thank you for a great workshop!
TRANSCRIPT
![Page 1: Thank you for a great workshop!](https://reader031.vdocument.in/reader031/viewer/2022032902/623f9c6d3e8c6774d655d3d7/html5/thumbnails/1.jpg)
![Page 2: Thank you for a great workshop!](https://reader031.vdocument.in/reader031/viewer/2022032902/623f9c6d3e8c6774d655d3d7/html5/thumbnails/2.jpg)
Thank you for a great workshop!• 65 submissions!
• 56 reviewers!
• 44 accepted papers!
• 11 speakers!
• Thank you to our co-organizers not at NIPS (Isabelle, Eugene, Christoph, Eduoard, Chris) !
![Page 3: Thank you for a great workshop!](https://reader031.vdocument.in/reader031/viewer/2022032902/623f9c6d3e8c6774d655d3d7/html5/thumbnails/3.jpg)
Semi-Supervised Learning!Weak Supervision!Transfer Learning!Representation Learning!Applications!Multi-Task Learning!Data Augmentation!Active Learning !Self-Training!Knowledge Distillation!
Topics of Accepted Papers
![Page 4: Thank you for a great workshop!](https://reader031.vdocument.in/reader031/viewer/2022032902/623f9c6d3e8c6774d655d3d7/html5/thumbnails/4.jpg)
Highlights from the Keynotes• Multi-task, multi-view---i.e. coupling---is critical!
– Tom Mitchell: “At NIPS 2027, we’ll look back and smile that we were tackling the hardest task in ML—learning single fns. in isolation”!
• Injection of domain expertise via more informative priors:!– Beyond L1 reg., GEC, logical constraints, etc.!
• More creative, higher-level, and responsive weak supervision types:!– GEC, logical constraints, feedback on explanations, AL, etc.!
![Page 5: Thank you for a great workshop!](https://reader031.vdocument.in/reader031/viewer/2022032902/623f9c6d3e8c6774d655d3d7/html5/thumbnails/5.jpg)
Highlights from the Keynotes• Intersection of structured prediction with
weaker supervision!
• Use of adversarial techniques!– For SSL setting!– Using GANs for SSL, data augmentation and generation via simulation,
domain adaptation!
• Panel: Insights from the applied side:!– Selection of appropriate problems for ML!– Representing label/task ambiguity!– Replacing versus assisting!!
![Page 6: Thank you for a great workshop!](https://reader031.vdocument.in/reader031/viewer/2022032902/623f9c6d3e8c6774d655d3d7/html5/thumbnails/6.jpg)
Weak supervision: The New New Alchemy?• One thing we’re excited about: weak
supervision!
• Proposition: Turn noisy, low-quality supervision into gold (labels)!
• Except here, we do have some theory! What can this help us to engineer?!
Excited to chat more about this!
![Page 7: Thank you for a great workshop!](https://reader031.vdocument.in/reader031/viewer/2022032902/623f9c6d3e8c6774d655d3d7/html5/thumbnails/7.jpg)
THANK YOU!