For our next little series introducing a different thing in science and how it works every week, I decided to focus on classifiers. With artificial intelligence becoming more and more prominent in our daily lives as of late, I thought this would be a good lead into the explicitly science-focused topics to come. So, what is a classifier? How does it work? And why does it matter?
At their core, classifiers are algorithms designed to categorize input data into predefined classes or categories. They learn patterns and relationships from labelled training data to make predictions on new, unseen data.
Once features are extracted, identified and quantified from labelled or annotated input data, mathematical models are employed for pattern recognition and predictions.
These models can range from simple decision trees to complex neural networks, each with its own strengths and weaknesses.
Training these models is an iterative process. That means to produce one good classifier, lots of classifiers were created in the process: Every time the pattern recognition is run, the annotated data is categorised by the classifier and compared to the annotation class. Prediction errors are corrected, and performance is optimised. This whole process is one iteration. How many iterations are required for a well-trained classifier varies widely and is largely dependent on the input data and application. For my tissue classifiers, it took up to 20,000 iterations.
Classifiers use these models to categorise unseen data into categories the user-defined at the start. In the figure, you can see my annotated histological slides from which the classifier extracted patterns to then classify the rest of the slide and entirely unseen slides into tumour (red), stroma (green) and background (blue) classes.
From identifying fraudulent transactions, filtering out junk mail, targeted advertising, and facial recognition to unlock your phone or diagnosing diseases, classifiers play a vital role in automating decision-making processes and driving advancements across a wide range of industries. Keep your eyes peeled, and you can find more classifiers in action all around you.
Written by Ronja Struck