SDS 580: Collecting Valuable Data — with Jon Krohn

Labeled Data

Jon Krohn emphasizes the critical role of labeled data in developing valuable machine learning models. He explains that labeled data, unlike unlabeled data, can be used to train models that predict commercially useful outcomes. However, acquiring labeled data is often challenging, requiring innovative strategies to automate the labeling process.

Generally speaking, labeled data are going to be much more valuable commercially than unlabeled data.

--- Jon Krohn

Jon suggests that while automated labeling can be complex, it can lead to unique datasets that set a product apart from competitors 1 2.

   

Automation

Automating data collection and labeling is a key focus for Jon Krohn. He provides practical examples, such as using IMDb star ratings to infer sentiment labels for movie reviews. This method illustrates how automation can streamline the data preparation process, making it more efficient and less costly.

You can sometimes automate the addition of labels to your data.

--- Jon Krohn

Despite the challenges, Jon notes that automation can significantly enhance the value of a dataset, providing a competitive edge in machine learning applications 1.