Aims & Scope The spectacular
successes in machine learning (ML) have led to a plethora of
Artificial Intelligence (AI) applications. However, the large
majority of these successful models, like deep neural networks,
support vector machines, etc. are black boxes, opaque, non-intuitive
and difficult for people to understand. There are critical domains
that demand more intelligent, autonomous, and symbiotic systems,
like medicine, security, legal, the military, finance and
transportation, to mention a few, for which performance is not the
only quality indicator. These are areas where decision-making faces
high risks due to the involvement of human lives, critical
infrastructure, very costly operations, national threats, etc. In
situations like these, decision makers need much more that numeric
performance in favor of alternative solutions that provide rationale
and are more knowledge-based.
The goal of Explainable AI (XAI) is to create a suite of ML
techniques that i) result in more explainable models, while
maintaining a high level of learning performance, but also ii)
enable human users to develop understanding to be able to trust the
model, and effectively manage a new generation of artificially
intelligent machine tools. Continued advances promise to produce
autonomous systems that will perceive, learn, decide, and act on
their own. However, the effectiveness of these systems is limited by
the machines' current inability to explain their decisions and
actions to human users.
This session will explore the performance-versus-explanation
trade-off space. This will include ML models that are interpretable
by design. Some, like fuzzy systems and rule induction, have general
function approximation properties. Very important are also
algorithms producing models in mathematic languages such as
algebraic functions and differential equations, piecewise non-linear
models, etc. Despite of the differences in the approaches, there are
common elements and basic methodologies that are present in many
applications. We will bring together researchers from
different fields to discuss key issues related to the research and
applications of XAI methods and to share their experiences of
solving common problems.
Topics that are of interest to this session include but are not
limited to:-
• Interpretable ML Models
• Query Interfaces for Deep Learning
• Interactive User Interfaces
• Active and Transfer learning
• Relevance and Metric Learning
• Practical Applications of Interpretable Machine
Learning
• Deep Neural Reasoning
Preliminary Dates
Paper submission: EXTENDED DEADLINE January 30, 2020
Paper acceptance notification: March 15, 2020
Session Chairs Julio J.
Valdés (Julio.Valdes@nrc-cnrc.gc.ca), National Research Council Canada Paulo
Lisboa, Sandra
Ortega-Martorell and Ivan
Olier ({P.J.Lisboa, S.OrtegaMartorell, I.A.OlierCaparroso}@ljmu.ac.uk),
Liverpool John Moores University, U.K. Alfredo Vellido
(avellido@cs.upc.edu), Universitat Politècnica de Catalunya, Spain