US-12619925-B2 - Context-level federated learning

Abstract

A method performed by a local client computing device is provided. The method includes training a local model using data from the local client computing device, resulting in a local model update; sending the local model update to a central server computing device; receiving from the central server computing device a first updated global model; determining that the first updated global model does not meet a local criteria, wherein determining that the first updated global model does not meet a local criteria comprises computing a score based on the first updated global model, wherein the score exceeds a threshold; in response to determining that the first updated global model does not meet a local criteria, sending to the central server computing device context information; and receiving from the central server computing device a second updated global model.

Inventors

• Perepu SATHEESH KUMAR
• SARAVANAN M
• Senthamiz Selvi ARUMUGAM

Assignees

• TELEFONAKTIEBOLAGET LM ERICSSON (PUBL)

Dates

Publication Date

20260505

Application Date

20200116

Claims (20)

1. 1 . A method performed by a local client computing device, the method comprising: training a local model using data from the local client computing device, resulting in a local model update; sending the local model update to a central server computing device; receiving from the central server computing device a first updated global model; determining that the first updated global model does not meet a local criteria; in response to determining that the first updated global model does not meet a local criteria, sending to the central server computing device context information; and receiving from the central server computing device a second updated global model.
2. 2 . The method of claim 1 , wherein determining that the first updated global model does not meet a local criteria comprises computing a score based on the first updated global model, wherein the score exceeds a threshold.
3. 3 . The method of claim 1 , wherein the computed score comprises an error information in a prediction and wherein the context information includes the computed score and is encoded prior to sending by using an auto-encoder.
4. 4 . The method of claim 1 , wherein the local model predicts items for stocking a store corresponding to the local client computing device to optimize supply chain management, and the context information comprises one or more of: an area of a market the store covers, a weight of the predicted items, a number of pre-orders for the predicted items, seasonal events related to the predicted items, and a price of the predicted items.
5. 5 . The method of claim 1 , wherein the local model predicts a travel itinerary for a user corresponding to the local client computing device to optimize travel plan selection.
6. 6 . A method performed by a central server computing device, the method comprising: receiving from a local client computing device a local model update; training a global model using the local model update, resulting in a first updated global model; sending to the local client computing device the first updated global model; receiving from the local client computing device a context information; training the global model using the local model update and the context information, resulting in a second updated global model; and sending to the local client computing device the second updated global model.
7. 7 . The method of claim 6 , wherein training the global model using the local model update and the context information, resulting in a second updated global model comprises using a modified objective function to incorporate the context information.
8. 8 . The method of claim 6 , wherein the context information includes an error information in a prediction from the local client computing device and the received context information is encoded by using an auto-encoder.
9. 9 . The method of claim 6 , wherein the local model predicts items for stocking a store corresponding to the local client computing device to optimize supply chain management, and the context information comprises one or more of: an area of a market the store covers, a weight of the predicted items, a number of pre-orders for the predicted items, seasonal events related to the predicted items, and a price of the predicted items.
10. 10 . The method of claim 6 , wherein the local model predicts a travel itinerary for a user corresponding to the local client computing device to optimize travel plan selection.
11. 11 . A local client computing device comprising: a memory; and a processor, wherein said processor is configured to: train a local model using data from the local client computing device, resulting in a local model update; send the local model update to a central server computing device; receive from the central server computing device a first updated global model; determine that the first updated global model does not meet a local criteria; in response to determining that the first updated global model does not meet a local criteria, send to the central server computing device context information; and receiving from the central server computing device a second updated global model.
12. 12 . The local client computing device of claim 11 , wherein determining that the first updated global model does not meet a local criteria comprises computing a score based on the first updated global model, wherein the score exceeds a threshold.
13. 13 . The local client computing device of claim 11 , wherein the computed score comprises an error in a prediction and wherein the context information includes the computed score and is encoded prior to sending by using an auto-encoder.
14. 14 . The local client computing device of claim 11 , wherein the local model predicts items for stocking a store corresponding to the local client computing device to optimize supply chain management, and the context information comprises one or more of: an area of a market the store covers, a weight of the predicted items, a number of pre-orders for the predicted items, seasonal events related to the predicted items, and a price of the predicted items.
15. 15 . The local client computing device of claim 11 , wherein the local model predicts a travel itinerary for a user corresponding to the local client computing device to optimize travel plan selection.
16. 16 . A central server computing device comprising: a memory; and a processor, wherein said processor is configured to: receive from a local client computing device a local model update; train a global model using the local model update, resulting in a first updated global model; send to the local client computing device the first updated global model; receive from the local client computing device context information; train the global model using the local model update and the context information, resulting in a second updated global model; and send to the local client computing device the second updated global model.
17. 17 . The central server computing device of claim 16 , wherein training the global model using the local model update and the context information, resulting in a second updated global model comprises using a modified objective function to incorporate the context information.
18. 18 . The central server computing device of claim 16 , wherein the context information includes error information from the local client computing device and the received context information is encoded by using an auto-encoder.
19. 19 . The central server computing device of claim 16 , wherein the local model predicts items for stocking a store corresponding to the local client computing device to optimize supply chain management, and the context information comprises one or more of: an area of a market the store covers, a weight of the predicted items, a number of pre-orders for the predicted items, seasonal events related to the predicted items, and a price of the predicted items.
20. 20 . The central server computing device of claim 16 , wherein the local model predicts a travel itinerary for a user corresponding to the local client computing device to optimize travel plan selection.

