Pith. sign in

REVIEW 16 cited by

TRAK: Attributing Model Behavior at Scale

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2303.14186 v2 pith:LVVOJVQB submitted 2023-03-24 stat.ML cs.LG

classification stat.MLcs.LG
keywords modelstrakattributiondatamethodsmodeltrainingwork
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
read the original abstract

The goal of data attribution is to trace model predictions back to training data. Despite a long line of work towards this goal, existing approaches to data attribution tend to force users to choose between computational tractability and efficacy. That is, computationally tractable methods can struggle with accurately attributing model predictions in non-convex settings (e.g., in the context of deep neural networks), while methods that are effective in such regimes require training thousands of models, which makes them impractical for large models or datasets. In this work, we introduce TRAK (Tracing with the Randomly-projected After Kernel), a data attribution method that is both effective and computationally tractable for large-scale, differentiable models. In particular, by leveraging only a handful of trained models, TRAK can match the performance of attribution methods that require training thousands of models. We demonstrate the utility of TRAK across various modalities and scales: image classifiers trained on ImageNet, vision-language models (CLIP), and language models (BERT and mT5). We provide code for using TRAK (and reproducing our work) at https://github.com/MadryLab/trak .

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 16 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. A Defense of the Quadratic Model

    cs.LG 2026-07 conditional novelty 7.0 of 10

    Local Taylor-expanded quadratic models reproduce a 150M-parameter LLM's validation loss for up to 10% of training late in the run, and LLM pretraining operates within a factor of 2 of a stochastic or deterministic edg...

  2. Small edits, large models: How Wikipedia advocacy shapes LLM values

    cs.CL 2026-04 unverdicted novelty 7.0 of 10

    Wikipedia edits by animal welfare advocates measurably influence LLM outputs on animal welfare topics, shown via retrieval and gradient attribution plus fine-tuning experiments.

  3. Sketching the Readout of Large Language Models for Scalable Data Attribution and Valuation

    cs.LG 2026-04 unverdicted novelty 6.0 of 10

    RISE applies CountSketch to dual lexical and semantic channels derived from output-layer gradient outer products, cutting data attribution storage by up to 112x and enabling retrospective and prospective influence ana...

  4. GUDA: Counterfactual Group-wise Training Data Attribution for Diffusion Models via Unlearning

    cs.LG 2026-01 reject novelty 6.0 of 10

    GUDA approximates leave-one-group-out counterfactual models with unlearning and ranks group influence by ELBO differences.

  5. BLISS: A Lightweight Bilevel Influence Scoring Method for Data Selection in Language Model Pretraining

    cs.LG 2025-10 conditional novelty 6.0 of 10

    A bilevel optimization method ranks pretraining data by training a small proxy model on weighted samples, yielding modest downstream-task gains without external pretrained models.

  6. Understanding Data Influence with Differential Approximation

    cs.LG 2025-08 conditional novelty 6.0 of 10

    This paper introduces Diff-In, an influence estimator that accumulates second-order approximations of influence differences across training steps and shows strong accuracy in data cleaning, deletion, and coreset selec...

  7. Better Training Data Attribution via Better Inverse Hessian-Vector Products

    cs.LG 2025-07 conditional novelty 6.0 of 10

    ASTRA, an EKFAC-preconditioned Neumann series iteration, computes more accurate inverse Hessian-vector products and improves training data attribution scores over EKFAC baselines.

  8. The Common Pile v0.1: An 8TB Dataset of Public Domain and Openly Licensed Text

    cs.CL 2025-06 conditional novelty 6.0 of 10

    A new 8TB openly-licensed text corpus trains 7B LLMs that are competitive with Llama 1/2, showing that performant models need not depend on unlicensed web data.

  9. Daunce: Data Attribution through Uncertainty Estimation

    cs.LG 2025-05 conditional novelty 6.0 of 10

    DAUNCE computes training-data attribution as the covariance of per-example losses across an ensemble of perturbed fine-tuned models, reporting state-of-the-art LDS scores and the first attribution runs on proprietary LLMs.

  10. Laplace Sample Information: Data Informativeness Through a Bayesian Lens

    cs.LG 2025-05 conditional novelty 6.0 of 10

    LSI ranks training samples by informativeness using the KL divergence between Laplace-approximated posteriors with and without each sample, and the ordering transfers from a small probe to larger models.

  11. KAIROS: Scalable Model-Agnostic Data Valuation

    cs.LG 2025-06 conditional novelty 5.0 of 10

    KAIROS derives a closed-form Maximum Mean Discrepancy influence score that approximates leave-one-out data rankings and detects noise, mislabels, and backdoors without retraining.

  12. TokenShapley: Token Level Context Attribution with Shapley Value

    cs.CL 2025-06 conditional novelty 5.0 of 10

    TokenShapley computes token-level Shapley attributions from context to response by treating context tokens as (prefix, token) data points in a KNN datastore.

  13. Rescaled Influence Functions: Accurate Data Attribution in High Dimension

    cs.LG 2025-06 conditional novelty 5.0 of 10

    Rescaled influence functions, which account for the change in the Hessian when a sample is removed, dramatically improve leave-T-out effect estimates compared to standard influence functions in high-dimensional logist...

  14. Efficient Data Selection at Scale via Influence Distillation

    cs.CL 2025-05 conditional novelty 5.0 of 10

    Influence Distillation selects LLM fine-tuning data by approximating each sample's gradient influence on a target task via landmarks and JVP embeddings, matching or beating RDS+ accuracy at roughly one third the selec...

  15. Scaling laws for activation steering with Llama 2 models and refusal mechanisms

    cs.LG 2025-07 conditional novelty 4.0 of 10

    Activation steering with contrastive vectors becomes less effective as Llama 2 models scale from 7B to 70B parameters, with peak effect at roughly 40% of the model's layers.

  16. Attributing Data for Sharpness-Aware Minimization

    cs.LG 2025-07 reject novelty 4.0 of 10

    SAM-HIF and SAM-GIF are proposed as data attribution scores for SAM-trained models, but SAM-GIF is TracIn with SAM gradients and SAM-HIF's derivation contains a load-bearing error.

Pith tools