AI model scheduling method and system for mobile application scenarios

Abstract

The present invention provides an AI model scheduling method and system for mobile application scenarios. By calculating and evaluating the overall resource usage average of each task in the task set, and then combining the task priority and the overall usage of mobile environment resources, priority downgrade and model replacement are performed. The model used by the task is adjusted in time according to the model accuracy, model size, and task priority. While ensuring the operation of the task, resource balance is maintained and maintained at 70% to 80%, ensuring that the system can still operate stably under power capacity constraints, and better balancing resource utilization and task execution quality.

Description

Technical Field

[0001] The present invention belongs to the field of AI scheduling, and in particular relates to an AI model scheduling method and system for mobile application scenarios. Background Art

[0002] Mobile edge computing, a cutting-edge technology, primarily performs computing tasks on mobile devices or temporarily deployed edge nodes. These devices, such as drones, vehicle-mounted systems, and mobile base stations, face a major challenge: resource constraints. With the rapid development of the Internet of Things and 5G technologies, the role of mobile edge computing in key areas such as civilian life and emergency response is becoming increasingly prominent. The application of artificial intelligence (AI) technology in mobile edge computing is particularly crucial, encompassing target recognition, environmental perception, and decision support. However, the complexity of corresponding AI models and the high demand for computing resources create a significant conflict with the resource constraints of mobile edge devices. Existing resource scheduling strategies, whether static resource allocation or dynamic resource adjustment, have limitations, making it difficult to efficiently match resources with AI task requirements. As the complexity and accuracy of AI models continue to increase, the demand for computing resources is also growing. Efficiently running multiple AI tasks within limited computing resources has become a pressing issue.

[0003] The shortcomings of existing technologies include: Model selection methods are often fixed, and the model size and accuracy cannot be dynamically adjusted based on the real-time needs and priorities of the task. This results in low computing resource utilization and inefficient task execution. When computing resource power consumption exceeds the power supply capacity, the model's operating status cannot be adjusted in a timely manner. Furthermore, it is impossible to effectively control the efficient use of computing resources while ensuring model accuracy, resulting in wasted resources and reduced task execution quality. Summary of the Invention

[0004] The main problem solved by the present invention is how to maximize the utilization of computing resources without affecting the accuracy of the model when the power consumption of computing resources exceeds the preset value. It provides an AI model scheduling method and system for mobile application scenarios.

[0005] In order to solve the above technical problems, the technical solutions adopted are:

[0006] An AI model scheduling method for mobile application scenarios includes the following steps:

[0007] Step 1: Build an AI model library. The attributes of each model in the model library include model resource configuration information, model accuracy, and model size.

[0008] Step 2: Obtain the tasks to be executed and sort them in order according to their execution sequence to form a task set. Each task in the task set has execution sequence and priority attributes.

[0009] Step 3: Match each task in the task set with a corresponding model based on pre-set task and AI model selection rules, and then evaluate the energy consumption of each task in the task set;

[0010] Step 4: If the energy consumption evaluation is within the resource balancing range, the AI model matching the current task is used to execute the tasks according to the execution sequence and terminate the scheduling. Resource balancing means that the average overall resource usage is between 70% and 80%.

[0011] If the resources are insufficient or sufficient after the energy consumption evaluation, the matching model of the task being executed in the task set is replaced and return to step 3 for energy consumption evaluation until the energy consumption evaluation shows that the resources are balanced. The insufficient resources refer to that the overall average resource usage exceeds 80%, and the sufficient resources refer to that the overall average resource usage is less than 70%.

[0012] Furthermore, the energy consumption assessment method is:

[0013] Step 3.1: Obtain the current overall power consumption of the mobile edge environment, the maximum capacity limits of the GPU and memory MEM, as well as the current power consumption of the mobile edge environment, and the actual resource utilization of the GPU and memory MEM;

[0014] Step 3.2: Calculate the energy consumption required to perform the task up to the current moment;

[0015] Step 3.3: Perform a minimum energy consumption game for the tasks in the task set: For each task in the task set, select a low-precision, small-volume model and calculate the minimum energy consumption value from the moment the first task is executed to the current moment;

[0016] Conduct an optimal energy consumption game for the tasks in the task set: select a model for each task in the task set based on priority. For high-priority tasks, the model is selected based on model accuracy, while for low-priority tasks, the model is selected based on model size. Calculate the optimal energy consumption value from the moment the first task in the task set is executed to the current moment.

