Analyzing EEG Using Cloud

Computing Techniques
Tauqueer Alam
Dr. MGR Educational & Research Institute
talam2@gmail.com
Contents
1. INTRODUCTION
2. BACKGROUND
a) Brain Computer Interface
b) Gathering EEG data.
c) Neural Network
i) Artificial Neural Network
ii) Biological Neural Network
d) Cloud Computing Technique
i) Benefits of cloud computing
3. METHODOLOGY
4. APPROACH
5. FRAME WORK USED
a) Snowfall Network Setup
b) Granules Network Setup
6. PLANNING
7. CONCLUSION
8. FUTURE WORK
Introduction
 Brain computer interfaces (BCI ‘s) allow user to interact with computer via EEG signals
generated by the brain .
 Our goal is to be able to train the neural network and classify the EEG signals from multiple
users to infer there intended actions in a distributed environment.
 EEG signals gathered using electrodes placed on the user’s scalp are analyzed using a neural
network to reason about the actions initiated by the user. Our focus here is to investigate the
possibility of performing such analysis using our cloud runtime, granules.
 Granules supports two of the most dominant models for cloud computing Map Reduce and
data flow graphs.
 Since granules arranges computations on the set of available machines, even-sized cluster can
support a fairly large number of users.
 Through this project we can process three challenges:
1. We can process these EEG streams in real time.
2. We can interleave processing of EEG streams from multiple users on
the same time.
3. We can scale up to a sufficiently large number of users on a set of
available machines.
BACKGROUND
Brain Computer Interface (BCIs)

 Allows users who have lost voluntary motor control to interact with
 a computer
 BCIs work by analyzing electroencephalograms (EEGs) to interpret
 the users intent
 EEG signals are gathered in a non-invasive method
 Typing interface (Doug Hains, Elliott Forney)
 Wheelchair (Millan)
Gathering EEG data

 Non-invasive methods:-
User wears a cap which holds electrodes to the scalp
 Electrode placement followed the international 10-20 system of electrode
placement.
NEURAL NETWORK
 THEY ARE CLASSIFIED AS:-

1.Artificial neural network

2. Biological neural network
Artificial Neural Networks
 Number of input and output nodes are defined by the data
 Number of hidden units can vary:
More hidden units can model more complex data
More hidden units take longer to train
 Weights are added between input and hidden and hidden and output layers.
Biological neural network

 Also are known as the circuit of biological network.

 They are made of biological neurons or nodes, and hence used in the
neuroscience-they are ought to perform biological functions.

 We use in EEG is the artificial neural network.

Cloud computing technique
 It’s a location dependent computing whereby shared server users provides
resources, software, and data to the computers and the other device on
demand termed.
 It can be also termed as computer network diagrams
 Cloud often appears as single source of data .
Benefits of the cloud Computing

 Current BCI applications are limited, All computation happens with the users
Mobile BCI applications (such as a wheelchair) are tied to a laptop.

 A single user is classified by a single machine a dedicated machine for a

single user is under-utilized.

 Computing capabilities are limited NN complexity is limited by what can be

trained on a laptop.

 Multiple users can access the same cloud Aggregation of data. More data
leads to better trained neural network, Cloud servers are separate from the
user. Users not limited to the computational power. Possibility for massive
scaling Thousands of users can be supported simultaneously
 .
 Complex pipelines for classification can be developed
Computations can be chained through Map Reduce or graph-based
paradigms.
Methodology

 BCI (Brain computer interface) allows users to interact with a computer via EEG
signals generated by their brains, so the BCI plays vital role in this project.
 The aim’s process is developed using the Map Reduce framework.
 It uses cloud computing runtime , Granules to classify these EEG streams.
 BCI software has been developed in R (JRI) which is an interpreted language
designed for the fast computation of matrix multiplication, making it an effective
language for the development of artificial neural network.
Approach
 R backend
Optimized for matrix multiplication.
Existing code available for EEG manipulation, as well as neural network
code.
 Group of experts approach
Fits the map reduce framework - mappers classify, reducer produces
expert opinion
 3 sets of experiments:
Baseline times in R
Cloud communication overhead with Snowfall
Cloud and bridge communication overhead with Granules and JRI
Frame Works Used
Snowfall
 Parallel computing package for R.
 Builds on the Snow package.
 Executes sequential code on multiple machines simultaneously.
 Does not require strong parallel computing background.
Granules
 Lightweight cloud computing runtime.
 Java based.
 Allows user to specify run semantics - can enter a dormant state.
 while waiting for more data to become available.
JRI
 Java R Interface.
 Allows R computations to be run through Java.
 Communication is string-based.
Snowfall Network Setup
Granules Network Setup
PLANNING
 Planning aspect is based upon following....
1. Computation of EEG on more users on single machine-in this matter of
care is that the data loss occurring due to time being distributed among
all the users .
2. Sometime it leading to network congestion
3. For it R script paradigm is to reduced and also ought to depend on
parallel computing, so the network congestion could be reduced..
4. That is what we have to work on increasing network bandwidth.
5. What equipment we use today those parameters can’t be altered at the
run –time, Requirement is that to modify in way that allows to process run
time alteration as well as offline computation of cloud. For that along with
the currently used r- script java we have to work with using the c, c++
fundamentals
CONCLUSIONS
 Granules is a viable option for real-time EEG classification in the
cloud.

 While a pure R implementation can train a network more quickly,

there is no native R support for continuous streaming data.

 JRI carries a heavy overhead for communications.

 Compression is needed to scale further:

- With 150 users, we are processing 1GB of EEG signals every 83
seconds.
- At this rate, over 1TB of data is processed in a day.
FUTURE WORK

 Develop a byte-based Granules Bridge for R

 Implement an online learning algorithm

 Implement compression