Periodic Reporting for period 1 - INTENSS (INformation-Theoretic analysis of Embodied and Situated Systems)
Reporting period: 2018-10-01 to 2020-09-30
This project proposes and develops an analytical framework based on information-theoretic concepts and mathematical tools that may overcome such limitations. The focus is the study of human and robotic behaviour using kinematic data utilising as a case study the notion of multiple affordances, i.e. action possibilities offered by the environment to the cognitive agent.
A crucial objective of the project consists of the development and refinement of the individual skills of the researchers. Specifically, the Fellow gained the necessary abilities to lead a psychological experiment and organise and lead research teams, extending his scientific network globally and locally.
WP1. The Fellow utilised mathematical models such as coupled chaotic systems to developed the analytical framework grounded in information-theoretic concepts. The information-theoretic measures are applied to variables whose interactions are known a priori, thus assessing the validity of the analysis.
A particular focus has been on the notion of synergy, which assesses the additional contribution of a set of variables considered as a whole compared to the same variables considered in isolation.
A critical issue tackled in this preliminary work is estimating the probabilities from the data, a necessary and critical step for a correct evaluation of such measures.
WP2. The Fellow implemented two experiments based on robotic systems controlled by artificial neural networks. The first experiment employs a two-wheeled real robotic platform performing a wall-following in a simple maze. The second experiment trains a simulated robotic arm interacting with objects populating the environment. Data recorded from sensors and actuators are analysed with the developed information-theoretic approach.
The impossibility of finalising the experiment of behavioural psychology caused by the ongoing pandemic, the Fellow is developing an information-theoretic algorithm to minimise the number of parameters (e.g. the size of the neural network) in deep learning models. This topic is a current challenge for the machine learning community, and the research work conducted in this project naturally transfers to this field.
Furthermore, the Fellow collaborated with the Laboratory of Embodied Computational Neuroscience to develop a cognitive architecture based on deep neural networks capable of autonomously reducing the complexity of the visual input during training depending on the task.
WP3. The Fellow was involved as a collaborator in two works of experimental psychology, receiving excellent training by collaborating with his Supervisor and the members of the laboratory.
Two experiments strictly related to the research plan led by the Fellow were under implementation in the final months of the project. These studies aimed at investigating the phenomenon of multiple affordances utilising an immersive virtual reality headset and a kinematic glove. Unfortunately, the pandemic outbreak is preventing data collection imposing a suspension of data collection.
To finalise his training in experimental psychology, the Fellow is leading and coordinating an experiment investigating cross-cultural differences in abstract and concrete words processing. This research involves an international team that includes a Japanese private company part of the Fellow's scientific network.
The Fellow contributed to the preparation of a position article on the topic of abstract concept processing from an embodied perspective.
The analytical approach may find application in the health sector aiding the early assessment of neurological diseases affecting motor responses or supporting the rehabilitation of patients with motor impairments. The developed data analysis may also find application in an industrial setting, for example detecting faults in autonomous robots.
From a scientific perspective, the proposed method is not limited to find application in the field of embodied cognitive science. The case study is a behavioural experiment analysed with a data-driven and model-free methodology. Hence, the framework is agnostic towards strong assumptions concerning the underlying systems and the adherence to a specific scientific theory. This leads to possible applications to other scientific fields. Furthermore, the study of both artificial and biological embodied systems secures great visibility in the field of embodied cognitive science.