weak supervision
In plain wordsTraining AI with labels that are cheap but imperfect, such as rough rules of thumb, partial tags or noisy guesses, instead of carefully hand-checked answers.
A family of approaches that train models from supervision that is noisier, coarser or less complete than hand-annotated ground truthground truth, n.: The reference answers treated as correct when checking or training an AI system — for example, the confirmed diagnoses that a medical-imaging model's guesses are compared against., to reduce the cost of labelingdata labeling, n.: The work of attaching correct answers to examples — often done by large paid workforces, sometimes by experts — so that AI systems can learn from them or be tested against them.. Zhi-Hua Zhou's taxonomy distinguishes incomplete supervision (only some examples labeled, overlapping with 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. and active learningactive learning, n.: A way to train a model with fewer labeled examples by letting it choose which unlabeled examples a person should label next, usually the ones it is most unsure about.), inexact supervision (labels coarser than the task requires, as in multiple-instance learning or image-level tags used to train object detectors) and inaccurate supervision (noisy labelslabel noise, n.: Mistakes in the answers attached to training or test examples — a dog photo marked 'cat' — which can mislead what a model learns and distort how good it seems.). Practical sources of weak labels include heuristic rules, keyword patterns, knowledge bases (distant supervision), crowd workers and, increasingly, large language modelslarge language model, n.: A computer program trained on enormous amounts of text to predict the words that come next. At vast scale, that simple skill lets it chat, answer questions, translate, summarize and write code. acting as labelers. In data programming, as in the Snorkel system, users write many labeling functions whose conflicting votes a learned label model combines into probabilistic training labels.
Thirty regex rules and a drug database labeled the corpus, and the weakly supervised model learned to generalize beyond them.
Origin
Weakly supervised methods include Mike Mintz and colleagues' 'distant supervision' for relation extraction (2009). Alexander Ratner, Christopher Ré and colleagues at Stanford formalized data programming (2016) and the Snorkel system (2017), and Zhi-Hua Zhou's 2018 review in National Science Review supplied a standard taxonomy.
Usage note
Usage varies: some authors treat weak supervision as an umbrella that includes semi-supervised learning, others as a synonym for it, and others reserve it for programmatic, noisy labeling. Models trained this way inherit the systematic biases of their labeling heuristics.
Related
- 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.
- label noiselabel noise, n.: Mistakes in the answers attached to training or test examples — a dog photo marked 'cat' — which can mislead what a model learns and distort how good it seems.
- data labelingdata labeling, n.: The work of attaching correct answers to examples — often done by large paid workforces, sometimes by experts — so that AI systems can learn from them or be tested against them.
- active learningactive learning, n.: A way to train a model with fewer labeled examples by letting it choose which unlabeled examples a person should label next, usually the ones it is most unsure about.
- supervised learningsupervised learning, n.: Teaching a computer by example: you give it many inputs paired with the right answers, such as photos labeled 'cat' or 'dog', and it learns to answer correctly for new inputs.
- synthetic datasynthetic data, n.: Data created by computer programs or AI models rather than collected from the real world — such as practice math problems written by a model, or simulated street scenes for self-driving cars.
- ground truthground truth, n.: The reference answers treated as correct when checking or training an AI system — for example, the confirmed diagnoses that a medical-imaging model's guesses are compared against.
- supervised learningsupervised learning, n.: Teaching a computer by example: you give it many inputs paired with the right answers, such as photos labeled 'cat' or 'dog', and it learns to answer correctly for new inputs.
Sources
- Ratner, A. et al. (2017). Snorkel: Rapid Training Data Creation with Weak Supervision. Proceedings of the VLDB Endowment, 11(3).
- Zhou, Z.-H. (2018). A Brief Introduction to Weakly Supervised Learning. National Science Review, 5(1), 44–53.
- Mintz, M., Bills, S., Snow, R. & Jurafsky, D. (2009). Distant Supervision for Relation Extraction without Labeled Data. Proceedings of ACL-IJCNLP 2009.
Mentioned in
labellabel, n.: The correct answer attached to an example in a dataset — 'cat' on a photo of a cat, 'positive' on an upbeat review — that a learning system is trained to predict., active learningactive learning, n.: A way to train a model with fewer labeled examples by letting it choose which unlabeled examples a person should label next, usually the ones it is most unsure about., data labelingdata labeling, n.: The work of attaching correct answers to examples — often done by large paid workforces, sometimes by experts — so that AI systems can learn from them or be tested against them., 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., supervised learningsupervised learning, n.: Teaching a computer by example: you give it many inputs paired with the right answers, such as photos labeled 'cat' or 'dog', and it learns to answer correctly for new inputs., WhisperWhisper, prop. n.: A free OpenAI model that writes down what is said in recordings in many languages, and can translate it into English; it learned from hundreds of thousands of hours of online audio., label noiselabel noise, n.: Mistakes in the answers attached to training or test examples — a dog photo marked 'cat' — which can mislead what a model learns and distort how good it seems., sound event detectionsound event detection, n.: Teaching computers to recognize everyday sounds in a recording, such as a dog barking, glass breaking or a siren, and to say exactly when each one starts and stops., weak-to-strong generalizationweak-to-strong generalization, n.: The finding that a strong AI trained only on answers from a weaker, error-prone teacher can end up outperforming that teacher — studied as a stand-in for humans supervising smarter-than-human AI.