Disclosed are an artificial intelligence learning method and an operating method of a robot using the same. An on-screen label is generated based on image data acquired through a camera, an off-screen label is generated based on data acquired through other sensors, and the on-screen label and the off-screen label are used in learning for action recognition, thereby raising action recognition performance and recognizing a user's action even in a situation in which the user deviates from a camera's view.

An artificial intelligence robot for managing movement of an object using artificial intelligence includes a driving motor, a camera configured to acquire image data, a memory configured to store an object recognition model used to recognize the object from the image data and store a delivery location inference model for inferring a delivery location of the recognized object and a processor configured to recognize the object from the image data using the object recognition model, determine the delivery location of the recognized object from identification data of the recognized object using the delivery location inference model, and control the driving motor to move the artificial intelligence robot to the determined delivery location.

A system and method for assisted-learning with a portable computing device that includes requesting that a user complete a mathematical challenge by arranging real-world objects in an environment to form an arrangement according to the mathematical challenge, optionally receiving an input from the user that the arrangement is complete, activating a camera of a portable computing device located in the environment with the user to capture an image of the arrangement, wherein the image is received from the portable computing device over a network, evaluating the arrangement using a visual recognition engine to determine whether the arrangement successfully completes the mathematical challenge, and providing at least one of a visual feedback and an audible feedback to the user.

Methods, apparatuses, and non-transitory machine-readable media associated with data transmission are described. Data transmission management can include receiving, from an edge device via a radio at a first device, instructions associated with data transmission between a second device in communication with the first device and a cloud service in communication with the first device. Data transmission management can also include managing, at the first device and based on the instructions from the edge device, data received from a memory resource of the second device for transmission to the cloud service and data received from the cloud service for transmission to the memory resource of the second device. Data transmission management can further include enabling transmission of some, none, or all of the data between the cloud service and the memory resource of the second device and vice versa based on the management of the data.

This patent concerns novel technology for detecting backdoors of neural network, particularly deep neural network (DNN), classifiers. The backdoors are planted by suitably poisoning the training dataset, i.e., a data-poisoning attack. Once added to input samples from a source class (or source classes), the backdoor pattern causes the decision of the neural network to change to a target class. The backdoors under consideration are small in norm so as to be imperceptible to a human, but this does not limit their location, support or manner of incorporation. There may not be components (edges, nodes) of the DNN which are dedicated to achieving the backdoor function. Moreover, the training dataset used to learn the classifier may not be available. In one embodiment of the present invention which addresses such challenges, if the classifier is poisoned then the backdoor pattern is determined through a feasible optimization process, followed by an inference process, so that both the backdoor pattern itself and the associated source class(es) and target class are determined based only on the classifier parameters and a set of clean (unpoisoned attacked) samples from the different classes (none of which may be training samples).

This patent concerns novel technology for detecting backdoors of neural network, particularly deep neural network (DNN), classifiers. The backdoors are planted by suitably poisoning the training dataset, i.e., a data-poisoning attack. Once added to input samples from a source class (or source classes), the backdoor pattern causes the decision of the neural network to change to a target class. The backdoors under consideration are small in norm so as to be imperceptible to a human, but this does not limit their location, support or manner of incorporation. There may not be components (edges, nodes) of the DNN which are dedicated to achieving the backdoor function. Moreover, the training dataset used to learn the classifier may not be available. In one embodiment of the present invention which addresses such challenges, if the classifier is poisoned then the backdoor pattern is determined through a feasible optimization process, followed by an inference process, so that both the backdoor pattern itself and the associated source class(es) and target class are determined based only on the classifier parameters and a set of clean (unpoisoned attacked) samples from the different classes (none of which may be training samples).

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating and executing biological convolutional neural network layers. One of the methods obtaining a network input; and processing the network input using a neural network to generate a network output, wherein the neural network is configured to perform operations comprising: generating a layer input to a convolutional neural network layer based on the network input; and generating a layer output of the convolutional neural network layer based on the layer input, comprising applying a convolutional kernel to the layer input, wherein the convolutional kernel corresponds to a specified neuron in a brain of a biological organism and values of parameters of the convolutional kernel are based on synaptic connectivity between the specified neuron and each of a plurality of other neurons in the brain of the biological organism.

Systems and methods for controlling and implementing a concurrent multi-agent BDI architecture to control complex systems in real time and exhibiting high parallelism. The system comprises BDI agent modules storing a BDI state of the whole system, an emotional and an internal the BDI agent modules and a world model. Each BDI agent module concurrently evaluates the activation of its potential goals and prioritizes its intentions. Each BDI agent module may be hardware-implemented, software-implemented or hybrid-domain-implemented, depending on whether they are executed on application-specific hardware, general purpose hardware or a combination thereof. The domain of each BDI agent might be migrated in real time during operation, optimizing resources and improving performance.

An operation control device for a robot comprises: an input part inputting at least one operation candidate, and a captured image including an object to be processed; a first learning device that has finished learning performed according to first learning data to output a first evaluation value indicating evaluation of each operation candidate when the robot performs a first processing operation upon input of the captured image and the operation candidate; a second learning device that has finished learning performed according to second learning data which differs from the first learning data, to output a second evaluation value indicating evaluation of each operation candidate when the robot performs a second processing operation upon input of the captured image and the operation candidate; and an evaluation part that, based on at least one of the first evaluation value and the second evaluation value, calculates a command value.

A robot according to an embodiment of the present disclosure includes a body which is provided with a battery therein, a head connected to a front or an upper side of the body, a mouth formed on one side of the head and include a fixed portion and a rotatable portion disposed below the fixed portion, a mouth driver configured to rotate the rotatable portion in a vertical direction, a biometric information sensor disposed inside the mouth and exposed to the outside during the lower rotation of the rotatable portion, and a processor configured to acquire health state information of a user through the biometric information sensor.