Probabilistic Explanations for Linear Models

Authors: Bernardo Subercaseaux, Marcelo Arenas, Kuldeep S. Meel

AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Theoretical Formal XAI is an emerging field that focuses on providing explanations with mathematical guarantees for the decisions made by machine learning models. [...] We propose the notion of pδ, ϵq SR, a simple relaxation of δ-SRs, and show that this kind of explanations can be computed efficiently over linear models. [...] Theorem 2. Given a linear model L and an input x, we can compute a pδ, εq-min-SR successfully with probability 1 γ in time r O d ε2γ2 ; that is, polynomial in d, 1{ε, and 1{γ.
Researcher Affiliation Collaboration Bernardo Subercaseaux1, Marcelo Arenas2, 3, 4, Kuldeep S. Meel5, 6 1Carnegie Mellon University 2Pontificia Universidad Cat olica de Chile 3IMFD Chile 4Relational AI 5Georgia Institute of Technology 6University of Toronto
Pseudocode Yes Algorithm 1: Linear Monte Carlo Explainer Input: Linear model L, instance x, parameters δ P p0, 1q Parameter: ε P p0, 1q, γ P p0, 1q Output: A value δ P rδ ε, δ εs together with a minimum δ -SR explanation for x. [...] Algorithm 2: Monte Carlo Estimation Input: Linear model L, a partial instance y, an instance x, and a number of samples M Output: An estimate pv of Prz Upyqr Lpzq Lpxqs.
Open Source Code No The paper does not provide an explicit statement about the release of open-source code, a link to a repository, or mention of code in supplementary materials.
Open Datasets No The paper uses illustrative examples with synthetic linear models and instances (e.g., Example 1: 'Consider a binary classifier M defined as Mpxq px1 _ x3q px2 _ x1q px4 _ x3q , and the input instance x p1, 1, 0, 1q.'; Example 2: 'Consider a linear model L of dimension d 1000 with parameters t 1250 and w p1000, 1, 1, 1, 1, . . . , 1q.'). It does not refer to any publicly available or open datasets.
Dataset Splits No The paper does not use any external datasets, therefore, there is no mention of dataset splits for training, validation, or testing.
Hardware Specification No The paper discusses the theoretical runtime complexity of its algorithms but does not provide any specific details about the hardware (e.g., CPU, GPU models, memory) used for conducting experiments.
Software Dependencies No The paper does not specify any particular software, libraries, or their version numbers that were used for implementation or experiments.
Experiment Setup No As a theoretical paper focusing on mathematical proofs and algorithm design, it does not describe an experimental setup including specific hyperparameter values, training configurations, or system-level settings.