Pith. sign in

REVIEW 2 cited by

Better Training Data Attribution via Better Inverse Hessian-Vector Products

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 2507.14740 v1 pith:DHPN2GYZ submitted 2025-07-19 cs.LG stat.ML

Better Training Data Attribution via Better Inverse Hessian-Vector Products

classification cs.LG stat.ML
keywords astradataihvpiterationstrainingaccurateapproximationattribution
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Training data attribution (TDA) provides insights into which training data is responsible for a learned model behavior. Gradient-based TDA methods such as influence functions and unrolled differentiation both involve a computation that resembles an inverse Hessian-vector product (iHVP), which is difficult to approximate efficiently. We introduce an algorithm (ASTRA) which uses the EKFAC-preconditioner on Neumann series iterations to arrive at an accurate iHVP approximation for TDA. ASTRA is easy to tune, requires fewer iterations than Neumann series iterations, and is more accurate than EKFAC-based approximations. Using ASTRA, we show that improving the accuracy of the iHVP approximation can significantly improve TDA performance.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 2 Pith papers

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

  1. On the Fragility of Data Attribution When Learning Is Distributed

    cs.LG 2026-05 unverdicted novelty 6.0

    A single adversary in distributed training inflates its attribution value via latent optimization on synthetic batches without degrading accuracy or triggering basic defenses.

  2. LLM generation novelty through the lens of semantic similarity

    cs.LG 2025-10 conditional novelty 6.0

    An output is called novel when no semantically similar pretraining context can be retrieved; using this embedding-based measure, SmolLM models show longer-range reuse and instruction tuning increases novelty.