The world of artificial intelligence is an interesting place to be a part of these days. Due to the cheap and ubiquitous access to computing power in form of CPUs and GPUs has made it easier for researchers and organizations to harness the power of predictive modelling and analysis.
In order to understand the nature of the problem, I typically ask the following questions:
- What is the nature, size and quality of your data?
- What hardware do you have to run the model?
- What are you trying to find? (if you don’t know what you’re looking for then yer dun goofed!)
But in order to understand when to use a model from the plethora of options we need to know the types of problems we can solve and in which category your problem falls under.
Machine Learning falls under the following 5 categories: Supervised learning, Unsupervised learning, Semi-supervised learning, Reinforcement learning and Recommendation Engine (have I committed a cardinal sin by putting RE in it own category!)
In Supervised learning we have labelled data. If we have categorical data we can use classification algorithms if the data is continuous we are better off with regression. Anomalies fall under anomaly detection.
In Unsupervised Learning our data is non-labelled and we need to make sense of this data by putting them in clusters (clustering) or reduce the dimensions to find the meaningful ones (dimension reduction).
Semi-supervised Learning contains a mix of labelled and unlabeled data.
In Reinforcement Learning we could have both types of data. The model takes an action to maximize the reward function in an environment.
Reinforcement learning can be better explained by reward feedback or the reinforcement signal.
Recommendation Engine is an approach where the algorithm finds patterns in historical data to give accurate and meaningful recommendations. Due to its complexity I like to give Recommendation Engine its own category!
Now that we have grouped Machine Learning is 5 broad-categories we can start talking about the many algorithms we can use. (Continued Part 2)
Source: Deep Learning on Medium