Abstract: Various embodiments include systems and methods for generating a prompt for a large generative AI model (LXM).

Abstract: Various embodiments include systems and methods for generating a prompt for a large generative AI model (LXM). A computing device may be configured to receive a user prompt, process the received user prompt to recognize whether the prompt includes privacy information or will cause an LXM to provide a response that will reveal privacy information; and use the LXM to provide a response to the user prompt in a manner that will avoid disclosure of privacy information.

Abstract: Techniques for inertial navigation aided with multi-interval pose measurements are disclosed. The techniques can include obtaining inertial measurement unit (IMU) data from an IMU, generating, based on the IMU data, a respective pose measurement vector according to each of a plurality of machine-learning models, resulting in a plurality of pose measurement vectors, wherein each of the plurality of pose measurement vectors is associated with a respective one of multiple motion classes, and determining a device pose estimate based on the IMU data and the plurality of pose measurement vectors.

Abstract: In some aspects, a device may obtain a signal of a vibration sensor. The device may perform compressive sampling of the signal by combining each set of data points, of a plurality of sets of data points of the signal, with a compressive sampling matrix to obtain a plurality of compressive measurements of the signal. The device may determine whether the plurality of compressive measurements are indicative of a person present in a vicinity of the device. The device may cause the device to exit a low-power mode responsive to a determination that the plurality of compressive measurements are indicative of the person present in the vicinity of the device. Numerous other aspects are described.

Abstract: In some aspects, a pose tracking device may receive usability information from a sensor system that includes a plurality of sensors based on current operating conditions associated with the plurality of sensors. The pose tracking device may select a set of sensor modalities associated with the sensor system based on the usability information. The pose tracking device may select a pose tracking model based on the set of sensor modalities selected and one or more key performance indicator (KPI) requirements related to a current context associated with a pose tracking configuration for a client application. The pose tracking device may estimate a pose associated with an object using the pose tracking model based on sensor inputs associated with the one or more sensors selected from the plurality of sensors. Numerous other aspects are described.

Abstract: Systems and methods for detecting an unauthorized virtual object in a mobile device are disclosed. In an aspect, a mobile device may receive an image from a camera. The mobile device may detect a virtual object displayed on a display of a mobile device included in the image. The mobile device may receive data from an inertial measurement unit (IMU) after a movement of the mobile device. The mobile device may determine an estimated new position of the virtual object based on the data received from the IMU. The mobile device may determine an actual position of the virtual object after receiving the data from the IMU. The mobile device may determine a difference between the estimated new position and the actual position of the virtual object. The mobile device may determine whether the virtual object is the unauthorized virtual object based on the difference.

Abstract: A method for energy-efficient state change detection and classification of streaming sequential data includes receiving via a first prediction model, sequential data from a sensor. The first prediction model determines a change in an activity state based on the sequential data. An indication that the activity state has changed is transmitted to a second prediction. The second prediction model determines an updated activity state based on the sequential data. The updated activity state is sent to the first prediction model, after which the second prediction enters an inactive state.

Abstract: Systems and techniques are provided for calibrating tracking devices. For example, a process can include generating, at a primary tracking device of a tracking device network, pose information of the primary tracking device; obtaining relative pose information of a secondary tracking device of the tracking device network, wherein the relative pose information of the secondary tracking device comprises a relative position and orientation between the primary tracking device and the secondary tracking device; obtaining absolute pose information of the secondary tracking device using the pose information of the primary tracking device and the relative pose information of the secondary tracking device; and transmitting, at the primary tracking device, the absolute pose information to the secondary tracking device for performing a calibration action.

Abstract: Systems and methods for detecting an unauthorized virtual object in a mobile device are disclosed. In an aspect, a mobile device may receive an image from a camera. The mobile device may detect a virtual object displayed on a display of a mobile device included in the image. The mobile device may receive data from an inertial measurement unit (IMU) after a movement of the mobile device. The mobile device may determine an estimated new position of the virtual object based on the data received from the IMU. The mobile device may determine an actual position of the virtual object after receiving the data from the IMU. The mobile device may determine a difference between the estimated new position and the actual position of the virtual object. The mobile device may determine whether the virtual object is the unauthorized virtual object based on the difference.

Abstract: In some aspects, a device may receive measured camera motion information associated with an image. The device may generate a shared image based on the image. The device may apply, based on the measured camera motion information, an image stabilization process to the image to generate a true image. The device may store the true image in a memory associated with the device. The device may provide the shared image and the measured camera motion information to another device. Numerous other aspects are described.

