Technical Note—A Data-Driven Approach to Beating SAA Out of Sample
Bibliographic record
Abstract
A Little Pessimism Goes a Long Way Data-driven optimization is concerned with finding a decision, using data and perhaps a model, that performs well when it is applied on a new unseen data point. Data-driven optimization is challenging because data are limited or the model is wrong or the environment in which the decision is being applied is different from the one in which the training data were collected. Distributionally robust optimization (DRO), a worst case optimization method for finding decisions that are insensitive to model error, can sometimes but not always deliver a decision that has a larger out-of-sample expected reward than the sample average approximation (SAA). “A Data Driven Approach to Beating SAA out of Sample” by Jun-ya Gotoh, Michael Kim, and Andrew Lim shows that if worst case (DRO) solutions fail at this task, then the solution of a best case distributionally optimistic optimization problem will do the job. As good as this sounds, there is a catch: whereas an optimistic decision might beat SAA, the improvement is very modest and comes at the cost of being much more sensitive to model misspecification than both the SAA and the DRO decisions. Moreover, it is easy to make a mistake: it can be difficult to determine with a modestly sized data set whether the best or worst case solution will have the higher expected reward than SAA. In summary, data driven optimization is a trade-off between maximizing the expected reward and controlling the sensitivity of this expectation to model misspecification. When both are considered, a little bit of pessimism goes a long way.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.008 | 0.018 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.001 | 0.004 |
| Science and technology studies | 0.001 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.002 | 0.001 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".