Machine learning in resource-constrained environments

Abstract

In one embodiment, a method includes executing server operations for processing a data item, wherein the server operations are based on feature values associated with the data item, executing a heuristic to determine a heuristic result value based on the feature values, determining a prediction result by a machine-learning model based on the feature values and the heuristic result value, wherein the prediction result is associated with an assessment score indicating an effectiveness of the machine-learning model, invoking a feedback function configured to update the machine-learning model based on the prediction result and the assessment score, and updating the machine-learning model when resources on the server are available.

What is claimed is: 
     
         1 . A method comprising, by a server:
 executing one or more server operations for processing a data item, wherein the one or more server operations are based on one or more feature values associated with the data item;   executing a heuristic to determine a heuristic result value based on the one or more feature values;   determining, by a machine-learning model, a prediction result based on the one or more feature values and the heuristic result value, wherein the prediction result is associated with an assessment score indicating an effectiveness of the machine-learning model;   invoking, based on the prediction result and the assessment score, a feedback function configured to update the machine-learning model; and   updating the machine-learning model when resources on the server are available.   
     
     
         2 . The method of  claim 1 , wherein the machine-learning model is trained based on the one or more feature values, a heuristic value based on the feature values, and one or more assessment scores based on corresponding past predictions generated by the machine-learning model. 
     
     
         3 . The method of  claim 2 , wherein updating the machine-learning model comprises:
 generating one or more online updates for the machine-learning model based on the assessment score, the feature values, and the heuristic value when; and   updating the machine-learning model in accordance with the online updates.   
     
     
         4 . The method of  claim 3 , further comprising determining the resources on the server are available based on a resource availability condition being satisfied, wherein the resource availability condition is based on one or more of:
 a minimum time interval between updates to the machine-learning model;   a processor utilization level; or   a number of processors of the server.   
     
     
         5 . The method of  claim 2 , further comprising:
 sending, from the server to a remote system, the assessment score, the feature values, and the heuristic value, wherein the remote system generates one or more offline updates to the machine-learning model based on the assessment score, the feature values, and the heuristic.   
     
     
         6 . The method of  claim 5 , wherein the sending occurs when the resources on the server are not available, and wherein the resources on the server being not available is determined based on a resource availability condition not being satisfied. 
     
     
         7 . The method of  claim 5 , further comprising:
 receiving, at the server from the remote system, the offline updates to the machine-learning model; and   updating the machine-learning model based on the offline updates.   
     
     
         8 . The method of  claim 1 , wherein the machine-learning model comprises a reinforcement model. 
     
     
         9 . The method of  claim 1 , wherein the one or more server operations comprise admitting an object associated with the data item to a cache, and wherein the feature values comprise one or more attributes of the object. 
     
     
         10 . The method of  claim 9 , wherein the assessment score is based on a number of requests received for the object since the object was admitted to the cache. 
     
     
         11 . The method of  claim 1 , wherein the one or more server operations comprise invalidating a cache entry, and the feature values comprise one or more attributes of the cache entry. 
     
     
         12 . The method of  claim 11 , wherein the assessment score is based on a number of requests received for the cache entry since the cache entry was invalidated. 
     
     
         13 . The method of  claim 1 , further comprising:
 executing one or more further server operations based on the prediction result.   
     
     
         14 . The method of  claim 1 , wherein the heuristic value indicates whether to perform one or more further server operations based on one or more predetermined conditions that are based on the feature values. 
     
     
         15 . The method of  claim 1 , wherein the prediction result indicates whether to perform one or more further server operations. 
     
     
         16 . One or more computer-readable non-transitory storage media embodying software that is operable when executed to:
 execute one or more server operations for processing a data item, wherein the one or more server operations are based on one or more feature values associated with the data item;   execute a heuristic to determine a heuristic result value based on the one or more feature values;   determine, by a machine-learning model, a prediction result based on the one or more feature values and the heuristic result value, wherein the prediction result is associated with an assessment score indicating an effectiveness of the machine-learning model;   invoke, based on the prediction result and the assessment score, a feedback function configured to update the machine-learning model; and   update the machine-learning model when resources on the server are available.   
     
     
         17 . The media of  claim 16 , wherein the one or more server operations comprise admitting an object associated with the data item to a cache, and wherein the feature values comprise one or more attributes of the object. 
     
     
         18 . The media of  claim 16 , wherein the one or more server operations comprise invalidating a cache entry, and the feature values comprise one or more attributes of the cache entry. 
     
     
         19 . A system comprising:
 one or more processors; and   one or more computer-readable non-transitory storage media coupled to one or more of the processors and comprising instructions operable when executed by one or more of the processors to cause the system to:
 execute one or more server operations for processing a data item, wherein the one or more server operations are based on one or more feature values associated with the data item; 
 execute a heuristic to determine a heuristic result value based on the one or more feature values; 
 determine, by a machine-learning model, a prediction result based on the one or more feature values and the heuristic result value, wherein the prediction result is associated with an assessment score indicating an effectiveness of the machine-learning model; 
 invoke, based on the prediction result and the assessment score, a feedback function configured to update the machine-learning model; and 
 update the machine-learning model when resources on the server are available. 
   
     
     
         20 . The system of  claim 19 , wherein the one or more server operations comprise admitting an object associated with the data item to a cache, and wherein the feature values comprise one or more attributes of the object.