Feeding the world
Agriculture is a combination of science and skill. It refers to cultivating seeds, tending livestock, using pesticides, irrigation and harvesting plants, etc. Rather than dispersed ideas, it is a holistic integration of all these applications. Plants and animals both serve the purpose of food and material production. Food is the prime component of all human necessities. As per the Food and Agriculture Organization, by 2050, the global population is expected to increase by 20% to 9.8 billion people. To feed this increasing population, the agricultural industry will need to produce 60% more food while only using 5% more land. Continuing with traditional farming techniques could lead to a whopping 68% increase in cost.
The integration of AI can effectively increase productivity while keeping the cost at bay. Smart Agriculture combines the latest technologies and evidence backed by scientific breakthroughs to optimise crops and livestock production.
AI in agriculture
Artificial intelligence can be used to filter quality seeds, study environmental conditions, estimate crop yield, predict cattle health and disease, and identify infestation or outbreaks. An AI system makes decisions based on what it learns from past mistakes and data trends.
In Machine learning (ML), a centralised model learns to perform by training on a single dataset. So, each record, data distribution, and raw data is accessible to the model.
Contrary to this setup, federated learning (FL) provides a way to distribute the learning process of AI systems across multiple devices, making it suitable for resource-constrained devices. A commercial farm from Wales, a farmer from the Scottish Highlands, and a backyard gardener in England can track their plant growth via images. Everyone gets their own ML model to deduce leaf diseases from images and then redirects the learned parameters to the central server. The server is then in charge of detecting the disease every time a new image comes from either of these three regions. In this process, all parties benefit without the need for compromising data security and increasing computational power, as large datasets and sophisticated tasks are handled by the central server.
Privacy preserving AI
In practice, a model often struggles to learn across complex and non-linear relationships. In fact, real data is heterogeneous, multi-modal, and multi-regional. Accessibility to data is also limited due to usage policies, and privacy restrictions. Federated-learning-based-agriculture models leverage environmental and sensor data and can learn from multi-domain datasets.
Most farms and farmers might be sceptical about sharing data with others. The decentralised setup allows multiple farms to collaborate with each other irrespective of their location, device type, or computation strength and helps them fulfil their individual aims. Thus, cross-city, cross-country, and even cross-continent global models can be deployed. Privacy and security are also upheld in the process while preventing unnecessary competition as local farm devices only share the updated weights and parameters rather than raw data with the global aggregator model.
Why federated learning (FL)?
As agriculture leaps into an unforeseen future, federated learning stands as the stewardship to balancing the productivity versus privacy issue by offering proper collaborative learning while keeping hold of local ownership of data. By offloading high intensity computation to the central server, it allows farms to operate on low powered, cost-effective settings, further driving productivity. Signing up to a decentralised system will benefit small-scale farmers as well as commercial farms irrespective of their location and resources, and thus pave the path towards a fair, successful, and sustainable agriculture system.
Rate and Review
Rate this article
Review this article
Log into OpenLearn to leave reviews and join in the conversation.
Article reviews