Methods and apparatus to generate optimized models for internet of things devices

Abstract

Example systems, methods, and apparatus to generate optimized models for Internet of Things device are disclosed. An example apparatus includes a data receiver to collect data from a sensor of an internet of things device based a first sampling frequency and a buffer having a first buffer size; a model trainer to train a model based on the data collected from the sensor; a buffer analyzer to select a second sampling frequency and to reduce the buffer to a second buffer size, the model trainer to update the model based on the second buffer size; and a platform analyzer to: determine a duration of time that that internet of things device will take to analyze sensor data based on the updated model.

1 .- 30 . (canceled) 
     
     
         31 . An apparatus to distribute an analysis model in a computing system, the apparatus comprising:
 memory;   instructions;   processor circuitry to execute the instructions to:
 obtain first data, the first data collected by a first sensor of a device based on a sampling frequency and a buffer having a first buffer size; and 
 obtain second data, the second data collected by a second sensor of the device based on the sampling frequency and the buffer having the first buffer size; 
 calculate a classification accuracy of a trained model, the model trained based on the first data collected from the first sensor and the second data collected from the second sensor; 
 cause the first buffer size to be reduced to a second buffer size based on the classification accuracy; 
 cause the model to be updated based on the second buffer size; 
 determine an amount of memory for the device to analyze sensor data based on the updated model; and 
 responsive to determining that the amount of memory is greater than a memory budget allocated to the device, cause a further update to the model with an increased accuracy tolerance. 
   
     
     
         32 . The apparatus of  claim 31 , wherein the processor circuitry is to execute instructions to extract features of the first data and the second data for use in training the model. 
     
     
         33 . The apparatus of  claim 31 , wherein the processor circuitry is to execute instructions to cause the accuracy tolerance to decrease in response to determining that the amount of memory is less than a memory budget allocated to the updated model. 
     
     
         34 . The apparatus of  claim 31 , wherein the processor circuitry is to execute instructions to cause the accuracy tolerance to decrease in response to determining that the classification accuracy of the trained model satisfies a threshold value. 
     
     
         35 . The apparatus of  claim 32 , wherein the processor circuitry is to further execute instructions to compare an accuracy of the updated model to the accuracy tolerance and trigger further changes to the sampling frequency and the buffer when the accuracy meets the accuracy tolerance. 
     
     
         36 . The apparatus of  claim 35 , wherein the processor circuitry is to execute instructions to perform feature quantization. 
     
     
         37 . The apparatus of  claim 36 , wherein the processor circuitry is to further execute instructions to perform feature reduction to identify a reduced feature set. 
     
     
         38 . The apparatus of  claim 37 , wherein the processor circuitry is to further execute the instructions to perform model quantization. 
     
     
         39 . The apparatus of  claim 37 , wherein the processor circuitry is to execute the instructions to determine, after the feature reduction, if the classification accuracy of the updated model using the reduced feature set meets the accuracy tolerance. 
     
     
         40 . The apparatus of  claim 39 , wherein the processor circuitry is to execute the instructions to add features removed from the reduced feature set back to the feature set and select different features for removal in response to a determination that the classification accuracy of the updated model does not meet the accuracy tolerance. 
     
     
         41 . A method comprising:
 obtaining first data, the first data collected by a first sensor of a device based on a sampling frequency and a buffer having a first buffer size; and   obtaining second data, the second data collected by a second sensor of the device based on the sampling frequency and the buffer having the first buffer size;   calculating a classification accuracy of a trained model, the model trained based on the first data collected from the first sensor and the second data collected from the second sensor;   causing the first buffer size to be reduced to a second buffer size based on the classification accuracy;   causing the model to be updated based on the second buffer size;   determining an amount of memory for the device to analyze sensor data based on the updated model; and   responsive to determining that the amount of memory is greater than a memory budget allocated to the device, causing a further update to the model with an increased accuracy tolerance.   
     
     
         42 . The method of  claim 41 , further including extracting features of the first data and the second data for use in training the model. 
     
     
         43 . The method of  claim 41 , further including causing the accuracy tolerance to decrease in response to determining that the amount of memory is less than a memory budget allocated to the updated model. 
     
     
         44 . The method of  claim 41 , further including causing the accuracy tolerance to decrease in response to determining that the classification accuracy of the trained model satisfies a threshold value. 
     
     
         45 . The method of  claim 42 , further including comparing an accuracy of the updated model to the accuracy tolerance and trigger further changes to the sampling frequency and the buffer when the accuracy meets the accuracy tolerance. 
     
     
         46 . The method of  claim 45 , further including performing feature quantization. 
     
     
         47 . The method of  claim 46 , further including performing feature reduction to identify a reduced feature set. 
     
     
         48 . The method of  claim 47 , further including performing model quantization. 
     
     
         49 . The method of  claim 47 , further including determining, after the feature reduction, if the classification accuracy of the updated model using the reduced feature set meets the accuracy tolerance. 
     
     
         50 . The method of  claim 49 , further including adding features removed from the reduced feature set back to the feature set and select different features for removal in response to a determination that the classification accuracy of the updated model does not meet the accuracy tolerance. 
     
     
         51 . A non-transitory computer readable storage medium comprising instructions that, when executed, cause a machine to:
 obtain first data, the first data collected by a first sensor of a device based on a sampling frequency and a buffer having a first buffer size; and
 obtain second data, the second data collected by a second sensor of the device based on the sampling frequency and the buffer having the first buffer size; 
 calculate a classification accuracy of a trained model, the model trained based on the first data collected from the first sensor and the second data collected from the second sensor; 
 cause the first buffer size to be reduced to a second buffer size based on the classification accuracy; 
 cause the model to be updated based on the second buffer size; 
 determine an amount of memory for the device to analyze sensor data based on the updated model; and 
 responsive to determining that the amount of memory is greater than a memory budget allocated to the device, cause a further update to the model with an increased accuracy tolerance. 
   
     
     
         52 . The non-transitory computer readable storage medium of  claim 51 , wherein the instructions, when executed, extract features of the first data and the second data for use in training the model. 
     
     
         53 . The non-transitory computer readable storage medium of  claim 51 , wherein the instructions, when executed, cause the accuracy tolerance to decrease in response to determining that the amount of memory is less than a memory budget allocated to the updated model. 
     
     
         54 . The non-transitory computer readable storage medium of  claim 51 , wherein the instructions, when executed, cause the accuracy tolerance to decrease in response to determining that the classification accuracy of the trained model satisfies a threshold value. 
     
     
         55 . The non-transitory computer readable storage medium of  claim 52 , wherein the instructions, when executed, compare an accuracy of the updated model to the accuracy tolerance and trigger further changes to the sampling frequency and the buffer when the accuracy meets the accuracy tolerance.