Blog Posts

Modern machine learning models require a lot of training data to be successful. Over the past few years, we've been studying programatically creating labels with weak supervision to address this training data bottleneck; instead of relying on manual labels, data programming uses weak supervision---multiple noisy label sources---to automatically generate labeled... [Read More]

Four years ago, the idea that you could use a statistical process to help people label training data from noisy sources sounded absolutely crazy. But since then, the ideas that we formalized into data programming and weak supervision in Snorkel have had a surprisingly broad degree of impact. In fact,... [Read More]

Data augmentation techniques have proven powerful in building machine learning models in applications such as image and text classification. However, most of the machine learning research today is still carried out solving fixed tasks. In the real world, machine learning models in deployment can fail in various ways, due to... [Read More]

As we have seen in the previous blog post, data augmentation techniques have achieved remarkable gains when applied to neural network models. In this blog post, we reflect on the success story of various augmentation techniques and review our recent work that study theoretical properties of data augmentation. [Read More]

Instead of performing manual search, automated data augmentation approaches hold promise to search for more powerful parameterizations and compositions of transformations. Perhaps the biggest difficulty with automating data augmentation is how to search over the space of transformations. This can be prohibitively expensive due to the large number of transformation... [Read More]