Convolutional neural networks deepa
Upcoming SlideShare
Loading in...5
×
 

Convolutional neural networks deepa

on

  • 1,180 views

Convolutional neural networks

Convolutional neural networks

Statistics

Views

Total Views
1,180
Views on SlideShare
1,180
Embed Views
0

Actions

Likes
2
Downloads
45
Comments
0

0 Embeds 0

No embeds

Accessibility

Categories

Upload Details

Uploaded via as Microsoft PowerPoint

Usage Rights

© All Rights Reserved

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
    Processing…
Post Comment
Edit your comment

Convolutional neural networks deepa Convolutional neural networks deepa Presentation Transcript

  • Technical seminar onConvolutional Neural Networks for P300 Detection with Application to Brain-Computer Interfaces Date: 21-03-2012 Presented By: Deepa D. Shedi 1KS08CS026 8th sem CSE
  • OVERVIEW Introduction The P300 Speller Matrix and Detection Database Existing Systems Convolutional Neural Network Input Normalization Neural Network Topology Learning Classifiers Result for P300 Detection Network Analysis Character Recognition Rate Information Transfer Rate Discussion Conclusion
  • INTRODUCTIONBrain Computer Interface :A Brain-Computer Interface (BCI) is a specific type of human-computer interface that enables the direct communicationbetween human and computers by analyzing brainmeasurements.Eelectroencephalogram(EEG):It is a measure of brain’s voltage fluctuations as detected fromscalp electrodes. It captures typical patterns of P300 signals.P300 :EPRs are voltage fluctuations in the EEG induced with in thebrain that are time locked to sensory or motor events.The P300 is positive bump in the ERP named so because itstarts about 300 milliseconds after an event. 1 View slide
  • THE P300 SPELLER MATRIX AND DETECTIONThe two classification problems.  P300 detection.  Character Recognition. Fig 1: P300 detection Fig 2 : Character recognition. 2 View slide
  • DATABASEData set contains a complete record of P300 evokedpotentials from two subjects.The subject was presented with a matrix of size 6X6.2 out of 12 intensifications for rows/columns.The number of samples for both databases and for eachsubject is presented. Table 1: Database Size for Each Subject 3
  • EXISTING SYSTEMSThis section describes some of the best techniques that have beenproposed during the III BCI competition.  Support Vector Machine(SVM)  Band-pass Filtering  Frequency Filtering And Principal Component Analysis (PCA)  Gradient Boosting Method  Component Classifier Linear Discriminant Analysis (LDA) 4
  • CONVOLUTIONAL NEURAL NETWORKThe classifiers that are used for the detection of P300responses are based on a convolutional neural network(CNN).Neural network is a multilayer perceptron (MLP).Neural network is used for object recognition andhandwriting character recognition.A classifier based on a CNN seems to be a good approachfor EEG classification 5
  • INPUT NORMALIZATION Steps in Normalization: Step 1: Subsampling of EEG signal to reduce the size of the data to analyze and divided by two. Step 2: Bandpass filtering of signal to keep only relevant frequencies. Normalized Signal :Fig. 3. Electrode map. 6
  • NEURAL NETWORK TOPOLOGY Fig. 4. Neural Network Architecture. 7
  • CONTD.The network topology is described as follows:L0: The input layer. Ii,j with 0 ≤ i < Nelec and 0 ≤ j < Nt.L1: The first hidden layer is composed of Ns maps. We defineL1Mm, the map number m. Each map of L1 has the size Nt.L2: The second hidden layer is composed of 5Ns maps. Eachmap of L2 has six neurons.L3: The third hidden layer is composed of one map of 100neurons. This map is fully connected to the different maps of L2.L4: The output layer. This layer has only one map of twoneurons, which represents the two classes of the problem (P300and no P300). This layer is fully connected to L3. 8
  • LEARNINGA neuron in the network is defined by n(l, m, j).This sigmoid function for only one map in the layer:Convolution of the input signal for first two hidden layers: The classical sigmoid function is used for the two last layers: 9
