A nomenclature guide for Automation and Machine Learning for Grid-Power Use Minimization in Sustainable Residential Architecture and its appendices

Table of Contents

Version/Revision History:

Version/Revision Date Published Details
V00, Rev.01 2021-12-26 Initial Draft
V01, Rev.00 2021-12-26 Midterm Submission

Glossary

  • Accuracy – For machine learning, unity minus the error rate. 1
  • Architecture – The functional form of the machine learning model, i.e. the mathematical model being applied without any specific parameters achieved by learning labelled data. 1
  • API – Aplication programming interface.
  • Buoyancy-Driven Cooling – A building energy industry term used to describe convective cooling that occurs in buildings due to temperature gradients through the building and across openings to the exterior environment.
  • CNN – Convolutional neural network; a type of neural network that works particularly well for computer vision tasks. 1
  • Epoch – One complete pass through the input data. 1
  • Fine Tuning – A transfer learning technique where the weights of a pre-trained model are updated by training for additional epochs using a different task to that used for pretraining. 1
  • Fit – The process of updating a model’s parameters such that the predictions using the input data match the target labels. 1
  • Label – The data assigned to points in the training and validation sub-data-sets that will be predicted in the application of the model. 1
  • Loss – A quantitative measure of a model’s performance, which measures the difference between labels and predictions. 1
  • Low-Energy Consumption Homes – Homes designed and built with low energy consumption in mind. Synonymous with “Sustainable Residential Architecture” for this study.
  • Metric – A quantification of the quality of a model’s predictions created for human interpretation using a validation data-set. 1
  • Model – The final combination of an architecture and a specific set of parameters achieved through training. 1
  • Overfitting – Training a model in such a way that it remembers specific features of the training dataset and, for that reason, does not perform well when applied to other data subsets. 1
  • Parameters – Statistical weights used by a model to perform its task, obtained by learning labelled data. 1
  • Predictions – Results obtained by the model using independent variables (without labels). 1
  • Pretrained Model – A model which has already been trained, typically using an expansive data-set on a task similar but not identical to the one it will finally be assigned to. The pre-trained model will be “fine-tuned” on data more relevant to this final task. 1
  • Sustainable Residential Architecture – Residential buildings designed and built with low energy consumption in mind. Synonymous with “Low Energy Consumption Homes” for this study.
  • Train – A synonym for “fit.” 1
  • Training Set – The data used for fitting the model. It does not include the validation data set. 1
  • Transfer Learning – A method of pretraining a model where the model is trained on a training data set, not directly applicable to the data that will be used in its final application. 1
  • Validation Set – A subset of labelled data held out of training and used for measuring how good a model is. 1

References Revisited

  1. Howard and Gugger, 2021: Deep Learning for Coders with fastai & PyTorch  2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18