Adaptive Data Analysis and Error-Control via Information Measures
Event details
| Date | 21.08.2018 |
| Hour | 14:00 › 16:00 |
| Speaker | Amedeo Esposito |
| Location | |
| Category | Conferences - Seminars |
EDIC candidacy exam
Exam president: Prof. Olivier Lévêque
Thesis advisor: Prof. Michael Gastpar
Co-examiner: Prof. Volkan Cevher
Abstract
There is an increasing concern that most current published research findings are false. The main cause seems to lie in the fundamental disconnection between theory and practise in data analysis. While the former typically relies on statistical independence, the latter is an inherently adaptive process: new hypotheses are formulated based on the outcomes of previous analyses. A recent line of work tries to mitigate these issues using mechanisms, like Differential Privacy, that compose adaptively while degrading gracefully and thus provide statistical guarantees even in adaptive contexts. Our contribution consists in the introduction of a new approach, based on the concept of Maximal Leakage, an information-theoretic measure of leakage of information. We will see how this measure ensures generalization guarantees, composes adaptively and is robust under post-processing, making it a good candidate for being used in this framework.
Background papers
An operational measure of Information Leakage, by Issa, I., et al.
Generalization in Adaptive Data Analysis and Holdout Reuse by Cynthia Dwork et al.
Max-Information, Differential Privacy and Post-selection Hypothesis Testing, Rogers, R., et al.
Exam president: Prof. Olivier Lévêque
Thesis advisor: Prof. Michael Gastpar
Co-examiner: Prof. Volkan Cevher
Abstract
There is an increasing concern that most current published research findings are false. The main cause seems to lie in the fundamental disconnection between theory and practise in data analysis. While the former typically relies on statistical independence, the latter is an inherently adaptive process: new hypotheses are formulated based on the outcomes of previous analyses. A recent line of work tries to mitigate these issues using mechanisms, like Differential Privacy, that compose adaptively while degrading gracefully and thus provide statistical guarantees even in adaptive contexts. Our contribution consists in the introduction of a new approach, based on the concept of Maximal Leakage, an information-theoretic measure of leakage of information. We will see how this measure ensures generalization guarantees, composes adaptively and is robust under post-processing, making it a good candidate for being used in this framework.
Background papers
An operational measure of Information Leakage, by Issa, I., et al.
Generalization in Adaptive Data Analysis and Holdout Reuse by Cynthia Dwork et al.
Max-Information, Differential Privacy and Post-selection Hypothesis Testing, Rogers, R., et al.
Practical information
- General public
- Free
Contact
- EDIC - [email protected]