Memo archive
Memos
-
1. Introducing Data Counterfactuals
Many data-centric techniques in machine learning can be unified under the concept of 'data counterfactuals': how does changing data affect model outcomes.
-
2. Formalisms
A draft memo lining up several tasks that fit the data counterfactuals frame: influence, valuation, active learning, distillation, unlearning, privacy, and poisoning.
-
3. Glossary
Working definitions for the key terms used across the memo, explorer, and related areas and papers page.