control, anomaly detection, and classification tasks. NIDA networks are a type of spiking neural network, a non-traditional network type that captures dynamics throughout the network. We demonstrate the utility of our visualization tool in exploring and understanding the structure and activity of NIDA networks. Finally, we describe several extensions to the visual analytics tool that will further aid in the development and improvement of NIDA networks and their associated design method.

この発表はNeural Networkのアプリケーションについてでした.この発表についても英語がほぼ聞き取れなかったためにスライドをみるかスライドに書いてある簡単な英語を読むことでしか内容を推測できず,あまりよくわからなかったです.英語を聞き取れるようになればもう少し理解が深まると思うので英語の勉強は大切だと感じました.