University of California

Hewlett-packard Development Company, L.P.

Hewlett Packard Enterprise Development LP

Prasant Mohapatra

Kyu Han Kim

Jindan Zhu

Parth H Pathak

Amit Pande

Shaxun Chen

Souvik Sen

Li Zhang

Kai Zeng

Muchen Wu

Setareh Rafatirad

Han Wang

Yunze Zeng

Houman Homayoun

The embodiments disclose a system and method including a random number generator coupled to the randomized machine learning-based malware detector configured for determining changes of settings and selections of parameters, candidate classifiers integrated with the randomized machine learning-based malware detector and configured to be initiated by a random number to avoid transferable learning, a set of feature combinations for random feature selection including monitoring granularity and detection prediction latency and a random number for identifying a set of feature combinations that minimize the overhead and maintain enough variance in data for baffling the malware adversarial attacks.

Adversarial resilient malware detector randomization method and devices

The disclosed embodiments relate to the design of a system that uses an accelerometer in a mobile device to detect hotwords, which activate a voice interface to recognize subsequent voice input. During operation, the system gathers samples comprising readings from the accelerometer in the mobile device. Next, the system calculates features from the gathered samples. The system then classifies the calculated features using a classifier, which has been trained to detect hotwords based on calculated features. Finally, if the classifier detects one or more hotwords, the system causes the mobile device to launch a voice-control system that recognizes subsequent voice input received from an audio microphone.

Energy-efficient, accelerometer-based hotword detection to launch a voice-control system

The disclosed embodiments relate to the design of a system that identifies a person. During operation, the system receives channel state information (CSI) for a set of orthogonal frequency division modulation (OFDM) subcarriers while the person moves in a region that includes two or more nodes that use the set of OFDM subcarriers to communicate with one another. Next, the system analyzes the CSI to obtain an analysis result. The system then determines the identity of the person based on the analysis result.

WiFi-based person-identification technique for use in smart spaces

Systems and techniques for user authentication are described. A portable electronic device can include an imaging sensor and a motion sensor. Multiple images of a face of a user can be acquired by using an imaging sensor. Motion information can be obtained by using a motion sensor. Next, a first motion vector can be determined based on the multiple images, and a second motion vector can be determined based on the motion information. The user can then be authenticated based on (1) determining a similarity measure between the first motion vector and the second motion vector, and (2) performing facial recognition on at least one of the multiple images. Note that the similarity measure can be used to detect a virtual-camera attack. Specifically, authentication can fail if the similarity measure is less than a threshold.

Sensor-assisted user authentication

According to an example, a signal space based navigation apparatus may receive traces of wireless signal observations and corresponding pedestrian deadreckoning (PDR) traces between different ones of the wireless signal observations from a plurality of user devices. A plurality of wireless clusters may be generated based on the traces of wireless signal observations. A PDR displacement vector may be generated between two or more of the wireless clusters based on the two or more of the wireless clusters and a subset of the PDR traces corresponding to the two or more of the wireless clusters. A sensing map may be generated based on the plurality of wireless clusters and a plurality of PDR displacement vectors including the PDR displacement vector.

Signal space based navigation

During a technique for authenticating a user using a portable electronic device (such as a smartphone) while preventing successful 2D media attacks and virtual-camera attacks, the portable electronic device provides an instruction to a user to move the portable electronic device relative to the user's face. Then, using an imaging sensor in the portable electronic device, the portable electronic device acquires, a series of images while the portable electronic device is at different positions or orientations relative to the user's face, where the series of images include at least two images relative to a symmetry axis of the user's face (such as the midline). Moreover, the portable electronic device determines a change in a sign of a nose angle in at least the two images. Next, the portable electronic device authenticates the user based on at least one of the series of images using a facial-recognition technique.

Sensor-assisted biometric authentication for smartphones

Examples disclosed herein relate to utilizing Bluetooth beacons for location determination of a mobile device. A Wi-Fi signal scan and a Bluetooth beacon scan are initiated at the mobile device. A location of the mobile device is determined based on at least on at least one Bluetooth beacon detectable by the mobile device, where the at least one Bluetooth beacon is transmitted from a first location, and where the first location is the location of the mobile device. A user of the mobile device is prompted for location information input, in response to determining that no Bluetooth beacons are detectable by the mobile device or when the at least one Bluetooth beacon detected is determined to be transmitted from a second location different from the location of the mobile device.

Growing community of inventors

Davis, CA, United States of America

Prasant Mohapatra