overfitting
In plain wordsWhen a model learns its training examples too well — quirks and noise included — so it does great on data it has seen but poorly on anything new, like a student who memorizes answers without understanding.
The failure in which a model, while learning from its training examples, also absorbs their accidents — noise, coincidences and quirks of that one sample — and treats them as signal, so that it scores much better on the data it was trained on than on new data. It shows up as a widening generalization gapgeneralization gap, n.: The difference between how well a model does on the examples it trained on and how well it does on new ones. A big gap means it memorized rather than learned., often with validation loss rising while training loss keeps falling. Overfitting is more likely when capacity is high relative to the amount of data, when training runs long, or when many models or hyperparametershyperparameter, n.: A setting chosen by people before training, such as how fast a model learns or how big it is, as opposed to the internal values the model learns by itself from data. are tried against the same validation data — which overfits the validation set itself, and, on public leaderboards, the 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.. Remedies include more or augmented data, regularizationregularization, n.: Any technique that stops a model from memorizing its training examples too closely — often by penalizing complexity — so that it works better on new data., early stoppingearly stopping, n.: Stopping training at the point where the model does best on practice data it has not learned from, before it starts memorizing quirks of its training examples., dropoutdropout, n.: A training trick that randomly switches off a fraction of a network's neurons at each step, so the network cannot lean too heavily on any one of them and learns sturdier patterns., simpler models, pruningpruning, n.: Cutting away the parts of an AI model that contribute least, as a gardener trims branches, so the model becomes smaller or faster while still working nearly as well., ensemblingensemble learning, n.: Combining the predictions of several models — by voting or averaging — so their individual mistakes partly cancel out, giving more accurate and stable results than any one model. and honest evaluation with 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. and untouched test setstest set, n.: A set of examples kept completely apart from training and used at the end to measure how well a model works on data it has never seen, like a sealed final exam.. Some overparameterized models fit noisy training data exactly yet still generalize well (benign overfittingbenign overfitting, n.: When a model fits its training data perfectly — even the errors and noise — yet still predicts well on new data, defying the textbook warning that this always backfires.), complicating the classical picture.
Training accuracy hit 99% but validation stalled at 71% — classic overfitting.
Origin
The danger is old in statistics: with enough free parameters one can fit almost anything, as in the anecdote, told by Freeman Dyson, of Enrico Fermi quoting John von Neumann about fitting an elephant with four parameters. 'Overfitting' was used in statistics before machine learning and appears in neural-network literature by the late 1980s; its first use has not been established. Peter Bartlett and colleagues named 'benign overfitting' (preprint 2019; published 2020).
Usage note
Training on leaked test items (data contaminationdata contamination, n.: When the test questions used to grade an AI system were already in its training data, so a high score may reflect memorized answers rather than real ability.) or tuning against a test set inflates reported scores in a related way. Not the same as memorizationmemorization, n.: When an AI model stores specific pieces of its training data — a paragraph, a phone number, an image — closely enough to reproduce them, instead of only learning general patterns., though memorizing noise is one form of overfitting.
Related
- underfittingunderfitting, n.: When a model is too simple, or trained too little, to capture the real pattern in the data, so it does poorly even on the examples it learned from.
- bias–variance tradeoffbias–variance tradeoff, n.: The balancing act between models too simple to capture a pattern and models so flexible that they also learn the noise. The best predictions usually come from somewhere in between.
- regularizationregularization, n.: Any technique that stops a model from memorizing its training examples too closely — often by penalizing complexity — so that it works better on new data.
- generalization gapgeneralization gap, n.: The difference between how well a model does on the examples it trained on and how well it does on new ones. A big gap means it memorized rather than learned.
- early stoppingearly stopping, n.: Stopping training at the point where the model does best on practice data it has not learned from, before it starts memorizing quirks of its training examples.
- 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.
- double descentdouble descent, n.: A surprising pattern where a model's error on new data first falls, then rises, then falls again as the model grows bigger — contradicting the old rule that oversized models always do worse.
- data leakagedata leakage, n.: A mistake in building a model that lets it peek at clues it will not have in real use, such as hints about the very answer it must predict, so it shines in testing and then disappoints.
- underfittingunderfitting, n.: When a model is too simple, or trained too little, to capture the real pattern in the data, so it does poorly even on the examples it learned from.
