The Mechanical Neural Network

What is the Mechanical Neural Network?

The Mechanical Neural Network is a physical model of an artificial neural network. Artificial neural networks are the basis of almost all current AI technologies, such as AI chatbots and AI models for image generation. The Mechanical Neural Network is used to explain the basics and functioning of artificial neural networks to students, pupils, and anyone else who is interested.

What advantages does the MNN offer?

The Mechanical Neural Network is a purely mechanical model, which means no computer and therefore no programming knowledge is required to understand AI. This distinguishes it from all other teaching methods that convey AI. However, it is possible to transfer data to the computer and feed the computer-calculated results back into the Mechanical Neural Network.

Furthermore, the Mechanical Neural Network can be used to cover AI topics in varying depth. It is possible to explain the basics of training a classification task in just a few minutes, as well as to motivate and understand complex mathematical methods such as the backpropagation algorithm with the Mechanical Neural Network.

The Mechanical Neural Network uses the Experiential Learning Model (ELM). Learners go through the cycle of active experimentation, concrete experience, observation and reflection, as well as abstract concept formation, and achieve greater learning success compared to traditional teaching methods.

How can the MNN be used in teaching and learning?

A modular system is available for using the Mechanical Neural Network in teaching and learning. From this system, various teaching modules can be selected and adapted in depth to the age, learning objectives, time frame, etc. Ideally, small groups experiment with one Mechanical Neural Network each. Each small group goes through David Kolb’s Experiential Learning Model (ELM) once or several times per module by experimenting with the MNN and gathering concrete experiences, which they can then reflect on in the whole group. Together with the teacher, they then form abstract concepts.

What topics can be taught with the MNN?

The Mechanical Neural Network can be used to teach basic machine learning concepts, such as classification and supervised learning, the structure of artificial neural networks, i.e., the network architecture with its layers, and how artificial neural networks works, such as bias, weighting, and activation function. Iterative learning or training using forward and backward passes can also be taught. The details are described on the page of the modular system.

Which artificial neural networks and which functions are modeled by the mechanical neural network?

The Mechanical Neural Network models a multilayer perceptron or fully connected network with two input neurons, four neurons in a hidden layer, and two output neurons. The ReLU function is used as the activation function .

The functioning of the neurons, i.e., the application of the activation function to the weighted input values and the bias, is modeled as well as the the complete forward pass. The backpropagation algorithm can be motivated with the model and tested manually. Results of a learning step that have been calculated manually or with a computer can be set and traced on the mechanical neural network.

How exactly does the MNN work mechanically?

The Mechanical Neural Network is constructed from rockers that represent the neurons. These rockers are connected by ropes that correspond to the connections between the neurons of an MLP. Tabs that are moved on the rockers model the weights between the neurons. The weights are adjusted by moving the tabs. The summation of the input values is performed by pulleys, and the activation function is modeled by metal rods on the rockers.