A self-learning magnetic Hopfield neural network with intrinsic gradient descent adaption.

Area of Science:

• Materials Science
• Neuroscience
• Computer Science

Background:

• Physical neural networks (PNN) offer energy-efficient artificial intelligence but face training challenges.
• Current PNN training relies heavily on external computing resources.
• Physical self-learning, using intrinsic material properties for training, is an emerging solution.

Purpose of the Study:

• To demonstrate a spintronic system capable of physical self-learning for Hopfield neural networks (HNN).
• To show intrinsic adaptation of learning rules via autonomous physical processes.
• To eliminate the need for external computation in training PNN.

Main Methods:

• Implemented a spintronic system mimicking HNN using magnetic texture-defined conductance matrices as trainable weights.
• Applied external voltage inputs to drive the evolution of the conductance matrix.
• Demonstrated unsupervised learning through the natural evolution of physical parameters.

Main Results:

• The conductance matrix evolved and adapted Oja's learning algorithm in a gradient descent manner.
• The self-learning HNN demonstrated scalability.
• The system successfully achieved associative memories on patterns with high similarities.

Conclusions:

• A real spintronic system for physical self-learning HNN was successfully demonstrated.
• Intrinsic adaptation of learning rules via material property evolution eliminates external computation needs.
• Spintronic platforms offer a promising avenue for efficient, autonomous material-based training.