I had a meeting with my collaborator at the CONFIRM, and we agreed that we need a list of things to do. so here it is!
current concept of the project:
wearing a neural network at work as if one is wearing an apprentice that observes and learn from you. Training neural network needs tens of thousands of training iteration. By wearing it every day at work over long period of time, eventually the network will learn your work, and more over you, as a master. And when we apply the inceptionism to this network, so we can observe what it understood as your work, and you as a person.
current plan of the artwork/object:
a garment that is equipped with a lot of bend/stretch sensors to observe the movements of the wearer. They are connected to microcontrollers which act as neurons of the input layer of the ANN. The garment is covered with microcontrollers and optic fibers acting as artificial neurons and connections between them. Produced signal is transferred between the neurons/microcontrollers as light (LED-LDR). The garment can recognize the action of the wearer (forward propagation with trained variables) or learn the action of the wearer (back propagation model with both direction communication network).
– specify: type of data, number of data, what I am classifying (or the goal) >> this will lead me to discuss about the design of the ANN architecture.
– finish 12neuron forward propagation model on textile
next next steps: Forward Propagation model
– decide on what this e-textile ANN classifies. What does it do?
– decide on the sensor specification(number, placements) of the garments >> design the ANN architecture
– design the garment >> develop the pattern >> circuit layout on fabric
– implement forward propagation network on garment
next next next steps: Backward propagation model (learn-able)
– prototype back propagation circuit on textile (LED-LED data exchange)
– choice of the IC, design of the circuit for the more elaborate network
– prototype on how to store the calculated data (EPROM?)
– implementation on garment (pattern, circuit design, construction…)
(and here down is either just a story, or maybe to implement…. )
– inceptionism with the learned data
– visualizing the result of the inceptionism in 3D model?
– applying the result of the inceptionism onto another person (muscle stimulation?)
Questions to be answered:
– how do you classify/recognize things that takes place in duration of time in real-time? (for example, dance movements or speech recognition?)
– how does the inceptionism of neural network work?
I need a consultation on:
– if the current implementation of ANN (architecture, number of layers, calculation of back propagation) is correct.
– if I could use more modern method of ANN on e-textile ANN.
– tip/rule of thumb on how one should design the ANN network (number of layers, number of neurons, what is connected to what).