- generalizationgeneralization, n.: How well a model copes with new examples it has never seen, rather than just memorizing its training data, like a student who understands the method, not just the answers.
Sources
- Hawkins, D. M. (2004). The Problem of Overfitting. Journal of Chemical Information and Computer Sciences, 44(1), 1–12.
- Bartlett, P. L., Long, P. M., Lugosi, G. & Tsigler, A. (2020). Benign Overfitting in Linear Regression. Proceedings of the National Academy of Sciences, 117(48), 30063–30070.
Mentioned in
fine-tuningfine-tuning, n.: Taking an AI model that has already learned a lot and training it a little more on a smaller, focused set of examples so it gets better at one job or behaves a certain way., machine learningmachine learning, n.: A way of building software that learns from examples instead of following rules written by a programmer. Shown thousands of labeled photos, for instance, a program can learn to recognize cats., trainingtraining, n.: The process by which an AI model learns: it is shown huge numbers of examples and its internal settings are adjusted, bit by bit, until its answers get better., training settraining set, n.: The examples a machine-learning model learns from, like the worked problems a student practices on, kept separate from the examples later used to check how well it learned., bias–variance tradeoffbias–variance tradeoff, n.: The balancing act between models too simple to capture a pattern and models so flexible that they also learn the noise. The best predictions usually come from somewhere in between., coefficient of determinationcoefficient of determination, n.: A score for numerical predictions showing how much of the variation in the true values a model accounts for, compared with always guessing the average: 1 is perfect, 0 is no better than the average., 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., curse of dimensionalitycurse of dimensionality, n.: The problems that pile up when each item is described by very many measurements: even huge datasets end up thinly scattered across the possibilities, and 'near' and 'far' stop meaning much., data augmentationdata augmentation, n.: Creating extra training examples by changing existing ones in ways that keep their meaning — a flipped, cropped or recolored photo of a cat still shows a cat — so the model learns more robustly., data leakagedata leakage, n.: A mistake in building a model that lets it peek at clues it will not have in real use, such as hints about the very answer it must predict, so it shines in testing and then disappoints., decision boundarydecision boundary, n.: The dividing line, or surface, where a classifier switches from predicting one category to another. Simple models draw straight lines; flexible ones draw curves., decision treedecision tree, n.: A model that makes a prediction by asking a series of yes-or-no questions about the input, like a flowchart, with each answer leading to the next question until it reaches a final answer., dropoutdropout, n.: A training trick that randomly switches off a fraction of a network's neurons at each step, so the network cannot lean too heavily on any one of them and learns sturdier patterns., early stoppingearly stopping, n.: Stopping training at the point where the model does best on practice data it has not learned from, before it starts memorizing quirks of its training examples., empirical risk minimizationempirical risk minimization, n.: The basic recipe of most machine learning: from the models on offer, pick the one with the smallest errors on the training examples, trusting that score as a guide to how it will do on new data., epochepoch, n.: One complete pass through all the training examples. Smaller models are often trained for many epochs, seeing each example many times; the largest language models may see most of their data only once., feature selectionfeature selection, n.: Choosing which of the available input measurements a model should use and dropping the rest, so the model becomes simpler, faster and often more accurate., generalizationgeneralization, n.: How well a model copes with new examples it has never seen, rather than just memorizing its training data, like a student who understands the method, not just the answers., generalization gapgeneralization gap, n.: The difference between how well a model does on the examples it trained on and how well it does on new ones. A big gap means it memorized rather than learned., gradient boostinggradient boosting, n.: A method that builds a strong predictor by adding small decision trees one at a time, each trained to correct the errors that remain after all the trees before it., KaggleKaggle, prop. n.: A popular website, owned by Google, where people compete to build the most accurate prediction models on shared datasets, publish their code and data, and learn data science., learning curvelearning curve, n.: A graph showing how a model's performance improves as it gets more training — more examples or more time. Its shape tells you whether more data or a bigger model would help., LightGBMLightGBM, prop. n.: A widely used tool for making predictions from very large tables of data, quickly and in little memory, by combining many small decision trees. Microsoft released it as free software in 2016., loss curveloss curve, n.: A graph of a model's error as training goes on. A healthy curve drops steeply and then keeps sloping gently downward; bumps, jumps or flat stretches can warn engineers that something is wrong.