· Xiaojing Yang · Explainability and Responsible AI · 4 min read
中文Interview Notes: Explaining My Training-Data Attribution Thesis
A compact interview narrative for a training-data attribution thesis: research question, method, contribution, limitations, and PhD extensions.
Core idea
A strong interview answer separates what I completed, what I understand, and what I would extend in a PhD. That honesty makes the research story stronger, not weaker.
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 should I explain the thesis clearly in an interview?
Intuition
Interviewers do not only want the final metric. They want to know whether I can define a research problem, justify methodology, understand limitations, and propose the next scientifically meaningful step.
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
A concise structure is: problem -> gap -> method -> evidence -> limitation -> next step. For this thesis, the problem is understanding which training data groups influence domain MT behaviour. The method is group-level coalition attribution with carefully defined utilities and reliability checks.
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
I would explain that I focus on low-resource/domain MT because data quality and composition matter strongly. Instead of only reporting one best model, I ask which data groups contribute to quality, terminology, or style. This connects explainability, evaluation, and data-centric AI.
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
My thesis investigates training-data attribution for domain machine translation. I define interpretable data groups, evaluate their marginal contributions through coalition-style experiments, and analyse how different groups affect quality, terminology, and style. The main contribution is not only a set of scores, but a framework for asking data-centric questions about multilingual model behaviour. The limitations are scalability, approximation error, and the need for stronger intervention-based validation. In a PhD, I would extend this toward hierarchical attribution, instance-level validation, and attribution for factuality and style in LLMs.
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.