[0017] Step 3.4: Calculate the overall mean usage of the mobile edge environment resources corresponding to the minimum energy consumption game and the optimal energy consumption game respectively;

[0018] Step 3.5: If the overall resource usage average is between 70% and 80%, use the current model configuration.

[0019] If the overall resource usage average is lower than 70% or higher than 80%, the matching model of the executing task in the task set is replaced.

[0020] Furthermore, the method to replace the matching model of the task being executed in the task set is:

[0021] If the overall resource usage average exceeds 80%, starting with the highest priority task currently being executed, a model with one level lower accuracy than the model matched by the current task will be used to replace it. The system will then determine whether the overall resource usage average is between 70% and 80%. If not, the current model will be reduced by one level. The system will then determine whether the overall resource usage average is between 70% and 80%. If not, the next high-priority task currently being executed will be selected for model replacement until the overall resource usage average is between 70% and 80%.

[0022] If the overall average resource usage is lower than 70%, starting from the lowest priority task being executed, use a model with one level higher accuracy than the model matched by the current task to replace it, and judge whether the overall average resource usage is between 70% and 80%. If not, select a model with a model volume one level larger than the model matched by the current task, and judge again whether the overall average resource usage is between 70% and 80%. If not, continue to select the next low-priority task being executed to start model replacement until the overall average resource usage is between 70% and 80%.

[0023] Furthermore, the pre-set task and AI model selection rules are:

[0024] For high-priority tasks, models are selected with the goal of high accuracy, while for low-priority tasks, models are selected with the goal of model minimization.

[0025] Furthermore, the method for calculating the energy consumption value required to perform the task up to the current moment is:

[0026] ;

[0027] in, Indicates that a fixed duration The energy consumption value required to perform tasks within the system; For the current moment , the overall mean usage of resources in the mobile edge environment;

[0028] ;

[0029] 、 and Respectively indicate at the current moment Power consumption, GPU and memory MEM actual resource utilization in mobile edge environments;

[0030] ;

[0031] in, Represents a binary decision variable: If the task The model was selected ,but Established; otherwise, Established; 、 and Respectively represent the current edge computing environment in Power consumption, GPU and memory resource usage at each moment; N represents the number of models in the model set, and M represents the number of tasks; 、 and They represent the current maximum capacity limits of the overall power consumption, GPU, and memory MEM of the mobile edge environment respectively;

[0032] ;

[0033] Indicates the first i The model is for the task set j Tasks, according to j The priority of each task Select i The accuracy of the model and size The corresponding resource demand value, include , , , respectively represent the resource requirements of the overall power consumption, GPU, and memory MEM of the current mobile edge environment of the i-th model, Indicates the j The task priority of each task, Respectively represent i The accuracy and size of the model, Indicates the accuracy and size of the selected i-th model.

[0034] The present invention also provides an AI model scheduling system for mobile application scenarios, using the steps of an AI model scheduling method for mobile application scenarios.

[0035] By adopting the above technical solution, the present invention has the following beneficial effects:

[0036] The present invention provides an AI model scheduling method and system for mobile application scenarios. The method calculates and evaluates the overall resource usage average of each task in the task set, and then combines the task priority and the overall usage of mobile environment resources to perform priority downgrade and model replacement. The model used by the task is adjusted in time according to the model accuracy, model size, and task priority. While ensuring the operation of the task, the resource balance is maintained and maintained at 70% to 80%, ensuring that the system can still operate stably under power capacity constraints, and better balancing resource utilization and task execution quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 A flow chart of the system provided by the present invention;

[0038] Figure 2 Load a Gantt chart for the AI model based on task timing. DETAILED DESCRIPTION

[0039] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0040] Figure 1 A specific embodiment of an AI model scheduling method for mobile application scenarios of the present invention is shown, including the following steps:

[0041] Step 1: Build an AI model library. The attributes of each model in the model library include the model's resource configuration information, model accuracy, and model size.

