bootstrap
In plain wordsEstimating how much a result might vary by repeatedly re-drawing samples, with replacement, from the data you already have — like re-running a study many times without collecting new data.
- 1
Statistics
A resampling method that treats the observed sample as a stand-in for the population: it draws many new samples of size n from the n data points with replacement, recomputes the statistic of interest on each, and reads the statistic's standard error, bias or confidence intervalconfidence interval, n.: A range of plausible values for something being estimated, such as 'accuracy 82 percent, give or take 2 points', reflecting the uncertainty that comes from measuring only a sample. off the spread of the recomputed values. Computation thus takes the place of analytic formulas, so the method serves estimators from a median to a model's AUCAUC, abbr.: A single number from 0 to 1 summarizing how well a model ranks real positive cases above negative ones: 0.5 is no better than a coin flip and 1.0 is perfect. or benchmarkbenchmark, n.: A fixed, shared test — a set of questions or tasks with an agreed way of scoring — used to compare AI systems with one another and to track progress over time. score, though it can fail for extremes, very small samples or dependent data, which call for block or other variants. In machine learning, bootstrap samples underlie baggingbagging, n.: Training many copies of a model, each on a different random resample of the data, and averaging their predictions. The averaging cancels out much of each copy's random error. and random forestsrandom forest, n.: A prediction method that builds hundreds of decision trees, each from a random sample of the data and features, and lets them vote. The crowd is far more reliable than any single tree., and bootstrapped intervals are a standard way to report uncertainty in evaluation results.
We bootstrapped the test set 10,000 times; the 95% interval for the accuracy gain was 0.4 to 2.1 points.
- 2
Reinforcement learning
Updating an estimate of value from other current estimates rather than from complete observed returns, as in temporal difference learningtemporal difference learning, n.: A way of learning predictions by adjusting each guess toward the next, slightly better-informed guess instead of waiting for the final outcome — like revising a weather forecast every day as the date approaches. and Q-learningQ-learning, n.: A trial-and-error method in which an agent keeps a score for every move in every situation and nudges each score toward the reward it got plus the best score available next.. Bootstrapping lowers variance and allows learning before an episodeepisode, n.: One complete run of a task from start to finish — a single game of chess, one attempt at a maze — after which everything resets and the agent tries again. ends, at the cost of bias inherited from the estimates' own errors.
- 3
General use
Starting a system from a small seed and improving it with its own outputs — for example, training a model on a few labeled examples, labeling unlabeled data with it and retraining on the most confident labels (self-training), or a compiler compiling itself. All senses share the image of pulling oneself up by one's bootstraps.
Origin
The statistical bootstrap was introduced and named by Bradley Efron in his 1977 Rietz Lecture, published as 'Bootstrap Methods: Another Look at the Jackknife' (Annals of Statistics, 1979); the name alludes to pulling oneself up by one's bootstraps. The reinforcement-learning sense was popularized by Richard Sutton and Andrew Barto's textbook (1998).
Related
- baggingbagging, n.: Training many copies of a model, each on a different random resample of the data, and averaging their predictions. The averaging cancels out much of each copy's random error.
- random forestrandom forest, n.: A prediction method that builds hundreds of decision trees, each from a random sample of the data and features, and lets them vote. The crowd is far more reliable than any single tree.
- confidence intervalconfidence interval, n.: A range of plausible values for something being estimated, such as 'accuracy 82 percent, give or take 2 points', reflecting the uncertainty that comes from measuring only a sample.
- cross-validationcross-validation, n.: A fair way to test a model when data are limited: split the data into several parts, train on all but one, test on the part left out, and rotate so every part takes a turn.
- temporal difference learningtemporal difference learning, n.: A way of learning predictions by adjusting each guess toward the next, slightly better-informed guess instead of waiting for the final outcome — like revising a weather forecast every day as the date approaches.
- semi-supervised learningsemi-supervised learning, n.: Machine learning that combines a small set of labeled examples with a much larger pile of unlabeled ones, which are cheaper to collect, to learn better than from the labeled data alone.
Sources
- Efron, B. (1979). Bootstrap Methods: Another Look at the Jackknife. The Annals of Statistics, 7(1), 1–26. projecteuclid.org/euclid.aos/1176344552
Mentioned in
baggingbagging, n.: Training many copies of a model, each on a different random resample of the data, and averaging their predictions. The averaging cancels out much of each copy's random error., cross-validationcross-validation, n.: A fair way to test a model when data are limited: split the data into several parts, train on all but one, test on the part left out, and rotate so every part takes a turn., Leo BreimanLeo Breiman, prop. n.: An American statistician who invented random forests and bagging, decision-tree methods still in wide use, and urged statisticians to judge models by how well they predict., random forestrandom forest, n.: A prediction method that builds hundreds of decision trees, each from a random sample of the data and features, and lets them vote. The crowd is far more reliable than any single tree., conformal predictionconformal prediction, n.: A way to wrap any prediction model so that, instead of one guess, it gives a range or set of answers that contains the truth a chosen share of the time, such as 90%.