Leveraging Unlabeled Image Data With Self-Supervised Learning or Pseudo Labeling With Mateusz Opala

On this episode of MLOps Live we have Mateusz Opala as our guest. Mateusz shares his experience and answers your questions about leveraging unlabeled image data with SSL or pseudo labeling.

MLOps Live is a biweekly Q&A show where practitioners doing ML at a reasonable scale answer questions from other ML practitioners. Every episode focused on one specific subject related to MLOps. Only the juicy bits, the things you won’t find in a company blog post.

00:00 Introduction to MLops Live
00:35 Mateusz introduction
02:08 How would you explain pseudo labeling in one minute?
04:30 What are some different use cases at Brainly where you apply self-supervised models and pseudo labeling for your image data?
09:33 Is your experience at Brainly your first in working on semi-supervised learning problems? If not, what other experiences can you share with us?
11:47 How does Mateusz choose image augmentation to train the SSL model?
12:51 What are the challenges you have encountered so far with using supervised learning techniques and pseudo labeling for your image tasks at Brainly?
16:22 What are the most likely errors a small team would make when applying pseudo-labeling, and how do you think they should be addressed?
18:54 Did you try other semi-supervised learning techniques? If you did, how did they stack up against pseudo-labeling?
22:55 Pseudo-labeling is known to enhance the robustness of the model. How does it achieve this in your use case?
25:54 Are there other issues that affect the efficacy of pseudo labeling?
29:35 Labeled data is expensive and difficult to get, while unlabeled data is abundant and cheap. How does this impact pseudo-labeling?
34:08 How do you set up the architecture for the data processing and training of the models?
38:11 What is still actively being researched in self-supervised learning and pseudo learning that you cannot take to production yet but you would like to see if production-ready?
40:00 What is the self-supervised learning technique?
41:07 What would you say is your biggest challenge with MLOps right now?
44:07 Where can you find Mateusz?

Follow us & stay updated:
► MLOps Community: (#neptune-ai)