Device authentication based on behavior classification using convolution neural network

1. A method of authenticating a user on a mobile device, comprising:

receiving, at the mobile device from a remote device, a pre-trained convolutional neural network previously trained with behavioral data from a first plurality of remote users;

updating, at the mobile device, the pre-trained convolutional neural network based on behavioral data of the user gathered from a plurality of sensors of the mobile device during a first period of time to generate an updated convolutional neural network;

gathering, at the mobile device, samples of the behavioral data of the user from the plurality of sensors of the mobile device during a second period of time, each sensor generating a different number of samples during the second period of time;

normalizing, at the mobile device, the samples to obtain a same number of samples for each sensor during the period of time;

extracting, at the mobile device with the updated convolutional neural network, features from the normalized samples; and

authenticating, at the mobile device, the user based on the extracted features.

2. The method of claim 1 , in which the normalizing further comprises:

interpolating samples of each sensor to generate a pre-determined number of samples for each sensor; and

sampling the pre-determined number of samples of each sensor with at least a same starting point, a same frequency, and the same number of samples to generate the normalized samples.

3. The method of claim 2 , further comprising sampling the pre-determined number of samples of each sensor to have a same length and the same frequency based on a distribution of samples obtained from pooled behavioral data from a plurality of users.

4. The method of claim 1 , in which:

the samples are time based;

the samples have different lengths for different gestures at one sensor; and

the samples have different lengths across different sensors.

5. The method of claim 1 , further comprising generating a multi-channel one-dimensional signal from the samples, each channel corresponding to a different sensor.

6. A mobile device for authenticating a user based on behavioral data of the user, comprising:

a memory; and

at least one processor coupled to the memory, the at least one processor configured:

to receive, from a remote device, a pre-trained convolutional neural network previously trained with behavioral data from a first plurality of remote users;

to update the pre-trained convolutional neural network based on behavioral data of the user gathered from a plurality of sensors of the mobile device during a first period of time to generate an updated convolutional neural network;

to gather samples of the behavioral data of the user from the plurality of sensors of the mobile device during a second time period, each sensor generating a different number of samples during the second period of time;

to normalize the samples to have a same number of samples for each sensor during the period of time;

to extract, with the updated convolutional neural network, features from the normalized samples; and

to authenticate the user based on the extracted features.

7. The mobile device of claim 6 , in which the at least one processor is further configured to normalize by:

interpolating samples of each sensor to generate a pre-determined number of samples for each sensor; and

sampling the pre-determined number of samples of each sensor with at least a same starting point, a same frequency, and the same number of samples to generate the normalized samples.

8. The mobile device of claim 7 , in which the at least one processor is further configured to sample the pre-determined number of samples of each sensor to have a same length and the same frequency based on a distribution of samples obtained from pooled behavioral data from a plurality of users.

9. The mobile device of claim 6 , in which:

the samples are time based;

the samples have different lengths for different gestures at one sensor; and

the samples have different lengths across different sensors.

10. The mobile device of claim 6 , in which the at least one processor is further configured to generate a multi-channel one-dimensional signal from the samples, each channel corresponding to a different sensor.

11. An apparatus for authenticating a user based on behavioral data of the user, comprising:

means for receiving, at the apparatus from a remote device, a pre-trained convolutional neural network previously trained with behavioral data from a first plurality of remote users;

means for updating, at the apparatus, the pre-trained convolutional neural network based on behavioral data of the user gathered from a plurality of sensors of the mobile device during a first period of time to generate an updated convolutional neural network;

means for gathering, at the apparatus, samples of the behavioral data of the user from the plurality of sensors of the apparatus during a second period of time, each sensor generating a different number of samples during the second period of time;

means for normalizing, at the apparatus, the samples to have a same number of samples for each sensor during the period of time;

means for extracting, at the apparatus with the updated convolutional neural network, features from the normalized samples; and

means for authenticating, at the apparatus, the user based on the extracted features.

12. The apparatus of claim 11 , in which the means for normalizing further comprises:

means for interpolating samples of each sensor to generate a pre-determined number of samples for each sensor; and

means for sampling the pre-determined number of samples of each sensor with at least a same starting point, a same frequency, and the same number of samples to generate the normalized samples.

13. The apparatus of claim 12 , further comprising means for sampling the pre-determined number of samples of each sensor to have a same length and the same frequency based on a distribution of samples obtained from pooled behavioral data from a plurality of users.

14. The apparatus of claim 11 , in which:

the samples are time based,

the samples have different lengths for different gestures at one sensor, and

the samples have different lengths across different sensors.

15. The apparatus of claim 11 , further comprising means for generating a multi-channel one-dimensional signal from the samples, each channel corresponding to a different sensor.