Wednesday 19 March 2025
Scientists have long struggled to develop artificial neural networks that can learn new tasks without forgetting old ones. The problem, known as catastrophic forgetting, has stumped researchers for years, making it difficult to create intelligent machines that can adapt to changing environments. Now, a team of scientists has made significant progress in addressing this issue by developing a novel approach inspired by the fly’s olfactory system.
The key to their success lies in the way they’ve structured their neural network. Unlike traditional networks, which are designed with a fixed number of layers and connections, the new approach uses a modular architecture that can learn and adapt as needed. This allows the network to selectively forget or update old memories while still retaining new information.
But how does it work? The team’s design is based on the fly’s olfactory system, which is capable of detecting and distinguishing between thousands of different odors. In this system, a group of neurons called projection neurons (PNs) receive input from sensory receptors and send it to another group of neurons called Kenyon cells (KCs). These KCs process the information and then send it to the mushroom body, which is responsible for learning and memory.
The team’s neural network mimics this architecture by dividing it into three layers: a projection neuron layer, a Kenyon cell layer, and an output layer. The PNs receive input from the environment and send it to the KCs, where it is processed and sent to the output layer. This modular design allows the network to learn and adapt in real-time, without forgetting old information.
The team tested their approach on several benchmark tasks, including image recognition and language processing. In each case, their neural network outperformed traditional networks in terms of accuracy and ability to generalize to new data. The results suggest that this novel approach could be a game-changer for artificial intelligence, enabling machines to learn and adapt to changing environments with greater ease.
One of the most impressive aspects of the team’s design is its efficiency. Unlike traditional neural networks, which require massive amounts of computational power and memory, the fly-inspired network is relatively lightweight and can run on even modest hardware. This makes it potentially deployable in a wide range of applications, from robotics to autonomous vehicles.
Of course, there are still many challenges to overcome before this technology becomes widely available. For example, the team’s design is currently limited to relatively simple tasks, such as image recognition.
Cite this article: “Artificial Intelligence Breakthrough: Neural Network Inspired by Flys Olfactory System”, The Science Archive, 2025.
Artificial Intelligence, Neural Networks, Catastrophic Forgetting, Modular Architecture, Fly’S Olfactory System, Projection Neurons, Kenyon Cells, Mushroom Body, Image Recognition, Language Processing.
