Abstract: A method of slicing a deep learning model for a heterogeneous embedded system includes collecting, by a model slicing apparatus, an execution time and power consumption when each layer corresponding to one layer of a deep learning model including a plurality of layers is executed in each computing device of the heterogeneous embedded system, predicting, by the model slicing apparatus, a performance cost and a power cost when each of the layers is executed in each of the computing devices using the execution time and the power consumption, predicting, by the model slicing apparatus, a communication cost when transmitting information from each of the layers to a next layer in each of the computing devices, and slicing, by the model slicing apparatus, the plurality of layers so that different sliced layers are allocated to each of the computing devices based on the performance cost, the power cost, and the communication cost in a given execution time limit condition using a reinforcement learning model.

Abstract: A method and apparatus for partitioning memory bandwidth of a system determines state combinations of techniques for partitioning the memory bandwidth of the system based on at least one application program executed on cores of the system, determines candidate state combinations corresponding to characteristics of the application program, among the state combinations, by profiling the application program based on the state combinations, selects a final state combination for the application program by exploring a state space of the system corresponding to the candidate state combinations, and partitions the memory bandwidth of the system based on the final state combination.

Abstract: A method and apparatus for partitioning memory bandwidth of a system determines state combinations of techniques for partitioning the memory bandwidth of the system based on at least one application program executed on cores of the system, determines candidate state combinations corresponding to characteristics of the application program, among the state combinations, by profiling the application program based on the state combinations, selects a final state combination for the application program by exploring a state space of the system corresponding to the candidate state combinations, and partitions the memory bandwidth of the system based on the final state combination.

Abstract: The present invention provides an unmanned aerial vehicle that can maintain stability by changing positions of rotating rotors when one of the rotating rotors malfunctions, and a method for controlling stability of the unmanned aerial vehicle. The unmanned aerial vehicle includes: a main body; a plurality of support bars that are arranged while forming an angle with each other along a circumferential direction of the main body and extended to an outer side from the main body; a plurality of rotating rotors that are respectively provided to the support bars and generate thrust; motors that are respectively connected to the rotating rotors to drive the rotating rotors; drivers that change positions of the respective rotating rotors along the circumferential direction of the main body by moving the support bars with respect to the main body; and a controller that maintains horizontal stability of the main body by controlling the drivers.

Abstract: A lightweight architecture for an aliased memory operation includes loading data by performing an aliased memory operation, and storing data by performing an aliased memory operation, the loading of data by performed by an aliased memory operation that includes: requesting an aliased address through an HAT without a specific operation for converting an address by means of a CPU processor, converting an address input through the HAT into an aliased address in response to the request for an aliased address, stopping the aliased memory operation through the HAT and calling up a miss handler when the conversion of an address into an aliased address in response to the request for an aliased address fails, and loading data by accessing an SAT using the aliased address when conversion of an address into an aliased address succeeds.

Abstract: The present invention provides an unmanned aerial vehicle that can maintain stability by changing positions of rotating rotors when one of the rotating rotors malfunctions, and a method for controlling stability of the unmanned aerial vehicle. The unmanned aerial vehicle includes: a main body; a plurality of support bars that are arranged while forming an angle with each other along a circumferential direction of the main body and extended to an outer side from the main body; a plurality of rotating rotors that are respectively provided to the support bars and generate thrust; motors that are respectively connected to the rotating rotors to drive the rotating rotors; drivers that change positions of the respective rotating rotors along the circumferential direction of the main body by moving the support bars with respect to the main body; and a controller that maintains horizontal stability of the main body by controlling the drivers.

Abstract: Reliability-aware scheduling of processing jobs on one or more processing entities is based on reliability scores assigned to processing entities and minimum acceptable reliability scores of processing jobs. The reliability scores of processing entities are based on independently derived statistical reliability models as applied to reliability data already available from modern computing hardware. Reliability scores of processing entities are continually updated based upon real-time reliability data, as well as prior reliability scores, which are weighted in accordance with the statistical reliability models being utilized. Individual processing jobs specify reliability requirements from which the minimum acceptable reliability score is determined. Such jobs are scheduled on processing entities whose reliability score is greater than or equal to the minimum acceptable reliability score for such jobs. Already scheduled jobs can be rescheduled on other processing entities if reliability scores change.

Abstract: The described implementations relate to resource aware programming. In one case a program is obtained that is configured to perform a task in accordance with one or more quantitative metrics. An approximate version can be generated from the program. The approximate version is configured to perform the task in a manner that satisfies the one or more quantitative metrics while using fewer computer resources than the program.

Abstract: A simplified hardware management communication protocol comprises defined request packets, which are utilized to transmit requests to lower layers of management functionality or to managed resources, and it also comprises defined response packets, which are utilized to transmit responses back to the source of the request. A request packet comprises an identification of a type of device, an identifier of that device, an address of the sending entity, a session identifier, a sequence number, a function identifier, and a payload that comprises encapsulated communications or data directed to the request target. A response packet can comprise an identification of the sender of the request, a session identifier, a sequence number, a completion code identifying whether and how the request was completed, and a payload. Managed asset type specific drivers translate into communications utilizing communicational protocols that are specific to the managed assets.

Abstract: A simplified hardware management communication protocol comprises defined request packets, which are utilized to transmit requests to lower layers of management functionality or to managed resources, and it also comprises defined response packets, which are utilized to transmit responses back to the source of the request. A request packet comprises an identification of a type of device, an identifier of that device, an address of the sending entity, a session identifier, a sequence number, a function identifier, and a payload that comprises encapsulated communications or data directed to the request target. A response packet can comprise an identification of the sender of the request, a session identifier, a sequence number, a completion code identifying whether and how the request was completed, and a payload. Managed asset type specific drivers translate into communications utilizing communicational protocols that are specific to the managed assets.

Abstract: Reliability-aware scheduling of processing jobs on one or more processing entities is based on reliability scores assigned to processing entities and minimum acceptable reliability scores of processing jobs. The reliability scores of processing entities are based on independently derived statistical reliability models as applied to reliability data already available from modern computing hardware. Reliability scores of processing entities are continually updated based upon real-time reliability data, as well as prior reliability scores, which are weighted in accordance with the statistical reliability models being utilized. Individual processing jobs specify reliability requirements from which the minimum acceptable reliability score is determined. Such jobs are scheduled on processing entities whose reliability score is greater than or equal to the minimum acceptable reliability score for such jobs. Already scheduled jobs can be rescheduled on other processing entities if reliability scores change.

Abstract: Various techniques of managing a computer cluster are disclosed herein. In one embodiment, a method for managing a computer cluster includes receiving a request for a computing operation, obtaining information of utility for the computer cluster, and determining an execution profile of the computing operation identified by the received request based at least in part on the obtained information. The information includes at least one of a configuration or condition of power, heating, cooling, ventilation that supports the computer cluster. The method also includes executing the computing operation in the computer cluster in accordance with the determined execution profile.

Abstract: The described implementations relate to resource aware programming. In one case a program is obtained that is configured to perform a task in accordance with one or more quantitative metrics. An approximate version can be generated from the program. The approximate version is configured to perform the task in a manner that satisfies the one or more quantitative metrics while using fewer computer resources than the program.