Memorization
Memorization is the retention of specific information or examples in a form that enables their later recall or reproduction.
The AI Pocket Book
A behavior where a model stores frequent patterns or solutions from training data rather than learning a general rule.
- How it is used
- Uses memorization to explain both brittle problem-solving and verbatim reproduction of training data.
- How deep it goes
- Defines memorization through examples and links it to overfitting, but stops short of explaining model-level mitigation methods.
- Examples
- Uses near-verbatim GPT-4 text, cow recognition, boat-crossing variants, and traffic-cone placement to expose memorized shortcuts.
- What you can do after
- Diagnose memorization by testing uncommon variants and checking whether outputs reproduce training examples nearly verbatim.
Introduction to Generative AI, Second Edition
Model memorization is the retention of specific training examples or sensitive information in a trained model's parameters or behavior.
- How it is used
- Frames memorization as a data privacy and safety risk in publicly deployed language models.
- How deep it goes
- Explains retention, extraction, rare-example memorization, and model-size effects, but does not detail prevention methods.
- Examples
- Uses GPT-2 extraction of contact details, conversations, code, and identifiers as a concrete example.
- What you can do after
- Evaluate privacy and legal risks arising when language models reproduce sensitive training data.
Read how Introduction to Generative AI, Second Edition explains memorization →
Evolutionary Deep Learning: Genetic algorithms and neural networks
Model memorization is the phenomenon in which a model fits particular training examples without learning patterns that transfer to new data.
- How it is used
- Diagnoses when oversized networks specialize on limited training data instead of generalizing to validation or test data.
- How deep it goes
- Explains causes, training signals, and interventions through a CNN experiment, but stops short of formal theory.
- Examples
- Uses a reduced-dataset CNN whose training accuracy nears 100% while validation performance declines, then tests pooling and batch normalization.
- What you can do after
- Diagnose memorization by comparing training and validation curves, then apply dropout, pooling, or batch normalization.
Read how Evolutionary Deep Learning: Genetic algorithms and neural networks explains memorization →
Sutskever's List
Memorization is a model's retention of specific training examples or their details rather than learning patterns that generalize.
- How it is used
- Frames memorization as a challenge in interpreting model training and benchmark performance.
- How deep it goes
- Explains factors shaping retention and gives random-label results, without detailing a mechanism for how examples are stored.
- Examples
- Uses randomized-label AlexNet experiments and possible SQuAD and GLUE overlap with pretraining data.
- What you can do after
- Evaluate benchmark scores cautiously by considering training-data overlap and distinguishing low training loss from learned structure.
Related concepts
- generalizationcontrasts with · Generalization is the extension of a pattern, model, solution, or conclusion beyond the specific examples or cases from which it was derived.
- training datasetproduced by · A collection of examples or observations supplied to a learning procedure to fit or adapt a model or learn its patterns or parameters.
- dropoutaddressed by · Dropout is a neural-network regularization technique that randomly suppresses units, connections, or activations during training to reduce overfitting.
- fine-tuningproduced by · Fine-tuning is adapting a pretrained machine-learning model by training it further on targeted data or objectives for a task, domain, or behavior.
- trainingcontrasts with · Training is a process of developing or improving a person’s or model’s capabilities through instruction, experience, or adjustment using data.
These guides are generated from the text of Manning books by Manning's knowledge graph and link to the sections they draw on. The full books are available on liveBook and at manning.com.