Throughout my education and professional career I have gained experience and interest in various areas of Probability, Stochastic Processes, Operational Research, and Artificial Intelligence.

My academic achievements include a B.Sc. in Mathematics, major Probability and Statistics, by the University of the Basque Country (Basque Country, Spain) and a Ph.D. in Management Science by Lancaster University (Lancaster, UK). For my PhD, I developed a prototype distributed component- based simulation architecture in Java. The thesis tackles the issue of how to deal with simulation models that are becoming increasingly complex and large. The thesis demonstrates that it is possible and useful to apply component technologies to simulation approaches. The result of this work has been published in a book [Oses 2010].

My professional career spans over eight years, several companies and an evolving scientific direction.

My first job was at Barcrest Games (UK) where I worked as a mathematician developing probability models of slot machine games for the gambling industry. These models are necessary to calculate and adjust the long-term percentage return, i.e. the percentage of the collected money the game will return to the players in the long term. Modelling these games requires the use of Operational Research techniques, Probability and Stochastic Processes. This work provided material for several papers ([Oses and Freeman 2006], [Oses 2008], and [Oses 2009]). Techniques used during this phase included stochastic processes, optimisation, modelling and simulation, mathematical programming, probability, Markov chains, Markovian decision processes, and deterministic and stochastic dynamic programming.

I then moved on to Fundación Fatronik-Tecnalia (Spain) in order to develop my interests in Artificial Intelligence. These interests can be stated technically as ‘Probabilistic methods for uncertain reasoning’. I am interested in using probability under uncertainty for reasoning, planning, learning, perception, prediction, and decision making.

I also have an interest in decision theory. Sometimes knowing the probabilities of events is not enough to make a decision, we must also know the rewards or costs associated to the different options. A key concept from the science of economics is ‘utility’: a measure of how valuable something is to an intelligent agent. Precise mathematical tools have been developed that analyze how an agent can make choices and plan, using decision theory, decision analysis, information value theory. These tools include models such as Markov decision processes, dynamic decision networks, and game theory.

At Fatronik, I worked in the Neurengineering department. The department of neuroengineering was active in the field of applied neuroscience; our research primarily involved projects in the domains of Bio-inspired Devices and Algorithms through A.I. & Computational Neuroscience. As part of my work here, I travelled to Zürich (Switzerland) for a 4.5 month internship at the Artificial Intelligence Lab of the University of Zürich. My project while at the AI Lab was called ‘A robot thinking ahead’ and it was part of ‘From locomotion to cognition’, a Swiss National Science Foundation project (Grant Nr. 200020-122279/1). The objective of my project was to investigate the use of forward models in a quadruped robot.

The capability to be able to predict future states is a first step to cognition. The predictive model is a representation of the immediate sensory-motor reality and a potential basis for ‘thinking’ or mental imagery. In our case, we wanted the robot to be able to predict the consequences of its actions, where the ‘consequence’ was change in position and the actions were the gaits. In order to pose the right challenges to our robot, we came up with a predator-prey scenario. To conduct this research we developed a bio-inspired control architecture that allows a mobile robot to: (1) learn a model of its action repertoire (a forward model); (2) learn a model of an object's behavior (the prey model); (3) combine the forward model and the prey model to seek the prey. All the models are learned from scratch, without assumptions, work in egocentric coordinates, and are probabilistic in nature. This work was published in [Oses, Hoffmann, Koene 2010].

This project sparked my interest in learning algorithms and since then my long-term dream has been to develop learning algorithms so that we can have a robot learn and develop as a human baby. Babies learn by perceiving the world through their senses and interacting with it. So good robot perception and manipulation are essential.

Also as part of my work in the Neuroengineering department, and after my internship in Zürich, I discovered Jeff Hawkins‘ ‘On intelligence’ and Dileep George‘s ‘How the brain might work: a hierarchical and temporal model for learning and recognition’. I have been working with HTM for over a year. It is this technology that I have used to work on robot perception and manipulation during my internship at DFKI-Bremen as part of the Seegrip project. Specifically, at DFKI-Bremen I worked on the development of an HTM model that performs object recognition using tactile sensor data and also on the development of an HTM model to recognize the grasp being performed by an operator wearing a sensorized data glove.

References: