Intel / article

Machine Learning with Dynamics

Source: unconv.ai
Author: Matthew Storm Bull
Date: Published May 21, 2026

Context

A simulated collection of gyroscopes and springs can be trained to recognize handwritten digits from their stroke movements. In Unconventional AI's worked example, the strokes become forces applied to selected gyroscopes. Springs transmit their influence through the network, and a learned readout converts the resulting positions and velocities into a digit prediction.

Training adjusts simulated physical properties, including masses and spring strengths, alongside the input and output transformations. The reported run reached 83.4% validation accuracy. Without connections between the input and output gyroscopes, the task could not be learned. The example illustrates how trainable interactions make dynamics computationally useful; it reports a simulation, not a tested mechanical classifier.

Key takeaways · AI-distilled

The first gyroscope-on-rod-only design fails as a classifier: each gyroscope has no way to influence the others, so input information never reaches the output gyroscopes, and the system cannot be trained until springs connect them.
Connecting 100 gyroscopes in a 10x10 hexagonal grid with 261 springs, the trained system reached 83.4% accuracy on the PenDigits stroke-classification task, versus 56.2% for a similarly complex linear classifier and 89.6% for an LSTM.
Training uses ordinary backpropagation through a differentiable ODE model of the gyroscope-and-spring dynamics, the same machinery used for Neural ODEs, so gyroscope mass, rod length, and spring/gyroscopic coefficients are learned parameters.
After training on GPU, can run on the physical gyroscope-and-spring system itself, or be emulated on a different physical system such as an electrical circuit, where 1 second of gyroscope time can map to 1 microsecond of circuit time.
The authors note the emulated dynamics don't need to be physically realistic: a hardware surrogate could emulate 'negative mass' to boost the system's expressivity even though that value is not physically meaningful.

Terms in this piece · Glossary

inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.

Read the source unconv.ai