[0042] In this embodiment, the constructed AI model library includes model libraries of various precisions and sizes, covering convolutional neural networks (CNN), recurrent neural networks (RNN), long short-term memory networks (LSTM), etc. When executing model scheduling, models of different sizes and precisions are selected from the model library based on known computing resources, task timing, and the priority of the tasks to be scheduled.

[0043] A collection of models in the model library It can be expressed as ,in Indicates the first Models, N represents the number of models in the model library. The collection of resource configuration information for each model in the model library It can be expressed as , Indicates the first Resource information of a model, ,in 、 Respectively represent The accuracy and size of the model.

[0044] Step 2: Obtain the tasks to be executed and sort them in order according to their execution sequence to form a task set. Each task in the task set has execution sequence and priority attributes.

[0045] This embodiment comprehensively considers several factors that influence model scheduling, including the timing of task scheduling, the task's scheduling priority, model accuracy, model size, and computing power consumption. This means that when executing a task, both task conditions and model conditions must be considered to determine the most appropriate model for the task.

[0046] Task collections are marked as , M represents the number of tasks, and each task in the task set is a task that has been sorted in order according to the execution sequence of the task. The first task to be executed requires selecting a model of the corresponding type from the model library. Each task is represented by a tuple ,in Indicates the The priority of each task is divided into high priority and low priority. The high priority selects the model for the purpose of high accuracy, and the low priority selects the model for the purpose of model minimization; Indicates the The execution sequence of tasks.

[0047] Step 3: Match each task in the task set with a corresponding model based on the pre-set task and AI model selection rules, and then perform energy consumption evaluation on each task in the task set.

[0048] In this embodiment, the pre-set task and AI model selection rules are:

[0049] For tasks with high priority, the model is selected with the purpose of high accuracy, and for tasks with low priority, the model is selected with the purpose of minimizing the model. , from the model library Select the corresponding model.

[0050] In this embodiment, the energy consumption evaluation method is:

[0051] Step 3.1: Obtain the current overall power consumption of the mobile edge environment, the maximum capacity limits of the GPU and memory MEM, as well as the power consumption, GPU and memory MEM actual resource utilization of the mobile edge environment at the current time t; and They represent the maximum capacity limits of the overall power consumption, GPU, and memory MEM of the current mobile edge environment respectively.

[0052] 、 and Respectively indicate at the current moment Power consumption, GPU and memory MEM actual resource utilization in mobile edge environments;

[0053] (1)

[0054] in, Represents a binary decision variable: If the task The model was selected ,but Established; otherwise, Established; 、 and Respectively represent the current edge computing environment in Power consumption, GPU, and memory resource usage at each moment; N represents the number of models in the model set, and M represents the number of tasks.

[0055] Through the above formula, the model library With task set The game is formalized as a cooperative game. In this cooperative game, the players are the selected models. j tasks Priority High or low, select the resource requirement value corresponding to the i-th model with accuracy priority or size priority The relationship is defined as:

[0056] ;

[0057] Indicates the first i The model is for the task set j Tasks, according to j The priority of each task Select i The accuracy of the model and size The corresponding resource demand value, include , , , respectively represent the resource requirements of the overall power consumption, GPU, and memory MEM of the current mobile edge environment of the i-th model, Indicates the j The task priority of each task, Respectively representi The accuracy and size of the model, Indicates the accuracy and size of the selected i-th model.

[0058] Step 3.2: Calculate the energy consumption required to execute the task up to the current moment. This value measures the energy consumption of the mobile edge device over a fixed-duration task, as the power consumption of the mobile edge environment, GPU, and memory resources change. A higher energy consumption value indicates greater resource consumption, while a lower energy consumption value indicates lower resource consumption. While lower energy consumption reduces resource consumption, it also results in a loss of model accuracy.

[0059] In this embodiment, the method for calculating the energy consumption value required to execute the task from the moment when the first task is executed to the current moment is:

[0060] (2)

[0061] in, Indicates a fixed duration The energy consumption value required to perform tasks within the system; For the current moment , the overall mean usage of resources in the mobile edge environment;

[0062] (3).