Description

CROSS REFERENCE TO RELATED APPLICATION(S) This application is a 35 U.S.C. § 371 National Stage of International Patent Application No. PCT/IN2020/050047, filed Jan. 16, 2020. TECHNICAL FIELD Disclosed are embodiments related to context-level federated learning. BACKGROUND Machine learning techniques have become commonplace in business and other areas of the modern economy. For example, globalization and digitalization have led to an explosion of manufacturers, dealers, suppliers, and distribution areas. These same forces also increase the need for unfaltering management e.g. of a complex network such as a global or regional supply chain. Accordingly, businesses are looking for ways to improve visibility into their operations, including across departments and beyond the boundaries of individual locations. Machine learning provides one such possibility, and can help to streamline certain business processes and models for amplifying business growth. Real-time monitoring of customer demand, critical business events, key-performance indicators (KPIs), and business transactions, for instance, can improve control and cost efficiency and also render a business more agile by enhancing its responsiveness to complex situations. Ultimately, this can improve customer experience, such as minimizing customer impact when local problems arise affecting supply distribution. Just as accurate forecasting can improve a business's potential, inaccurate forecasting can plague businesses everywhere, especially ones in industries like healthcare, consumer goods, retail, automotive, logistics, etc. where such forecasting can help maintain a competitive edge. Instability in demand is driving businesses to adopt tools to gain real-time forecast ability, so that they are able to respond to highly volatile markets that have no tolerance for bottlenecks. Investing in real-time supply chain analytics can help businesses gain key inventory and forecasting metrics to combat the volatility of markets. Increasing operational costs inevitably affect budgets, working capital, cost of end-product, and cash flow. Systematic and timely analysis of critical data can help to achieve cost optimization in areas including material sourcing, load planning, fleet sizing, route and freight costing. Detailed analysis of finances, capability constraints, and potential supplier risk can minimize monetary loss in the later stages of supply chain management. Machine learning makes it possible to discover patterns, such as in supply chain data, by relying on algorithmic methods that quickly pinpoint the most influential factors to a supply networks' success, while constantly learning in the process. Discovering new patterns in data, such as supply chain data, has the potential to improve many businesses. Machine learning algorithmic methods are finding these new patterns without the need for manual intervention or the definition of a taxonomy to guide the analysis. The algorithmic methods iteratively query data, with many using constraint-based modeling to find the core set of factors with the greatest predictive accuracy. Key factors influencing inventory levels, supplier quality, demand forecasting, procure-to-pay, order-to-cash, production planning, transportation management, and more, are becoming known for the first time. New knowledge and insights from machine learning are revolutionizing supply chain management to improve the business processes. One of the most challenging aspects of managing a supply chain is predicting the future demands for production. Existing techniques range from baseline statistical analysis techniques including moving averages to advanced simulation modeling. Machine learning is proving to be very effective at considering factors where existing methods have no way of tracking or quantifying over time. Another aspect is that customer purchasing data will be observed from different locations and set to a central store to process and understand the demand indicated by the purchasing data, and also potentially making recommendation to streamline the business. Other types of businesses also experience different issues that machine learning is helping to address. Another example is travel agencies attempting to arrange itineraries for customers is another. Still another example is businesses trying to route people to particular destinations. Some issues, however, have arisen from this expanding use of machine learning. With the advancement of so-called “big data,” which drives the machine learning, data privacy and security has become a worldwide concern. Leaks of public data can cause great concern to a company and those whose data has been leaked, resulting in negative press in the media and worsening relations with the public. For example, a recent data breach involving Facebook has caused a wide range of protests. At the same time, countries are strengthening the protection of data security and privacy. The General Data Protection