Abstract: In a method of managing a plurality of computing paths in an artificial neural network (ANN) driven by a plurality of heterogeneous resources, resource information, preference level metrics, and a plurality of initial computing paths are obtained by performing an initialization. The resource information represents information associated with the heterogeneous resources. The preference level metrics represent a relationship between the heterogeneous resources and a plurality of operations. The initial computing paths represent computing paths predetermined for the operations. When a first event including at least one of the plurality of operations is to be performed, a first computing path for the first event is set based on the initial computing paths, the preference level metrics, resource environment, and operating environment. The resource environment represents whether the heterogeneous resources are available.

Abstract: A neural network system for predicting a polling time and a neural network model processing method using the neural network system are provided. The neural network system includes a first resource to generate a first calculation result obtained by performing at least one calculation operation corresponding to a first calculation processing graph and a task manager to calculate a first polling time taken for the first resource to perform the at least one calculation operation and to poll the first calculation result from the first resource based on the calculated first polling time.

Abstract: Provided are a system for managing a calculation processing graph of an artificial neural network and a method of managing a calculation processing graph by using the system. A system for managing a calculation processing graph of an artificial neural network run by a plurality of heterogeneous resources includes: a task manager configured to allocate the plurality of heterogeneous resources to a first subgraph and a second subgraph that are to be run, the first subgraph and the second subgraph being included in the calculation processing graph; a first compiler configured to compile the first subgraph to be executable on a first resource among the plurality of heterogeneous resources; and a second compiler configured to compile the second subgraph to be executable on a second resource among the plurality of heterogeneous resources, wherein the first subgraph and the second subgraph are respectively managed through separate calculation paths.

Abstract: An artificial neural network system for managing a task to be performed by heterogeneous resources executing an artificial neural network, the artificial neural network system including a model analyzer that receives an artificial neural network model and outputs sub-graph information generated based on the artificial neural network model including at least one of sub-graph, a detector that outputs awareness information about the heterogeneous resources, and a task manager that outputs a first request signal for performing a task with respect to each layer of first resource of the heterogeneous resources based on the sub-graph information and the awareness information, and a second request signal for performing an task with respect to each depth of a second resource of the heterogeneous resources.

Abstract: Provided are a mobile device and a data management method of the mobile device. The data management method that is processed by a central processing unit (CPU) of the mobile device includes: selecting data to be swapped from among data stored in a local memory of the mobile device, in response to a request to perform swapping on the data stored in the local memory; selecting one of at least two external memories which are located outside the mobile device and have different hierarchical levels; and swapping the selected data to the selected external memory.

Abstract: A multi-panel display device includes a first display device including a first display region and a first non-display region surrounding the first display region; a first optical member located on the first display device and including optical fibers; and a first protection film covering an outside of the first optical member.

Abstract: A method of processing data for a deep learning system driven by a plurality of heterogeneous resources is provided. The method includes, when a first task including at least one of a plurality of operations is to be performed, receiving first path information indicating a first computing path for the first task. The first computing path includes a sequence of operations included in the first task and a driving sequence of resources for performing the operations included in the first task. The method further includes setting data representation formats of the resources for performing the operations included in the first task based on data representation information and the first path information. The data representation information indicates an optimized data representation format for each of the plurality of heterogeneous resources.

Abstract: An operation method of a convolutional neural network system includes executing convolution loops to determine a correlation between at least one kernel and an input feature map; determining a maximum position on the input feature map, the maximum position corresponding to a maximum value of an output feature produced based on the convolution loops; and processing first data of a first area of the input feature map through a convolution loop using the at least one kernel, the first area being associated with the maximum position. The processing of the first data includes skipping a convolution operation of second data of a second area of the input feature map, the second area being unassociated with the maximum position.

Abstract: Provided are a neural network system, an application processor including the same, and a method of operating the neural network system. The neural network system includes an operation allocating manager configured to divide a request including a plurality of operations for neural network processing into a plurality of sub-requests each including at least one operation, allocate the plurality of sub-requests to dedicated hardware and one or more pieces of general-purpose hardware, and allocate memories for input and output for each of the plurality of sub-requests based on cost information indicating respective costs needed for hardware to access the memories; and a neural network device configured to execute the sub-requests by using the dedicated hardware and the one or more pieces of general-purpose hardware according to a result of the operation allocation. The operation allocating manager allocates different memories to at least two sub-requests of the plurality of sub-requests.

