Glossary
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