Seminarios 2022-2023

Título: Machine learning methods in forensic science

Ponente: Joanna Karłowska-Pik, Ph. D., Nicolaus Copernicus University in Toruń, Poland

Sesión:

• 2 marzo, de 16:30 a 18:00

Lugar: 7.2.J03

Organizador: Raquel Fuentetaja

Resumen:

Constructing predictive models based on data from the human genome can help scientists and specialists in various fields. Such models are primarily used to determine the risk of cancer or other genetic diseases. During the lecture, the use of machine learning methods to build models that can determine the presumed appearance of the perpetrator of the crime will be presented. The specificity of working with genetic data, including the so-called small n large p problem, will also be discussed.

Breve biografía:

Joanna Karłowska-Pik, Ph. D. Adjunct on the Faculty of Mathematics and Computer Science, Nicolaus Copernicus University in Toruń, Poland; employed at the Department of Mathematical Statistics and Data Mining. Since 2021 director of the Centre for Statistical Analysis of NCU. IBM SPSS software coach. Teacher of the winners of the Statistical Olympiad and the European Statistics Competition for secondary education students. Her research interests focus on applications of machine learning methods in forensics, genetics and life sciences.

Título: Dynamic Decision Models and Games

Ponente: Dr.ª Cleotilde González. Directora del Dynamic Decision Making Laboratory, Carnegie Mellon University

Sesiones:

• Viernes 24 de Marzo de 2023, de 16:30 a 19:30 horas
• Lunes 27 de Marzo de 2023, de 10 a 13:00 horas

Lugar: 3.S1.08. Campus de Leganés

Organizadora: Dr.ª María Isabel Sánchez Segura

Resumen:

This course will explore human decision making as a dynamic process resulting from human interactions with the environment. The course uses decision games to illustrate how humans learn and adapt to changing conditions of choice, and computational models to simulate decision processes and environmental dynamics.

Day 1

• Overview of Dynamic Decision Making (DDM).
• Learning and Instance-Based Learning Theory (IBLT)
• IBL models of Decisions from Experience (DfE) and binary choice

Day 2

• Practice on Binary Choice Tasks and models in Shiny IBL
• DfE in 2-person games
• DfE in Groups

Breve biografía:

Dr. Cleotilde Gonzalez is a Research Professor at the Department of Social and Decision Sciences at Carnegie Mellon University. Her research work focuses on the study of human decision making in dynamic and complex environments. She is the founding director of the Dynamic Decision Making Laboratory where researchers conduct behavioral studies on dynamic decision making using Decision Making Games, and create technologies and cognitive computational models to support decision making and training.

Dr. Cleotilde Gonzalez is affiliated faculty with the CyLab Security and Privacy Institute, The HCII Human-Computer Interaction Institute, The Societal Computing program, and The CNBC Center for Neural Basis of Cognition at Carnegie Mellon University. She is a lifetime fellow of the Cognitive Science Society and of the Human Factors and Ergonomics Society. She is also a member of the Governing Board of the Cognitive Science Society. She is a Senior Editor for Topics in Cognitive Science, a Consulting Editor for Decision, and Associate Editor for the System Dynamics Review. She is also a member of editorial boards in multiple other journals including: Cognitive Science, Psychological Review, Perspectives on Psychological Science, and others.

[+] personal website

Título: Advanced Techniques for State-Space Exploration: Beyond Explicit Search

Ponente: Álvaro Torralba, Aassociate Professor, Aalborg University

Sesiones:

• 12 y 14 de Junio: de 15:00 a 18:30
• 16 de Junio: de 15:00 a 18:00

Lugar: Por definir

Organizador: Carlos Linares

Resumen:

State-space exploration is a fundamental technique with applications in diverse areas such as AI planning or software verification, for example. The main principle is to reason about the future consequences of our actions, in order to find a plan (course of action) that best fulfills our goals. In the context of planning, explicit state space search has been the most popular approach in the last decades. In this seminar we will take a closer look at several ideas that can be used to explore the state space in different ways, with the potential of exponentially reducing the search effort compared to explicit search.

On the one hand, symbolic and decoupled search are two alternatives to explicit search that have recently caught interest. Having a main idea -- search over sets of states – in common, still have very individual strengths (and weaknesses). Symbolic search can Benefit from data-structures like BDDs to efficiently represent arbitrary sets of states. We will cover both the basics as well as recent advances like the combination with heuristics or its application for problems beyond classical planning.  Decoupled search constructs sets of states in a more selective manner, by identifying conditionally independent components to directly exploit the structure of the problem.

On the other hand, contrastive analysis techniques break the paradigm of analyzing all states independently of each other. Instead, by comparing states to each other, it can discard those alternatives that cannot lead to better solutions.

Topics include:

• State-space exploration: an ubiquitous problem-solving technique
• Symbolic Search: from single states to set of states
• Decoupled Search: leveraging the structure of your problem
• Contrastive Analysis: considering the relationship between states
• Applications beyond Classical Planning: including top-k planning (when you need more than a single plan) and Stackelberg planning (a restricted form of 2-player games)

Breve biografía:

Álvaro Torralba is Associate Professor at Aalborg University. His main research area is AI planning: dealing with how to design agents that are able to reason about the long-term consequences of their actions. His interests are both in developing the theory, to better understand the challenges behind the process of intelligent decision-making, and practical algorithms that can be used in diverse industrial applications.