We like to try out new ideas at Sunfleck. On this page you’ll find our latest experiments. Consider it our sandbox, our playground. At the moment, we’re having fun with machine learning (ML), an important subset of artificial intelligence. We’re looking at ways to incorporate these fascinating techniques into our games and apps.
In machine learning, the computer learns how to do a task. It is not given explicit instructions, but rather learns on it’s own. There are several fundamental types of machine learning:
- Supervised learning – this is a machine learning task where an algorithm trains the computer based on input-output pairs. The iris example in our app is based on this idea. The computer is trained with data for three different iris species (outputs) where the flower dimensions (inputs) are known for each. Then, when the computer is given a new set of flower dimensions that it has never seen before, the correct species can be predicted.
- Unsupervised learning – the computer is trained on data where there is no obvious relationship among inputs and there are no preconceived outputs. After training, the algorithm can recognize patterns in the data to create groupings. For example, the computer might sort potatoes based on inputs such as colour or size. It can then predict which group a new set of potatoes might belong to.
- Reinforcement learning – in this type of ML, agents learn in an interactive environment by train and error. They are given rewards for doing well, hence the term reinforcement. This is perhaps the most sophisticated type of ML, and is often compared to the way we as humans learn.
There are countless algorithms for each of these categories. Examples include genetic algorithms, neural networks, logistic regression, decision trees, and support vector machines. Whew!
Full disclosure: some of these projects we built from scratch, some use well-established ML libraries, and some are pre-built, third-party projects that we’ve tweaked.
Go ahead and play with some of these five interactive projects, and in the process learn a bit about how the magic is done. And check back once in a while for more experiments.