[0063] Step 3.3: Perform a minimum energy consumption game for each task in the task set: For each task in the task set, select a low-precision, small-volume model and calculate the minimum energy consumption value from the moment the first task in the task set is executed to the current moment. After matching the model to each task in the task set, use Equation 2 to calculate the minimum energy consumption value.

[0064] Perform an optimal energy consumption game for each task in the task set: For each task in the task set, select a model based on priority. For high-priority tasks, select a model based on accuracy, while for low-priority tasks, select a model based on size. Calculate the optimal energy consumption from the moment the first task in the task set is executed to the current moment. Use Equation 2 to calculate the optimal energy consumption value.

[0065] The game is carried out in two steps. First, the minimum energy consumption game is carried out. For each task, a low-precision, small-volume model is selected, and the minimum energy consumption value is finally calculated. ; Then the optimal energy consumption game is carried out, and each task selects a model based on priority. High-priority tasks prioritize model accuracy to ensure the scheduling and operation of the model, while low-priority tasks prioritize model size to ensure the scheduling and operation of the model. Substitute into the above formula and monitor the overall resource usage average Is the value between 70% and 80%? If not, adjust the model. The players will achieve a win-win situation. At this time, the expected fairness between each model in the model library reaches the maximum, and the optimal energy consumption value carried by the mobile edge device is finally calculated. .

[0066] Step 3.4: Calculate the overall mean usage of the mobile edge environment resources corresponding to the minimum energy consumption game and the optimal energy consumption game respectively;

[0067] Step 3.5: If the overall resource usage average is between 70% and 80%, use the current model configuration.

[0068] If the overall resource usage average is lower than 70% or higher than 80%, the matching model of the executing task in the task set is replaced.

[0069] Step 4: If the energy consumption evaluation is within the resource balancing range, the AI model matching the current task is used to execute the tasks according to the execution sequence and terminate the scheduling. Resource balancing means that the average overall resource usage is between 70% and 80%.

[0070] If the resources are insufficient or sufficient after the energy consumption evaluation, the matching model of the task being executed in the task set is replaced and return to step 3 for energy consumption evaluation until the energy consumption evaluation shows that the resources are balanced. The insufficient resources refer to that the overall average resource usage exceeds 80%, and the sufficient resources refer to that the overall average resource usage is less than 70%.

[0071] In this embodiment, the method for replacing the matching model of the task being executed in the task set is:

[0072] If the overall resource usage average exceeds 80%, starting with the highest priority task currently being executed, a model with one level lower accuracy than the model matched by the current task will be used to replace it. The system will then determine whether the overall resource usage average is between 70% and 80%. If not, the current model will be reduced by one level. The system will then determine whether the overall resource usage average is between 70% and 80%. If not, the next high-priority task currently being executed will be selected for model replacement until the overall resource usage average is between 70% and 80%.

[0073] If the overall average resource usage is lower than 70%, starting from the lowest priority task being executed, use a model with one level higher accuracy than the model matched by the current task to replace it, and judge whether the overall average resource usage is between 70% and 80%. If not, select a model with a model volume one level larger than the model matched by the current task, and judge again whether the overall average resource usage is between 70% and 80%. If not, continue to select the next low-priority task being executed to start model replacement until the overall average resource usage is between 70% and 80%.

[0074] In this embodiment, during model scheduling, the average overall resource usage is monitored to ensure resource balance (maintained between 70% and 80%) to maximize computing resource utilization. If resource balance is not achieved, the model matching the current task must be replaced and a new model of a different size must be selected. If GPU computing resources are insufficient when running multiple models simultaneously, the priority of high-priority tasks is automatically downgraded, ensuring that model accuracy does not impact business operations. Through this downgrade, the corresponding model is then replaced, effectively ensuring overall business operations and optimizing the use of existing computing resources.

[0075] Under the premise of ensuring the normal operation of AI recognition tasks, the present invention can intelligently select AI models of different precisions and sizes for scheduling, realize the optimal utilization of computing resources, and calculate the minimum energy consumption requirements and optimal energy consumption requirements required for task execution within a fixed time, ensuring that the energy equipment carried when setting out to perform the task can meet the basic operation of the task and the optimal guarantee of the task.