  • LEARNINGσm1(j) represents the scalar product between a set of inputneurons and the weight connections between these neuronsand the neuron number j in the map m in the layer l. σm1(j) for the four layers.For L1 :This layer aims at finding the best electrodes combination forthe classification.For L2 :This layer translates subsampling and temporal filters. 10
  • LEARNINGFor L3 :In this layer, each neuron has NsNf +1 input weights. L3contains 100(5*6*Ns) input connections.For L4 : Each neuron of L4 is connected to each neuron of L3. 11
  • LEARNINGLearning rate ϒ for layers L1 and L2 :Learning rate ϒ for layers L1 and L2 :The detection of a P300 wave : 12
  • CLASSIFIERSSingle classifiers: • CNN-1 • CNN-2a • CNN-2b • CNN-3Multiclassifiers : • MCNN-1 • MCNN-2 • MCNN-3 13
  • RESULT FOR P300 DETECTION Results of the P300 Detection for Subject A : Table 2: Results Obtained After the P300 detection of Subject A. Results of the P300 Detection for Subject B : 14 Table 3: Results Obtained After the P300 detection of Subject B.
  • RESULT FOR P300 DETECTION Measures for evaluating the quality of results : Subject B allows getting better results for the classification. 15
  • NETWORK ANALYSISSpatial filters obtained with CNN-1 for subject A. Fig. 5. Spatial filters obtained with CNN-1 for subject A.Spatial filters obtained with CNN-1 for subject B. 16Fig.6. Spatial filters obtained with CNN-1 for subject B.
  • NETWORK ANALYSISElectrode Ranking : Table 4: Ranking of ElectrodesCumulated probabilities of the P300 detection:The coordinate of the character are defined by : 17
  • CHARACTER RECOGNITION RATETable 5:Character Recognition Rate (in Percent) for the Different Classifiers 18
  • INFORMATION TRANSFER RATEInformation Transfer Rate :T is the time needed to recognize one character. T isdefined by :Best recognition rate when only 10 epochs are used. 19
  • INFORMATION TRANSFER RATEFig. 7. ITR (in bits per minute) in relation to the number of epochs. 20
  • DISCUSSIONThe database has two main interests. • It forces the system to reach the limit of the P300 detection. • It is an excellent challenge for the machine learning community.Steady-state visual evoked potentials (SSVEPs), the userhas to focus on some visual Stimuli.The interest of convolutional neural networks is double. • It allows a high performance in the classification. • They can allow deeper analysis of brain activity. 21
  • DISCUSSIONIn the P300 speller, it is possible that the subject may nothave always focused on the expected target. Table 6:Confusion of Character RecognitionComparisonTable 7:Comparison of the Recognition Rate and the ITR with Other Results. 22
  • CONCLUSIONThis model is based on a convolutional neural network.It outperforms the best method in two situations: •When the number of electrodes is restricted to 8. • when the number of considered epochs is 10.The recall of the P300 detection is the main feature thatdictates the overall performance of the P300 speller.The detection of P300 waves remains a very challengingproblem for both the machine learning and neurosciencecommunities. 23
  • REFERENCESC.W. Anderson, S.V. Devulapalli, and E.A. Stolz, “Determining Mental State fromEEG Signals Using Parallel Implementations of Neural Networks,” Proc. IEEEWorkshop Neural Networks for Signal in Processing, pp. 475-483, 1995.K. Chellapilla, S. Puri, and P.Y. Simard, “High Performance Convolutional NeuralNetworks for Document Processing,” Proc.10th Int’l Workshop Frontiers inHandwriting Recognition, 2006.B. Blankertz, K.-R. Muller, G. Curio, T.M. Vaughan, G. Schalk, J.R.Wolpaw, A.Schlogl, C. Neuper, G. Pfurtscheller, T. Hinterberger, M. Schroder, and N. Birbaumer,“The BCI Competition 2003: Progress and Perspectives in Detection andDiscrimination of EEG Single Trials,” IEEE Trans. Biomedical Eng., vol. 51, no. 6, pp.1044-1051, June 2004.D.J. Krusienski, E.W. Sellers, D. McFarland,T.M. Vaughan, and J.R. Wolpaw,“Toward Enhanced P300 Speller Performance,” J. Neuroscience Methods, vol. 167,pp. 15-21, 2008.G. Schalk, D.J. McFarland, T. Hinterberger, N. Birbaumer, and J. Wolpaw, “BCI2000:A General-Purpose Brain-Computer Interface(BCI) System,” IEEE Trans. BiomedicalEng., vol. 51, no. 6, pp. 1034-1043, June 2004.
  • Thank You!!Questions??? 27