· Xiaojing Yang · Explainability and Responsible AI · 3 min read
中文Scaling Data Attribution: Monte Carlo Shapley, Surrogate Models, and Hierarchical Attribution
Why exact Shapley is expensive and how scalable attribution uses sampling, surrogate models, datamodels, and group-to-document-to-example hierarchies.
Core idea
Exact Shapley scales exponentially with the number of units. Scalable attribution therefore needs approximation, structure, or both.
This note is part of my series Training Data Attribution for NLP and LLM Research. The series is written as both a research notebook and an interview preparation path: each article should help me explain the idea clearly, connect it to my thesis, and identify what would become future PhD work.
Guiding question: How can data attribution scale beyond a small number of data groups?

Intuition
With five data groups, coalition experiments may be feasible. With five thousand documents, they are impossible if treated naively. Scaling means deciding where to spend expensive retraining budget and where to use approximations.
In NLP and LLM research, this matters because model behaviour is deeply shaped by data mixture. A model may be fluent because of broad web text, domain-accurate because of specialised documents, safer because of curated instruction data, or biased because of repeated patterns in a subset of the corpus. Training-data attribution gives us language for asking these questions systematically instead of only saying “the data matters”.
Formal lens
Monte Carlo Shapley samples permutations or coalitions instead of enumerating all subsets. Surrogate models or datamodels learn to predict utility from data inclusion vectors. Hierarchical attribution first scores coarse groups, then drills down into high-impact groups at document or example level.
The important discipline is to define the attribution setup before interpreting the score:
| Design choice | Question to answer |
|---|---|
| Attribution unit | What receives credit: source, group, document, example, or token? |
| Utility function | Which behaviour is being explained: quality, terminology, style, factuality, or safety? |
| Intervention | Are we adding, deleting, reweighting, correcting, or retraining? |
| Estimator | Is the score exact, sampled, gradient-based, surrogate-based, or heuristic? |
| Uncertainty | How stable is the score across seeds, samples, metrics, and evaluation sets? |
NLP / LLM example
For domain MT, I might first attribute across source groups. If the petroleum regulatory group is high-impact, I can then split it into document families, then inspect sentence pairs with terminology errors. This preserves interpretability while controlling computation.
This is why I do not want to treat attribution as a generic interpretability topic. For my profile, the natural connection is multilingual and domain-specific NLP: low-resource settings, technical terminology, written-standard variation, and evaluation beyond one headline metric.
Connection to my thesis
In my thesis narrative, training-data attribution is useful because it turns a vague data question into an experimental design:
- define interpretable data units;
- define the model behaviour to explain;
- compare controlled data coalitions or interventions;
- estimate contribution;
- report uncertainty and limitations;
- decide what evidence is strong enough to support a causal-style claim.
That structure helps me avoid overclaiming. A score is not automatically a causal explanation. It is a measurement produced by a specific setup.
What I have done, understand, and would extend
| Level | Status |
|---|---|
| Already completed / thesis-ready | Group-level attribution, coalition thinking, metric-based utilities, cautious interpretation, random baselines, bootstrap-style reliability checks. |
| I understand but may not fully implement yet | Instance-level gradient attribution, influence functions, TracIn, Monte Carlo Shapley, surrogate/datamodel approximations. |
| Strong PhD extension | Hierarchical attribution, intervention-based validation, factuality/style-specific utilities, scalable attribution for LLM data mixtures. |
Interview answer
I would explain exact Shapley as conceptually clean but computationally limited. A realistic PhD direction is hierarchical attribution: start with interpretable groups, use Monte Carlo or surrogate estimates where needed, and only perform expensive retraining for the most important hypotheses.
References and reading path
- Lloyd Shapley, A Value for n-Person Games.
- Ghorbani and Zou, Data Shapley: Equitable Valuation of Data for Machine Learning.
- Koh and Liang, Understanding Black-box Predictions via Influence Functions.
- Pruthi et al., Estimating Training Data Influence by Tracing Gradient Descent.
- Ilyas et al., Datamodels: Predicting Predictions from Training Data.
- Rei et al., COMET: A Neural Framework for MT Evaluation.