[0076] The following is a detailed description using a real-time case:

[0077] Background: Consider a computing resource pool with a maximum power consumption of 400W, an 8-core GPU, and 32GB of memory. Four AI recognition tasks need to be completed within a day. The tasks have different timing requirements and priorities, and the model library offers models of varying sizes and accuracies. Resource balance should not exceed 80%.

[0078] like Figure 2 The Gantt chart shown is described as follows:

[0079] 1. AI model loading details:

[0080] (a) Task A: Starts at 8:00 AM, has high priority, and is suitable for high-precision models.

[0081] (b) Task B: starts at 9:00 am, has low priority, and is suitable for low-precision models.

[0082] (c) Task C: starts at 12:00 pm, has low priority, and is suitable for low-precision models.

[0083] (d) Task D: Starts at 2:00 PM, medium priority, suitable for high-precision models.

[0084] 2. Gantt chart description:

[0085] (a) Time axis: from 8:00 to 16:00, divided into hours.

[0086] (b) Task A: starts at 8:00, is expected to run for 6 hours, and ends at 14:00.

[0087] (c) Task B: starts at 9:00, is expected to run for 2 hours, and ends at 11:00.

[0088] (d) Task C: starts at 12:00, is expected to run for 2 hours, and ends at 14:00.

[0089] (e) Task D: starts at 14:00, is expected to run for 2 hours, and ends at 16:00.

[0090] 3. Changes in resource balance:

[0091] (a) 8:00: Task A begins. With sufficient resources, a high-precision, large-volume model is generated based on the pre-set task and AI model selection rules. The overall resource usage averages 70%.

[0092] (b) 9:00: Task B begins, and resources are balanced. A low-precision medium-volume model is matched based on the pre-set task and AI model selection rules. The overall resource usage averages 80%.

[0093] (c) 11:00: Task B ends, and the overall resource usage average drops to 70%.

[0094] (d) 12:00: Task C begins, and remaining resources are balanced. Based on pre-defined task and AI model selection rules, a low-precision, medium-volume model is matched. Overall resource usage averages rise to 85%, and power consumption exceeds 400W. As remaining resources run low, Task A's priority is lowered from high to medium, and a low-precision, small-volume model is matched for replacement scheduling. Resource balance drops to 75%, and power consumption drops to 350W.

[0095] (e) 2:00 PM: Task C ends and Task D begins. Remaining resources are balanced. Based on the pre-set task and AI model selection rules, a high-precision, medium-volume model is matched, raising resource balance to 85%. Remaining resources are insufficient. Tasks A and D are currently running, both with medium priority. Task A, with the same priority, is detected as downgraded to low-precision, small-volume. The downgrade operation continues with the next task, Task D. The model size of Task D is reduced, the running model is replaced, and the high-precision, small-volume model is used for scheduling. Resource balance drops to 80%, and power consumption drops to 320 W.

[0096] (f) 16:00: Tasks A and D finish at the same time, and the resource balance drops to 0%.

[0097] It can be seen that by selecting models of different precision and volume, the overall resource balance is maintained in the range of 70% to 80%, and the overall power consumption is controlled below 400W.

[0098] The present invention can dynamically adjust the size and accuracy of the AI model according to the timing and scheduling priority of the task, thereby significantly improving the utilization rate of GPU resources and the efficiency of task execution. By replacing the model, the use of GPU resources can be effectively controlled to ensure that the system can still operate stably under power capacity constraints. During the model replacement process, it is possible to achieve efficient utilization of GPU computing resources while ensuring the accuracy of the model, always maintaining it in the range of 70% to 80%, better balancing resource utilization and task execution quality, and calculating the optimal energy consumption requirements required to perform the task.

