Explainable AI

Apr 27, 2016

We develop methodology to explain the outcome of black box predictive models. We work with Shapley values, counterfactual explanations and related methods.

Martin Jullum

Senior Research Scientist

XAI (Shapley values, counterfactuals), machine learning, network analysis, model selection, point processes.

Events

XAI Tree

Brief introduction and demo of our interactive tool for choosing XAI method

Nov 21, 2022 NAV, Oslo

Martin Jullum

Global model agnostic* explanation methods

Brief introduction and comparison of four global explanation methods

Apr 8, 2022 Online

Martin Jullum

Global model agnostic* explanation methods

Shapley values for prediction explanation

Overview of the use of Shapley values for prediction explanation

Feb 3, 2022 Online

Martin Jullum

Shapley values for prediction explanation

Efficient and simple prediction explanations with groupShapley -- A practical perspective

Extending feature-wise Shapley value explanations to explanations through groups of features

Nov 30, 2021 Online

Martin Jullum, Annabelle Redelmeier, Kjersti Aas

Efficient and simple prediction explanations with groupShapley -- A practical perspective

Prediction explanation with Shapley values

Overview of the use of Shapley values for prediction explanation and the relevant work within BigInsight.

Oct 29, 2021 Online

Martin Jullum

Prediction explanation with Shapley values

Explaining individual predictions when features are dependent: More accurate approximations to Shapley values

Invited talk (+poster) presenting our JAI-paper, presenting different methods for accounting for dependence in Shapley value based prediction explanations with accompanying simulation studies.

Aug 24, 2021 Online

Martin Jullum, joint with Kjersti Aas and Anders Løland

Explaining individual predictions when features are dependent: More accurate approximations to Shapley values

Efficient Shapley value explanation through covariate groups

Presenting a method for explaining predictive models with Shapley values for groups of covariates

Jun 23, 2021 Online/Tromsø, Norway

Martin Jullum

Efficient Shapley value explanation through covariate groups

groupShapley: Efficient Shapley value explanation through feature groups

Poster/5-min lightning presentations: Extending feature-wise Shapley value explanations to explanations through groups of features.

Jan 27, 2021 Online

Martin Jullum

groupShapley: Efficient Shapley value explanation through feature groups

Proper prediction explanatin with shapr

Crash course in Shapley based methods for model explanation and demo of the R-package shapr

Jun 26, 2020 Online

Martin Jullum

Shapley-baserte modellforklaringer

Enkel introduksjon til modellforklaring og Shapley verdier + praktisk bruk av R-pakken shapr

Jun 26, 2020 Online

Martin Jullum

How to open the black box: Individual prediction explanation

Invited talk giving an introduction to prediction explanation with particular focus on our work on Shapley values.

Mar 29, 2020 NTNU, Trondheim

Martin Jullum

How to open the black box: Individual prediction explanation

Opening the black box: Individual prediction explanation

Invited talk presenting our groups work on dependence-aware prediction explanation with Shapley values.

Nov 14, 2019 Norwegian Institute of Public Health, Oslo

Martin Jullum

Opening the black box: Individual prediction explanation

How to open the black box: Individual prediction explanation

Introducing the topic of prediction explanation and Shapley values + some words on our methods for accounting for the feature dependence

Jun 19, 2019 Sola Strand Hotel, Stavanger

Martin Jullum