Lecture excerpt contrasting model-centric and data-centric approaches to AI development. The speaker explains that model-centric development holds data fixed and iteratively improves models (common in academic benchmark-driven research), whereas data-centric development prioritizes improving data quality (via error analysis and data augmentation) and holds code fixed. The speaker recommends considering a data-centric approach and previews a following segment on data augmentation.
Chapters
Lets say that error analysis has caused you to decide to focus on improving your learning algorithms performance on data with a certain category or tag, say speech with car noise in the background. Let's take a look at how you can take a data centric approach to improving your learning algorithms performance. You've heard me speak before about model centric versus data centric AI development. Here's a little more detail on what I mean. With a model centric view of AI development, you would take the data you have and then try to work really hard to develop a model that does as well as possible on the data. Because a lot of academic research in AI was driven by researchers downloading a benchmark data set and trying to do well on that benchmark. Most academic research on AI is model centric because the benchmark data set is a fixed quantity. So in this view, model centric development, you would hold the data fixed and iteratively improve. So in this model centric view, you would hold the data fixed and iteratively improve the code or the model. There's still an important role to play in trying to come up with better models, but there's a different view of AI development, which I think is more useful for many applications, which is to shift a bit from a model centric toward a data centric view. In this view, we think of the quality of the data as paramount, and you can use tools such as error analysis or data augmentation to systematically improve the data quality. And for many applications, I find that if your data is good enough, there are multiple models that will do just fine. So in this view, you can instead hold the code fix and iteratively improve the data. There's a role for model centric development and there's a role for data centric development. If you've been used to model centric thinking for most of your experience with machine learning, I would urge you to consider taking a data centric view as well. Where when you're trying to improve your learning algorithms performance, try asking, how can you make your data set even better? One of the most important ways to improve the quality of a data set is data augmentation. So let's go on to the next video where we'll start to take a look at data augmentation.
