BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Task Alignment - Vihari Piratla\, University of Cambridge
DTSTART:20240626T100000Z
DTEND:20240626T113000Z
UID:TALK218335@talks.cam.ac.uk
CONTACT:120952
DESCRIPTION:Imagine the implications of submitting an essay or a code rout
 ine generated by an LLM without double-checking. Deploying a task-specific
  model without verifying or modifying is even more consequential. As with 
 any other software\, ML models cannot be perfect and need constant monitor
 ing and patching. Yet\, the problem of making targeted bug fixes to ML mod
 els received little attention and love. We will discuss representative pap
 ers of the four broad solutions and their limitations. We will conclude wi
 th a critical evaluation of the current progress and future directions. Th
 e attached image is a non-exhaustive summary of related work. The papers t
 hat we will cover (likely) are further down below.\n\n**References**\n\nPa
 rameter editing approaches.\nShibani Santurkar\, Dimitris Tsipras\, Mahala
 xmi Elango\, David Bau\, Antonio Torralba\, and Aleksander Madry. Editing 
 a classifier by rewriting its prediction rules. Advances in Neural Informa
 tion Processing Systems\, 34:23359–23373\, 2021.\nTransparent model appr
 oaches.\nPang Wei Koh\, Thao Nguyen\, Yew Siang Tang\, Stephen Mussmann\, 
 Emma Pierson\, Been Kim\, and Percy Liang. Concept bottleneck models. In I
 nternational conference on machine learning\, pages 5338–5348. PMLR\, 20
 20.\nBhavana Dalvi Mishra\, Oyvind Tafjord\, and Peter Clark. Towards teac
 hable reasoning systems: Using a dynamic memory of user feedback for conti
 nual system improvement. arXiv preprint arXiv:2204.13074\, 2022.\nDense da
 ta annotation approaches\nRoss\, Andrew Slavin\, Michael C. Hughes\, and F
 inale Doshi-Velez. "Right for the right reasons: Training differentiable m
 odels by constraining their explanations." arXiv preprint arXiv:1703.03717
  (2017).\nSukrut Rao\, Moritz Böhle\, Amin Parchami-Araghi\, and Bernt S
 chiele. Studying how to efficiently and effectively guide models with expl
 anations. In Proceedings of the IEEE/CVF International Conference on Compu
 ter Vision\, pages 1922–1933\, 2023.\nData augmentation approaches\nShio
 ri Sagawa\, Pang Wei Koh\, Tatsunori B Hashimoto\, and Percy Liang. Distri
 butionally robust neural networks for group shifts: On the importance of r
 egularization for worst-case generalization. arXiv preprint arXiv:1911.087
 31\, 2019.
LOCATION:Cambridge University Engineering Department\, CBL Seminar room BE
 4-38.
END:VEVENT
END:VCALENDAR