Using Neural Networks for Optimization
How we can have a trained Model to cluster according to what we need
Panelizing has been more considered by covering the main free form but, there are other factors we need to look for in order to have a design gets optimized. This idea got us to check out for something different as an algorithm understands requirements and response as well. That was the place we started study about Machine Learning. The project used for this scenario was a special free form and the question was: How we can cluster panels after optimization in a dataset by their specifications from Vertices, Edges Lengths and Areas. Right after this process we were wondered to use different type of Machine Learning Algorithms in order for managing a great response. The clustering followed by this research was all about the delivery of fabrication in panels similarities all by the parameters mentioned above. Following the scenario above, we used LunchBox-ML for clustering by Gaussian Mixture Model. Right after taking advantage of visual programing benefits for getting to know something faster, we started to define a training model in TensorFlow. There are a lot to talk about TensorFlow but the first facing to it for us equals to check all our devices compatibilities for calculations and running the training. Turns out we had to test bunch of versions for taking the best result. In that case we found ourselves on installing and uninstalling different versions of CUDDnn and CUDA-Toolkit. After a series of trying the TensorFlow and TensorBoard finally ran great and the result got us to bring new ideas in the study. The next part was about training the model by all the parameters was used in Gaussian Mixture Model which is undergoing.
This project has been done as a PhD thesis part for Ms. Sara Beyraghi. We are honored by cooperating on the idea and developing it through the process.