[0099] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An AI model scheduling method for mobile application scenarios, characterized by: The following steps are involved: Step 1: Build an AI model library. The attributes of each model in the model library include model resource configuration information, model accuracy, and model size. Step 2: Obtain the tasks to be executed and sort them in order according to their execution sequence to form a task set. Each task in the task set has execution sequence and priority attributes. Step 3: Match each task in the task set with a corresponding model based on pre-set task and AI model selection rules, and then evaluate the energy consumption of each task in the task set; Step 3.1: Obtain the current overall power consumption of the mobile edge environment, the maximum capacity limits of the GPU and memory MEM, as well as the current power consumption of the mobile edge environment, and the actual resource utilization of the GPU and memory MEM; Step 3.2: Calculate the energy consumption required to perform the task up to the current moment; Step 3.3: Perform a minimum energy consumption game for the tasks in the task set: For each task in the task set, select a low-precision, small-volume model and calculate the minimum energy consumption value from the moment the first task is executed to the current moment; Conduct an optimal energy consumption game for the tasks in the task set: select a model for each task in the task set based on priority. For high-priority tasks, the model is selected based on model accuracy, while for low-priority tasks, the model is selected based on model size. Calculate the optimal energy consumption value from the moment the first task in the task set is executed to the current moment. Step 3.4: Calculate the overall mean usage of the mobile edge environment resources corresponding to the minimum energy consumption game and the optimal energy consumption game respectively; Step 3.5: If the overall resource usage average is between 70% and 80%, use the current model configuration. If the overall resource usage average is lower than 70% or higher than 80%, the matching model of the executing task in the task set is replaced; Step 4: If the energy consumption evaluation is within the resource balancing range, the AI model matching the current task is used to execute the tasks according to the execution sequence and terminate the scheduling. Resource balancing means that the average overall resource usage is between 70% and 80%. If the resources are insufficient or sufficient after the energy consumption evaluation, the matching model of the task being executed in the task set is replaced and return to step 3 for energy consumption evaluation until the energy consumption evaluation shows that the resources are balanced. The insufficient resources refer to that the overall average resource usage exceeds 80%, and the sufficient resources refer to that the overall average resource usage is less than 70%.

2. The AI model scheduling method according to claim 1, characterized in that: The method to replace the matching model of the task being executed in the task set is: If the overall resource usage average exceeds 80%, starting with the highest priority task currently being executed, a model with one level lower accuracy than the model matched by the current task will be used to replace it. The system will then determine whether the overall resource usage average is between 70% and 80%. If not, the current model will be reduced by one level. The system will then determine whether the overall resource usage average is between 70% and 80%. If not, the next high-priority task currently being executed will be selected for model replacement until the overall resource usage average is between 70% and 80%. If the overall average resource usage is lower than 70%, starting from the lowest priority task being executed, a model with one level higher accuracy than the model matched by the current task is used to replace it, and whether the overall average resource usage is between 70% and 80% is determined. If not, a model with a volume one level larger than the model matched by the current task is selected, and whether the overall average resource usage is between 70% and 80% is determined again. If not, the next low-priority task being executed is selected to start model replacement until the overall average resource usage is between 70% and 80%.

3. The AI model scheduling method according to claim 2, characterized in that: The pre-set tasks and AI model selection rules are: For high-priority tasks, models are selected with the goal of high accuracy, while for low-priority tasks, models are selected with the goal of model minimization.

4. The AI model scheduling method according to claim 3, characterized in that: The method to calculate the energy consumption value required to perform the task up to the current moment is: ； in, Indicates a fixed duration The energy consumption value required to perform tasks within the system; For the current moment , the overall mean usage of resources in the mobile edge environment; ； 、 and Respectively indicate at the current moment Power consumption, GPU and memory MEM actual resource utilization in mobile edge environments; ； Represents a binary decision variable: If the task The model was selected ,but Established; otherwise, Established; 、 and Respectively represent the current edge computing environment in Power consumption, GPU and memory resource usage at each moment; N represents the number of models in the model set, and M represents the number of tasks; 、 and They represent the current maximum capacity limits of the overall power consumption, GPU, and memory MEM of the mobile edge environment respectively; ； Indicates the first i The model is for the task set j Tasks, according to j The priority of each task Select i The accuracy of the model and size The corresponding resource demand value, include , , , respectively represent the resource requirements of the overall power consumption, GPU, and memory MEM of the current mobile edge environment of the i-th model, Indicates the j The task priority of each task, Respectively represent i The accuracy and size of the model, Indicates the accuracy and size of the selected i-th model.

5. An AI model scheduling system for mobile application scenarios, characterized by: Use the steps of an AI model scheduling method for mobile application scenarios as described in any one of claims 1 to 4.

Images