%0 Journal Article %1 zrnic2019natural %A Zrnic, Tijana %A Hardt, Moritz %D 2019 %K bounds generalization machine-learning mathematics probability stats theory %T Natural Analysts in Adaptive Data Analysis %U http://arxiv.org/abs/1901.11143 %X Adaptive data analysis is frequently criticized for its pessimistic generalization guarantees. The source of these pessimistic bounds is a model that permits arbitrary, possibly adversarial analysts that optimally use information to bias results. While being a central issue in the field, still lacking are notions of natural analysts that allow for more optimistic bounds faithful to the reality that typical analysts aren't adversarial. In this work, we propose notions of natural analysts that smoothly interpolate between the optimal non-adaptive bounds and the best-known adaptive generalization bounds. To accomplish this, we model the analyst's knowledge as evolving according to the rules of an unknown dynamical system that takes in revealed information and outputs new statistical queries to the data. This allows us to restrict the analyst through different natural control-theoretic notions. One such notion corresponds to a recency bias, formalizing an inability to arbitrarily use distant information. Another complementary notion formalizes an anchoring bias, a tendency to weight initial information more strongly. Both notions come with quantitative parameters that smoothly interpolate between the non-adaptive case and the fully adaptive case, allowing for a rich spectrum of intermediate analysts that are neither non-adaptive nor adversarial. Natural not only from a cognitive perspective, we show that our notions also capture standard optimization methods, like gradient descent in various settings. This gives a new interpretation to the fact that gradient descent tends to overfit much less than its adaptive nature might suggest.

@article{zrnic2019natural, abstract = {Adaptive data analysis is frequently criticized for its pessimistic generalization guarantees. The source of these pessimistic bounds is a model that permits arbitrary, possibly adversarial analysts that optimally use information to bias results. While being a central issue in the field, still lacking are notions of natural analysts that allow for more optimistic bounds faithful to the reality that typical analysts aren't adversarial. In this work, we propose notions of natural analysts that smoothly interpolate between the optimal non-adaptive bounds and the best-known adaptive generalization bounds. To accomplish this, we model the analyst's knowledge as evolving according to the rules of an unknown dynamical system that takes in revealed information and outputs new statistical queries to the data. This allows us to restrict the analyst through different natural control-theoretic notions. One such notion corresponds to a recency bias, formalizing an inability to arbitrarily use distant information. Another complementary notion formalizes an anchoring bias, a tendency to weight initial information more strongly. Both notions come with quantitative parameters that smoothly interpolate between the non-adaptive case and the fully adaptive case, allowing for a rich spectrum of intermediate analysts that are neither non-adaptive nor adversarial. Natural not only from a cognitive perspective, we show that our notions also capture standard optimization methods, like gradient descent in various settings. This gives a new interpretation to the fact that gradient descent tends to overfit much less than its adaptive nature might suggest.}, added-at = {2019-05-31T16:21:09.000+0200}, author = {Zrnic, Tijana and Hardt, Moritz}, biburl = {https://www.bibsonomy.org/bibtex/237ea2c19c7bc19a0c351772d8cca01d4/kirk86}, description = {[1901.11143] Natural Analysts in Adaptive Data Analysis}, interhash = {2f903bf5bb8cf1d1fd0ce406991645c2}, intrahash = {37ea2c19c7bc19a0c351772d8cca01d4}, keywords = {bounds generalization machine-learning mathematics probability stats theory}, note = {cite arxiv:1901.11143Comment: 22 pages}, timestamp = {2019-05-31T16:21:09.000+0200}, title = {Natural Analysts in Adaptive Data Analysis}, url = {http://arxiv.org/abs/1901.11143}, year = 2019 }