Introducing 

Prezi AI.

Your new presentation assistant.

Refine, enhance, and tailor your content, source relevant images, and edit visuals quicker than ever before.

Loading…
Transcript

Why the need of a new Biometric?

Conclusions

What we found

  • Brainprint seems very robust even with 50 people
  • If it becomes a reality:
  • What are the vulnerabilities?
  • Just more training by itself might not be enough

We learned a lot with Brainhack, but there is still much to explore

Cerebro Protocol

  • Although 2 participants only
  • Try this on more people

500 trials in 5 blocks of 5 minutes each: 30 minutes task

  • Design a new training protocol

GRE Words

Gabor Patches

  • Gradient instead

The Color Food

  • 100 randomly generated
  • Stimulates to V1
  • V1 has highly individual variability
  • Multimodal approaches?
  • 100 words
  • Low frequency
  • Different N400
  • 90 food items
  • 10 targets:

The Task

Celebrities and Food

75%

25%

  • 100 Celebrities
  • 100 Foods
  • Norming
  • 10 instances of 10 items

Press a button when you see color

Trends

The crosscorrelation starts at 0.5 in average and ends at 0.54

The target with itself has a crosscorelation > 0.9

We need an improvement of 0.3~0.4

Methods

  • 20 Subjects
  • 30 Active Electrodes
  • Cross-Correlation (x-corr)

Results

Difference in the cross correlation between the first training and the last training

Classification

  • Enrollment

...

Subject 1

Subject 50

  • Identification

BrainHack

System

Database

Subject

x-corr

6 Sessions: 2 hours each

1.

2.

3.

.

.

.

50.

0.91

0.54

0.67

0.89

200 times

for 5 categories

Hackers:

Aaron Baker

Aira Domingo

Results

Channel 30

41 out of 50 correctly classified!

Can we replicate entrainment?

Magnitude (dB)

60x2?

Frequency (Hz)

Voting Scheme (Democracy)

Voters:

30 Channels per each of the 6 categories for a total of 180 votes

. . .

1

2

3

180

1

5

3

43

1

43

5

1

3

43

1

Final vote: 1

Subjects

2

3

Vulnerabilities

Frequencies:

Subject 1: 8 votes

Subject 43: 4 votes

Subject 5: 2 votes

Subject 3: 2 votes

.

.

.

For fingerprints:

For Brainprint: ?

Entrainment

50

Accuracy Distribution

Can we combine different

sources of information?

Brainhack: Is there still a hope for brainwave biometrics?

Maria Virginia Ruiz-Blondet

15 Hz

60 Hz

180 Hz

15 Hz

30 Hz

45 Hz

1

16

11

µV

4

Accuracy

Gabor

V1

Color Targets

P300

Thank You!

And big thanks to...

Sarah Laszlo

Kenneth Kurtz

Zhanpeng Jin

Mallory Stites

Liz Sacchi

The people at Modeling Meeting

Our wonderful RAs!

Questions?

50

Hamburger

P300

.

.

.

?

Hacker's (1st Session) Target's = 0.5

Hacker's (6th Session) Target's = 0.6

0.10

Thursday, September 23, 2015

Subject 1

Learn more about creating dynamic, engaging presentations with Prezi