Muscle Memory Authenticator: MYO Device
​
Evaluation
​
In terms of the evaluation for this project, we found that the prototype met its intended purpose. Based upon the interviews and the observations, the prototype was able to effectively function as a secure, convenient authentication device. It showed to be a better form of password authentication in comparison to current alpha-numeric methods.
Because of time and accessibility, we were not able to accumulate much information in terms of our intended audience, that is, populations that suffer from memory retention such as those with dementia. However, in our user interviews, this question was also addressed. Though we cannot extrapolate from these studies, we can predict that the device has the potential to be successful within this population group as well.
​
The following graphs depict the user evaluation interviews relating to the evaluation of the product.
​
​
​
​
​
​
For more detailed information regarding the interviews and the interview participants, please see the "front-End" page by clicking HERE.
​
From this user feedback, it is demonstrative that the product meets the three goals of our project. In each of the graphs, no user depicts strong opinion against this form of authentication. In fact, in all three graphs pertaining to the security and convenience of the product, majority of the interviewees trusted the device and found it better than alpha-numeric passwords.
​
Those that ranked the convenience at a lower scale said that if the Myo device was easier to carry (such as built-in in a phone, like the fingerprint scanner present on an iPhone), it would could become the future form of authentication. This is something we hope to look into and ways that we could improve the product designed.
​
In terms of the accuracy of the device, through our experiments we found a positive success rate. In the following plots, black lines describe the EMG signals, and the red dots are dissimilarity measurements for a window starting at that point.
​
​
​
We learned from our preliminary experiments, that touch typists produced EMG signals which were much more reliable for our Muscle Memory Authentication. The amount of noise generated from non-touch typists could not be sufficiently reduced to properly authentication those subjects.
​
​
​
While working with both these touch, and non-touch typist, we worked on applying different amounts of smoothing of the data series.
In the figures showing the series of smoothing experiments, we started off with 0 smoothing, and gradually increased the size of the smoothing window until further smoothing was detrimental to the results. Finally, we tested the system against attack from a malicious user trying to enter anther user's passphrase (their name in our test cases). In this chart, we can see that the dissimilarity remains fairly high, and doesn't produce any significant drops in dissimilarity.
​
​
​
​
