In a conventional method of detecting a user position from footsteps of a user collected by microphones installed in a user's home, it is necessary to precisely align the microphones in advance to an extent of coordinates. This is inconvenient for both system and user. Provided is an information processing apparatus including an acquisition unit that acquires sound data recorded by a plurality of microphones installed in arbitrary places and relative positions, from the microphones, of footsteps included in the sound data, and a learning unit that generates a learning model by learning training data including the sound data as input and the relative positions as correct answers. As a result, precise positioning of the microphones in the user's home becomes unnecessary, and the user position can be detected in a more convenient manner for both the system and the user.

A system and method for controlling access to an on-device machine learning model without the use of encryption is described herein. For example, a request is received from an application executing on a device of a user. The request is to download a machine learning model to the device that enables a feature of the application, and the request includes information associated with the user and/or the device. The information is used to create an obfuscation key, and a derivative model can be generated using a reference copy of the machine learning model and the obfuscation key. The derivative model and the obfuscation key are then sent to the application. When the obfuscation key is provided to the derivative model at runtime, values derived from the obfuscation key are provided as additional inputs that enable the derivative model to function properly.

Techniques are disclosed in which a computer system generates a transaction network graph from an initial set of transactions including known labels and attributes. The computer system may generate first and second matrices using first and second graph embedding routines from a training set of transactions that includes a first subset of transactions in the network graph. The first routine is based on anomalies in related transactions occurring at nodes in the transaction network graph that are multiple hops away while the second routine is based on anomalies in neighborhoods of similar transactions. In some embodiments, the computer system generates a final embedded matrix from the first and second matrices and uses the final matrix and a testing set of transactions that includes a second subset of transactions in the graph to train a machine learning model, where the trained model usable to determine whether unlabeled transactions are anomalous.

A computer-implemented method of recommending information sources is provided. The method comprising collecting raw user log data of a browser user and transforming the raw user log data to remove data that is specific to the user, wherein the transformed user log data comprises only general pages visited. A number of key page mappings are appended to the transformed user log data. A machine learning model determines a user interaction function according to the transformed user log data and generates a number of web page recommendations, wherein each web page recommendation has a respective probability of engagement based on the user interaction function. The web page recommendations are then displayed to the user on an interface.

A hearing assistance device including: a microphone arranged to receive sound of an environment in which the hearing assistance device is located; a wireless communication device arranged to wirelessly communicate data with an external device; a controller operably connected with the microphone and arranged to process the received sound using sound processing settings that have been determined using a trained machine learning processing model; and a speaker operably connected with the controller and arranged to output the processed sound. The sound processing settings have been determined using the trained machine learning processing model based on a hearing response of a user and one or more properties of the environment.

3D NOR flash memory devices having vertically stacked memory cells are provided. In one aspect, a memory device includes: a word line/bit line stack with alternating word lines and bit lines separated by dielectric layers disposed on a substrate; a channel that extends vertically through the word line/bit line stack; and a floating gate stack surrounding the channel, wherein the floating gate stack is present between the word lines and the channel, and wherein the bit lines are in direct contact with both the channel and the floating gate stack. Techniques for configuring the memory device for neuromorphic computing are provided, as are methods of fabricating the memory device.

Disclosed herein are methods and systems for computing a launch power for an optical node by collecting data for an optical network segment and inputting the collected data and first power spectral density values into a machine learning model which are used to compute a first non-linear interference value. A first generalized-optical signal-to-noise ratio value is computed using the computed first non-linear interference value and amplified spontaneous emission values. At least one second generalized-optical signal-to-noise ratio value is computed using at least one second non-linear interference value, computed using at least one second power spectral density values, and the amplified spontaneous emission values. A highest generalized-optical signal-to-noise ratio value is determined by comparing the first generalized-optical signal-to-noise ratio value and the at least one second generalized-optical signal-to-noise ratio value. A launch power is computed using the power spectral density values associated with the highest generalized-optical signal-to-noise ratio.

A training technique trains a neural network having sparsely-activated sub-networks. It does so by processing plural batches of training data in two respective passes of the neural network, yielding first prediction information and second prediction information. For each batch, the technique randomly assigns different sub-networks in the first and second passes of the neural network to process the batch. Over the course of training, the technique attempts to minimize loss information, which describes the difference between the first prediction information and ground-truth information, and the difference between the second prediction information and the ground-truth information. Simultaneously, the technique attempts to minimize divergence information, which describes the divergence of the first prediction information from the second prediction information (and vice versa). The technique can produce an inference-stage model by arbitrarily selecting at least one of the trained sub-networks in the neural network, for use in a production system.

A computer implemented method for the automated analysis or use of data is implemented by a voice assistant. The method comprises the steps of: (a) storing in a memory a structured, machine-readable representation of data that conforms to a machine-readable language (‘machine representation’); the machine representation including representations of user speech or text input to a human/machine interface; and (b) automatically processing the machine representations to analyse the user speech or text input.

A method for knowledge answering includes: receiving a user question entered by a client, and obtaining a target insurance rule matching the user question from a plurality of insurance rules; searching for an answer in an insurance knowledge graph based on the target insurance rule, in which the insurance knowledge graph is generated based on the plurality of insurance rules; and returning the answer to the client.