Explainer Wrappers¶

quanda ships wrappers around several existing TDA libraries, exposing them through a single Explainer interface. The tables below list every wrapper class and cite the paper that introduced the underlying method. All wrapper sources live under quanda/explainers/wrappers/.

All wrappers can be imported directly from quanda.explainers.wrappers, for example:

from quanda.explainers.wrappers import (
    CaptumSimilarity,
    TRAK,
    Kronfluence,
    RepresenterPoints,
    DattriIFExplicit,
)

Captum¶

Wrappers around the influence methods provided by Captum. Source: quanda/explainers/wrappers/captum_influence.py.

Wrapper	Reference
`CaptumSimilarity`	Similarity between test and training samples in the representation space of a chosen layer. See Captum’s SimilarityInfluence docs.
`CaptumArnoldi`	Schioppa et al., 2022. Scaling Up Influence Functions. arXiv:2112.03052
`CaptumTracInCP`	Pruthi et al., 2020. Estimating Training Data Influence by Tracing Gradient Descent. NeurIPS 2020
`CaptumTracInCPFast`	Pruthi et al., 2020. Estimating Training Data Influence by Tracing Gradient Descent. NeurIPS 2020
`CaptumTracInCPFastRandProj`	Pruthi et al., 2020. Estimating Training Data Influence by Tracing Gradient Descent. NeurIPS 2020

Representer Point Selection¶

Source: quanda/explainers/wrappers/representer_points.py.

Wrapper	Reference
`RepresenterPoints`	Yeh et al., 2018. Representer Point Selection for Explaining Deep Neural Networks. NeurIPS 2018 — original implementation: chihkuanyeh/Representer_Point_Selection

TRAK¶

Source: quanda/explainers/wrappers/trak_wrapper.py.

Wrapper	Reference
`TRAK`	Park et al., 2023. TRAK: Attributing Model Behavior at Scale. ICML 2023 — original implementation: MadryLab/trak

Kronfluence¶

Source: quanda/explainers/wrappers/kronfluence.py.

Wrapper	Reference
`Kronfluence (incl. EK-FAC)`	Grosse et al., 2023. Studying Large Language Model Generalization with Influence Functions. arXiv:2308.03296 — original implementation: pomonam/kronfluence

Dattri¶

Wrappers around the unified TDA family provided by Dattri (Deng et al., 2024, arXiv:2410.04555). Source: quanda/explainers/wrappers/dattri_influence.py.

Wrapper	Reference
`DattriIFExplicit`	Koh and Liang, 2017. Understanding Black-box Predictions via Influence Functions. ICML 2017
`DattriIFCG`	Koh and Liang, 2017 (conjugate-gradient solver). ICML 2017
`DattriIFLiSSA`	Agarwal et al., 2017. Second-Order Stochastic Optimization for Machine Learning in Linear Time. JMLR 2017
`DattriIFDataInf`	Kwon et al., 2024. DataInf: Efficiently Estimating Data Influence in LoRA-tuned LLMs and Diffusion Models. ICLR 2024
`DattriArnoldi`	Schioppa et al., 2022. Scaling Up Influence Functions. arXiv:2112.03052
`DattriEKFAC`	Grosse et al., 2023. Studying Large Language Model Generalization with Influence Functions. arXiv:2308.03296
`DattriTracInCP`	Pruthi et al., 2020. Estimating Training Data Influence by Tracing Gradient Descent. NeurIPS 2020
`DattriGradDot`	Charpiat et al., 2019. Input Similarity from the Neural Network Perspective. NeurIPS 2019
`DattriGradCos`	Charpiat et al., 2019. Input Similarity from the Neural Network Perspective. NeurIPS 2019
`DattriTRAK`	Park et al., 2023. TRAK: Attributing Model Behavior at Scale. ICML 2023