Abstract: A resource hotplug managing method of a computing system includes accessing scenario data including a plurality of scenarios, evaluating the plurality of scenarios using context information about the computing system, and controlling hotplug-in or hotplug-out of a resource included in the computing system according to a satisfied scenario among the plurality of scenarios.

Abstract: A method of operating a neural network system includes merging, by a processor, a first operation group in a first neural network and a second operation group in a second neural network, including identical operations, as a shared operation group; selecting, by the processor, a first hardware to execute the shared operation group, from among a plurality of hardware; and executing the shared operation group by using the first hardware.

Abstract: A method of controlling computing operations in a deep neural network (DNN) is provided. A network structure of the DNN including a plurality of layers is analyzed. A hyper parameter is set based on the network structure and real-time context information of a system configured to drive the DNN. The hyper parameter is used for performing an early-stop function. Depth-wise jobs are assigned to resources included in the system based on the hyper parameter to execute the depth-wise jobs. Each of the depth-wise jobs includes at least a part of the computing operations. When an early-stop event for a first layer among the layers is generated while the plurality of depth-wise jobs are executed, a subset computing operations included in at least one second layer are performed and a remainder of the computing operations are stopped. The at least one second layer is arranged prior to the first layer.

Abstract: In a method of operating an application processor to control a display device including a non-rectangular valid display region, screen information regarding the non-rectangular valid display region is received, and a plurality of pieces of valid pixel data selected based on the screen information and corresponding to the non-rectangular valid display region are output to the display device.

Abstract: A neural network system is configured to perform a parallel-processing operation. The neural network system includes a first processor configured to generate a plurality of first outputs by performing a first computation based on a first algorithm on input data, a memory storing a first program configured to determine a computing parameter in an adaptive manner based on at least one of a computing load and a computing capability of the neural network system; and a second processor configured to perform the parallel-processing operation to perform a second computation based on a second algorithm on at least two first outputs from among the plurality of first outputs, based on the computing parameter.

Abstract: A method for operating a semiconductor device includes receiving a memory request for a memory; calculating a memory bandwidth such that the memory bandwidth is at least large enough to support allocation of the memory in accordance with the memory request; creating a memory path for accessing the memory using a memory hierarchical structure wherein a memory region that corresponds to the memory path is a memory region that is allocated to support the memory bandwidth; and performing memory interleaving with respect to the memory region that corresponds to the memory path.

Abstract: A method of operating a neural network system includes parsing, by a processor, at least one item of information related to a neural network operation from an input neural network model; determining, by the processor, information of at least one dedicated hardware device; and generating, by the processor, a reshaped neural network model by changing information of the input neural network model according to a result of determining the information of the at least one dedicated hardware device such that the reshaped neural network model is tailored for execution by the dedicated hardware device.

Abstract: Disclosed are electronic devices having speech recognition functionality and operating methods of the electronic devices. Operating methods may include selectively activating or deactivating speech recognition functionality of one or more electronic devices based on comparing priorities associated with the electronic devices, respectively. Speech recognition functionality of one or more devices may be controlled based on transmitting an activation or deactivation command to the one or more devices.

Abstract: An optical member for a multi-panel display device includes a first optical member located on a first display device and including optical fibers; a second optical member located on a second display device neighboring the first display device and including optical fibers; and an optical fiber triangular bar located to overlap a region where the first and second optical members are adjacent to each other, and including optical fibers, wherein each of the first and second optical members includes a chamfer portion corresponding to the optical fiber triangular bar at the region where the first and second optical members are adjacent to each other.

Abstract: An operation method of a convolutional neural network system includes executing convolution loops to determine a correlation between at least one kernel and an input feature map; determining a maximum position on the input feature map, the maximum position corresponding to a maximum value of an output feature produced based on the convolution loops; and processing first data of a first area of the input feature map through a convolution loop using the at least one kernel, the first area being associated with the maximum position. The processing of the first data includes skipping a convolution operation of second data of a second area of the input feature map, the second area being unassociated with the maximum position.

Abstract: In a method of managing a plurality of computing paths in an artificial neural network (ANN) driven by a plurality of heterogeneous resources, resource information, preference level metrics, and a plurality of initial computing paths are obtained by performing an initialization. The resource information represents information associated with the heterogeneous resources. The preference level metrics represent a relationship between the heterogeneous resources and a plurality of operations. The initial computing paths represent computing paths predetermined for the operations. When a first event including at least one of the plurality of operations is to be performed, a first computing path for the first event is set based on the initial computing paths, the preference level metrics, resource environment, and operating environment. The resource environment represents whether the heterogeneous resources are available.