By Naveen Joshi – Director at Allerin
Works on Data Analytics and Strategies, Process Automation, Connected Infrastructure (IoT)
Federated learning can open new horizons for AI (Artificial Intelligence) by enabling access to a broad set of data from various devices as opposed to a centralized system.
The AI market is currently dominated by a few players, such as Amazon, Google, Microsoft, and Apple. Their business model limits the users in the usage of the data generated and collected by these services. This might lead to the monopolization of the market. This will limit the competition in the market and even slow the progress of the technology. However, a more decentralized market trend is gaining prominence where the AI technology and the data to train AI systems are open to small, new businesses over a plethora of platforms, and this approach is known as federated learning.
Understanding federated learning
AI algorithms primarily require centralizing data on a single machine or a server. For example, when an e-retailer wants to develop an AI model to understand its users’ preferences for products, they run the various iterations of models on the data collected from the website or mobile application the user might have accessed the services from. Such data may include the user’s browsing history, the products shortlisted but not bought, the purchased products, etc. Typically, 10s to even 1000s of data points are collected on every user. Such data, however, is passed on to a centralized server for training the AI algorithm.
With federated learning, the machine learning process and the data is transferred to the edge. It enables mobile phones to share the machine learning model and keep the algorithm training data on the device itself. Federated learning allows for faster testing times of the AI models, providing low latency and low power consumption while also respecting the user’s privacy.
A mobile device incorporated in the federated learning architecture downloads the AI training model on the device itself. It then runs the model on the device locally and improves the AI model by analyzing the data stored on the device. The update to the model is then encrypted and sent to the central server. The model is then compared with previous models to identify and improve the best model suited to the user.
The federated learning model was relatively hard to implement in the past due to the limitations of the computing powers of mobile phones. However, with rapid advancements in mobile AI capabilities and easy availability made possible by tech giants such as Apple and Samsung, federated learning will only gain further prominence.
Federated learning opens up new opportunities for technology enthusiasts and businesses alike. As more and more smartphones with advanced AI capabilities become available to the typical consumer, federated learning will slowly but gradually reduce data centralization. It will allow many small and medium businesses to test out their AI models and make the AI market open for a large number of players. With data analytics and model processing being pushed to the edge, the time required to develop new AI models and data-driven products based on the AI models will reduce drastically. Existing businesses, new entrants, and even consumers will benefit hugely from the federated learning model.