Abstract: Cardiovascular or respiratory data of a subject is measured using a multi-sensor system. The multi-sensor system includes a mm-wave FMCW radar sensor, an IMU sensor, and one or more proximity sensors. The mm-wave FMCW radar sensor may be selected and its view angle adjusted based on positioning data regarding the subject obtained from the one or more proximity sensors. Each of the mm-wave FMCW radar sensor and the IMU sensor may acquire cardiovascular or respiratory measurements of the subject, and the measurements may be fused for improved accuracy and performance.

Abstract: Systems and techniques are described herein. For example, a process can include obtaining first sensor measurement data associated with a and second sensor measurement from one or more sensors. In some cases, the first measurement data can be associated with a first time and the second sensor measurement data can be associated with a second time occurring after the first time. In some aspects, the process includes determining that the first sensor measurement data and the second sensor measurement data satisfy at least one batching condition. In some examples, the process includes, based on determining that the first sensor measurement data and the second sensor measurement data satisfy the at least one batching condition, generating a sensor measurement data batch including the first sensor measurement data, the second sensor measurement data, and at least one target sensor measurement data. Ins examples the process includes outputting the sensor measurement data batch.

Abstract: Systems and techniques are described herein. For example, a process can include obtaining first sensor measurement data associated with a and second sensor measurement from one or more sensors. In some cases, the first measurement data can be associated with a first time and the second sensor measurement data can be associated with a second time occurring after the first time. In some aspects, the process includes determining that the first sensor measurement data and the second sensor measurement data satisfy at least one batching condition. In some examples, the process includes, based on determining that the first sensor measurement data and the second sensor measurement data satisfy the at least one batching condition, generating a sensor measurement data batch including the first sensor measurement data, the second sensor measurement data, and at least one target sensor measurement data. Ins examples the process includes outputting the sensor measurement data batch.

Abstract: A device includes a memory and one or more processors. The memory is configured to store instructions. The one or more processors are configured to execute the instructions to obtain electrical activity data corresponding to electrical signals from one or more electrical sources within a user's head. The one or more processors are also configured to execute the instructions to render, based on the electrical activity data, audio data to adjust a location of a sound source in a sound field during playback of the audio data.

Abstract: Innovative techniques utilize rotation vectors (RV) and game rotation vectors (GRV) for authentication are proposed. The proposed techniques enable auto-pairing of devices when the RVs/GRVs of the devices are aligned with each other. The proposed techniques also enable authentication of a user utilizing RVs/GRVs to a device.

Abstract: A device includes a memory and one or more processors. The memory is configured to store instructions. The one or more processors are configured to execute the instructions to obtain electrical activity data corresponding to electrical signals from one or more electrical sources within a user's head. The one or more processors are also configured to execute the instructions to render, based on the electrical activity data, audio data to adjust a location of a sound source in a sound field during playback of the audio data.

Abstract: A method for energy-efficient state change detection and classification of streaming sequential data includes receiving via a first prediction model, sequential data from a sensor. The first prediction model determines a change in an activity state based on the sequential data. An indication that the activity state has changed is transmitted to a second prediction. The second prediction model determines an updated activity state based on the sequential data. The updated activity state is sent to the first prediction model, after which the second prediction enters an inactive state.

Abstract: A machine learning model is trained for user activity detection and context detection on a mobile device. The machine learning model is configured to learn a statistical relationship between an always-on sensing modality of the mobile device and actual user context. Rather than user annotations, the machine learning model is enhanced and personalized for the always-on sensing modality by automated annotations obtained from non-always-on sensing modalities. The non-always-on sensing modality opportunistically provides an imperfect label of user context, where the imperfect label has a known associated probability of error.

Abstract: A method of controlling spatial audio rendering includes comparing a first heartbeat pattern to a second heartbeat pattern to generate a comparison result. The first heartbeat pattern is based on sensor information associated with a first sensor of a first sensor type, and the second heartbeat pattern is based on sensor information associated with a second sensor of a second sensor type. The method also includes, based on the comparison result, controlling a spatial audio rendering function associated with media playback.

Abstract: Cardiovascular or respiratory data of a subject is measured using a multi-sensor system. The multi-sensor system includes a mm-wave FMCW radar sensor, an IMU sensor, and one or more proximity sensors. The mm-wave FMCW radar sensor may be selected and its view angle adjusted based on positioning data regarding the subject obtained from the one or more proximity sensors. Each of the mm-wave FMCW radar sensor and the IMU sensor may acquire cardiovascular or respiratory measurements of the subject, and the measurements may be fused for improved accuracy and performance.