The Truth About Your Lying Calibrated Forecaster: How to Design Truthful Calibration Measures

Channel:
United States Simons Institute for the Theory of Computing
Subscribers:
68,700
Published on ● Video Link: https://www.youtube.com/watch?v=MTHI2onHYBc


Duration: 0:00
313 views
0

Nika Haghtalab (UC Berkeley)
https://simons.berkeley.edu/talks/nika-haghtalab-uc-berkeley-2024-11-18
Joint IFML/MPG Symposium



Other Videos By Simons Institute for the Theory of Computing


2024-12-04Weak-to-Strong Generalization
2024-12-04Fireside Chat
2024-11-19Robust Mixture Learning when Outliers Overwhelm Small Groups
2024-11-19Learning General Gaussian Mixtures With Efficient Score Matching
2024-11-19Talk by Mahdi Soltanolkotabi (University of Southern California)
2024-11-18Some Easy Optimization Problems Have the Overlap-Gap Property
2024-11-18Understanding Contrastive Learning and Self-training
2024-11-18Revisiting Scalarization in Multi-Task Learning
2024-11-18Beyond Decoding: Meta-Generation Algorithms for Large Language Models (Remote Talk)
2024-11-18Omnipredicting Single-Index Models with Multi-Index Models
2024-11-17The Truth About Your Lying Calibrated Forecaster: How to Design Truthful Calibration Measures
2024-11-14Scalably Understanding AI with AI
2024-11-14Domain Adaptation-- 20 years of theory chasing practice
2024-11-14Adaptive Data Collection via Autoregressive Generation
2024-11-14Panel Discussion
2024-11-14Open-Source and Science in the Era of Foundation Models
2024-11-13Toward Understanding the Extrapolation of Nonlinear Models to Unseen Domains or the Whole Domain
2024-11-13Language-guided Adaptation
2024-11-13On Spurious Associations and LLM Alignment
2024-11-13Causally motivated robustness to shortcut learning
2024-11-13Talk by Zachary Lipton