Lecture 5 - Models & Modelling II

Physical Models/Laboratory Models

Recreations of a real-world system on a much smaller scale with limited interactions/processes and certain assumptions that may not exactly reflect the real-world. They tend to use proxies that act like, look like, or behave like the real world components of a system to approximate them.

• This is usually done because it is impossible or hard to use the actual component due to cost, time, ethical reasons, and so on.

Data-Driven Models

Fully based on observations (not theory or processes). They are empirical models that are only based on the data and doesn’t consider the processes of the system. It uses the outputs of systems to determine Split further into statistical models and machine learning models.

There is a divide between people using this model and other models as others argue that we need to understand the components and interactions between them to understand the system. Nontheless, all models require data and parameterisation

Data-Driven Models: Statistical

Things such like line plots, scatter plots, and so on and so forth, requires some sort of computation in order to perform regressions, tests. Mostly done to test hypotheses using only data and observations. However, these models do require some understand on the system’s processes and interactions first (relationships) to come up with the model.

Data-Driven Models: Machine Learning

Models that also rely on only observations and data, but are processed and analysed by A.I. models instead rather than humans. They only need the data and are able to understand and predict based on those data by developing generalised predictors. They can skip the process of understanding complicated processes.

• The way A.I. models process the data makes it hard for us to understand the processes and steps they take to come up with the output or predictions. So, while the machine understands the system better, we don’t really get the chance to understand how that works.

• It also means that it relies on having access to large amounts of good quality data.

However, it is becoming more promising as we feed it more and more data for it to model the world. They can also be used in tandem with other models to support communication, or to just support and make processes easier.

They work by selecting data and then finding the mathematical issues to those situations, and answering it with the data. There are many approaches and technniques in machine learning, which has now been incorporated into many parts of life (academia, professional, research, etc).