AIAI Seminar - 29 November 2021 - Talks by Giannis Papantonis, Adarsh Prabhakaran and Thomas Fletcher Talk by Giannis Papantonis Title: Principled Diverse Counterfactuals in Multilinear Models Abstract: Machine learning applications have automated numerous real-life tasks, improving both private and public life. However, the black-box nature of many state-of-the-art models poses the challenge of model verification; how can one be sure that the algorithm bases its decisions on the proper criteria, or that it does not discriminate against certain minority groups? In this paper we propose a way to generate diverse counterfactual explanations from multilinear models, a broad class which includes Random Forests, as well as Bayesian Networks. Talk by Adarsh Prabhakaran Title: Studying the spread of smoking Abstract: Agent based models have been used to study a wide range of phenomenon from financial markets to the spread of infections. In this talk, I will be introducing an agent-based model developed to study the spread of smoking in a population. Empirical studies have shown that close social ties influence both smoking initiation and cessation. We incorporate these into the model through two types of non-smoker smoker interactions in a network. We show that the network topology influences the spread of smoking and that it can be used for tobacco control strategies. Talk by Thomas Fletcher Title: An overview of SMART (Statistical Methodology Advisor at Reasoning Time) – Architecture & Methods Abstract: SMART (Statistical Methodology Advisor at Reasoning Time) is a general-purpose statistics expert system (to be released as a Python package with an R back-end) developed in order to expand the capabilities of the FRANK query-answering system. Its purpose is to select (and carry out) appropriate processing, modelling and analysis steps based on the features of a given query and an accompanying dataset. Its behaviour is governed by a simple computational construct, akin to a Turing machine over a graph (implemented in the Python package Graph-State-Machine), which performs reasoning steps based on an ontology of features of inputs, outputs and all steps between them; this ontology constitutes the full “expertise” of the system and has the benefit of being an easily interpretable visual structure. Due to the abstracted nature of its logic and the modularity of its design, SMART is easily expandable, and already contains a broad selection of statistical methods (chosen to cover a range of new query types for FRANK), including multiple varieties of numerical and categorical generalised linear modelling, adaptive Gaussian process covariance discovery, analysis of variance and time series analysis. Nov 29 2021 14.00 - 15.00 AIAI Seminar - 29 November 2021 - Talks by Giannis Papantonis, Adarsh Prabhakaran and Thomas Fletcher AIAI Seminar talks hosted by Giannis Papantonis, Adarsh Prabhakaran and Thomas Fletcher Online
AIAI Seminar - 29 November 2021 - Talks by Giannis Papantonis, Adarsh Prabhakaran and Thomas Fletcher Talk by Giannis Papantonis Title: Principled Diverse Counterfactuals in Multilinear Models Abstract: Machine learning applications have automated numerous real-life tasks, improving both private and public life. However, the black-box nature of many state-of-the-art models poses the challenge of model verification; how can one be sure that the algorithm bases its decisions on the proper criteria, or that it does not discriminate against certain minority groups? In this paper we propose a way to generate diverse counterfactual explanations from multilinear models, a broad class which includes Random Forests, as well as Bayesian Networks. Talk by Adarsh Prabhakaran Title: Studying the spread of smoking Abstract: Agent based models have been used to study a wide range of phenomenon from financial markets to the spread of infections. In this talk, I will be introducing an agent-based model developed to study the spread of smoking in a population. Empirical studies have shown that close social ties influence both smoking initiation and cessation. We incorporate these into the model through two types of non-smoker smoker interactions in a network. We show that the network topology influences the spread of smoking and that it can be used for tobacco control strategies. Talk by Thomas Fletcher Title: An overview of SMART (Statistical Methodology Advisor at Reasoning Time) – Architecture & Methods Abstract: SMART (Statistical Methodology Advisor at Reasoning Time) is a general-purpose statistics expert system (to be released as a Python package with an R back-end) developed in order to expand the capabilities of the FRANK query-answering system. Its purpose is to select (and carry out) appropriate processing, modelling and analysis steps based on the features of a given query and an accompanying dataset. Its behaviour is governed by a simple computational construct, akin to a Turing machine over a graph (implemented in the Python package Graph-State-Machine), which performs reasoning steps based on an ontology of features of inputs, outputs and all steps between them; this ontology constitutes the full “expertise” of the system and has the benefit of being an easily interpretable visual structure. Due to the abstracted nature of its logic and the modularity of its design, SMART is easily expandable, and already contains a broad selection of statistical methods (chosen to cover a range of new query types for FRANK), including multiple varieties of numerical and categorical generalised linear modelling, adaptive Gaussian process covariance discovery, analysis of variance and time series analysis. Nov 29 2021 14.00 - 15.00 AIAI Seminar - 29 November 2021 - Talks by Giannis Papantonis, Adarsh Prabhakaran and Thomas Fletcher AIAI Seminar talks hosted by Giannis Papantonis, Adarsh Prabhakaran and Thomas Fletcher Online
Nov 29 2021 14.00 - 15.00 AIAI Seminar - 29 November 2021 - Talks by Giannis Papantonis, Adarsh Prabhakaran and Thomas Fletcher AIAI Seminar talks hosted by Giannis Papantonis, Adarsh Prabhakaran and Thomas Fletcher
