Build an Efficient Self-Organizing Neural Network
Neuton’s neural network learning is not based on backward propagation of errors and the stochastic
gradient descent algorithm, but rather uses a new efficient global optimization algorithm with an
excellent generalization capability, allowing for development of the optimal network structure
for one iteration.
Evaluate the Uniqueness
of our Approach by Comparing our Benchmarks
For example, Neuton’s Model for Combined Cycle Power Plant Data Set is 208.6 times
smaller
than TensorFlow, and 42.3 times smaller than TensorFlow Lite.
Algorithm
Size, kB
Coefficients
Target metric
Holdout metric
TensorFlow
31.08
2 338
MAE
3.35
TensorFlow Lite
6.31
n/a
MAE
3.36
Neuton
0.149
100
MAE
3.30
Combined Cycle Power Plant Data Set
The dataset contains 9,568 points collected from a Combined Cycle Power Plant over 6 years. Features
consist of hourly average ambient variables (Temperature, Ambient Pressure, Relative Humidity and
Exhaust Vacuum) used to predict the net hourly electrical energy output.
Create Tiny Models without Сompression
Neuton models maintain all of their original characteristics, without any reduction of
accuracy. Neuton does not reduce the model size after its creation.
Neuton does not use quantization, pruning, clustering, nor distillation.
Embed into Edge Devices
Neuton’s models can be built into microcontrollers and into other small compute devices, with limits as
challenging as the following characteristics:
Energy - 10s-100s mAh
Processor < 100 MHz
Memory < 100 Kb
Explore the Case Study
Determining Cardiac Arrhythmias Case Study
We would like to leverage the created model to predict whether there is an impending arrhythmia or not
by running inferences on the microcontroller, without sending the corresponding sensor data the cloud.
Improve Quality of your Models Ensembling them with Neuton Models
Thanks to unique model creation approach, models generated by the Neuton Neuron Network Framework
enhance any model in the ensemble
Benchmarks
To benchmark a model correctly, and allow for a clear comparison against other solutions, Neuton has
three measurements: number of coefficients, model size, and Kaggle score.