Periodic Reporting for period 1 - SYNERGINFO (Developments on the relevance of SYNERGistic INFOrmation sharing modes)
Periodo di rendicontazione: 2017-05-01 al 2019-04-30
"The goal of this project was to study the relevance of relationships between three or more variables that cannot be reduced to a combination of pairwise relationships. One of the main drivers behind Complexity Science is that to understand many systems is ""less the matter and more the patterns""; i.e. the relationship between the constituent elements might be more relevant that the constituent themselves. Traditionally, this view has lead to the growing discipline of ""Complex Networks"", where pairwise relationships are lumped into a graph/network, whose global properties are then studied. In contrast, this project is about studying which relationships cannot be reflected by such network.
This topic is very important for society, as it allow us to expand our understanding of what it means to be ""together"". Intuitively, we tend to think that ""togetherness"" can be only more or less, and hence can be quantified by a single number that assess it intensity. In contrast, this project explore different ""modes"" in which variables can be together. Given the political situation where trends towards a global society seem to be in direct opposition to nationalists movements that seek local independency, better understanding these modes might help to bring more dimensions into the discussion.
The goal of the project are twofold:
1) To advance our theoretical understanding of these diverse modes of multivariate interdependency, and
2) Explore the implications that these modes might have in real-word scenarios.
The scenarios studied in this project are very diverse, including the dynamics of the human brain, decentralised social systems, and data privacy."
This topic is very important for society, as it allow us to expand our understanding of what it means to be ""together"". Intuitively, we tend to think that ""togetherness"" can be only more or less, and hence can be quantified by a single number that assess it intensity. In contrast, this project explore different ""modes"" in which variables can be together. Given the political situation where trends towards a global society seem to be in direct opposition to nationalists movements that seek local independency, better understanding these modes might help to bring more dimensions into the discussion.
The goal of the project are twofold:
1) To advance our theoretical understanding of these diverse modes of multivariate interdependency, and
2) Explore the implications that these modes might have in real-word scenarios.
The scenarios studied in this project are very diverse, including the dynamics of the human brain, decentralised social systems, and data privacy."
"In the theoretical aspects, we developed a new framework to study multivariate correlations in systems described by multiple variables. Our framework contain various aspects:
- New notions of how information can be stored in multi-agent systems: We develop an information-theoretic framework to quantify various modes of information storage that can take place in a multi-agent system. Moreover, we shown how these modes can be useful to characterise various phases in self-organisation, and can provide useful characterisation of information-processing capabilities of complex systems.
- Generalized metrics of statistical synergy: We developed a new metric, called ""O-information"", which can be used to assess to what extend high-order interdependencies (i.e. ones that involve many variables, but cannot be reduced to interaction between smaller numbers of variables) dominate over pairwise or other low-order interdependencies. Our framework can be used for characterising fundamental organizational properties in a data-driven model-free fashion. With this framework we were able to show surprising properties of the statistics of the scores of J.S. Bach, which are not present in the music of some of his contemporaries.
- New notions of information dynamics. Traditional causal discovery frameworks only consider single 'cause'-variables that act on single 'effect'-variables. We extend this view developing a framework that considers high-order causes, which furthermore might act on high-order effects. With this, we developed a fine taxonomy of information dynamics phenomena, which can be used to disentangle dissimilar phenomena that is conflated by popular metrics such as transfer entropy.
In the field of applications of our theories, our results include the following areas:
- With respect to data privacy, we developed a new paradigm to disclose data which keeping critical contents secure. In particular, our approach proposes to disclose collective properties of databases which keep the actual value of each data-sample perfectly private. For this, we developed practical algorithms, and studied asymptotic performance limits.
- In computational neuroscience, we showed that the brain activity of musicians show a higher complexity when they improvise. High brain complexity has been associated with high state of alertness and awareness. Interestingly, some audience members also show this increase, although the effect is not so consistent.
- In neuroscience of psychedelics, we found that brain complexity also increases systematically with the richness of stimuli. Moreover, brain complexity also increases with the usage of LSD; interestingly, after LSD has been consumed, the outside stimuli still increase the complexity but the effect decreases. This contributes to the better understanding of the clinical usage of psychedelics.
- In computational social sciences, we shows that social learning algorithms can be used to guarantee network robustness against data falsification attacks. Additionally, we showed how social diversity can introduce stability and better network long-term network performance.
Our work with music improvisation received plenty of media coverage, being featured in The Times and BBC. Additionally, we made two concert+talk activities: one at the Imperial's Late Festival (06-dec-2018), and one in the Guildhall School of Music (01-may-2019). We have scheduled a new concert+talk that will take in the Exhibition Road Festival (30-june-2019). We also disseminated our results in the workshop ""The power of musical networks"", co-organised with the Orpheus Institute, which took place in Gent, Belgium (21/22-feb-2018).
The results of the project was presented at multiple international conferences and meetings, including CompNet2017, CCS2017 and 2018, NetSci2018, and Worshops at Oxford, Amsterdamn, Hong-Kong, Enschede, Gent, Valparaiso and Cuernavaca."
- New notions of how information can be stored in multi-agent systems: We develop an information-theoretic framework to quantify various modes of information storage that can take place in a multi-agent system. Moreover, we shown how these modes can be useful to characterise various phases in self-organisation, and can provide useful characterisation of information-processing capabilities of complex systems.
- Generalized metrics of statistical synergy: We developed a new metric, called ""O-information"", which can be used to assess to what extend high-order interdependencies (i.e. ones that involve many variables, but cannot be reduced to interaction between smaller numbers of variables) dominate over pairwise or other low-order interdependencies. Our framework can be used for characterising fundamental organizational properties in a data-driven model-free fashion. With this framework we were able to show surprising properties of the statistics of the scores of J.S. Bach, which are not present in the music of some of his contemporaries.
- New notions of information dynamics. Traditional causal discovery frameworks only consider single 'cause'-variables that act on single 'effect'-variables. We extend this view developing a framework that considers high-order causes, which furthermore might act on high-order effects. With this, we developed a fine taxonomy of information dynamics phenomena, which can be used to disentangle dissimilar phenomena that is conflated by popular metrics such as transfer entropy.
In the field of applications of our theories, our results include the following areas:
- With respect to data privacy, we developed a new paradigm to disclose data which keeping critical contents secure. In particular, our approach proposes to disclose collective properties of databases which keep the actual value of each data-sample perfectly private. For this, we developed practical algorithms, and studied asymptotic performance limits.
- In computational neuroscience, we showed that the brain activity of musicians show a higher complexity when they improvise. High brain complexity has been associated with high state of alertness and awareness. Interestingly, some audience members also show this increase, although the effect is not so consistent.
- In neuroscience of psychedelics, we found that brain complexity also increases systematically with the richness of stimuli. Moreover, brain complexity also increases with the usage of LSD; interestingly, after LSD has been consumed, the outside stimuli still increase the complexity but the effect decreases. This contributes to the better understanding of the clinical usage of psychedelics.
- In computational social sciences, we shows that social learning algorithms can be used to guarantee network robustness against data falsification attacks. Additionally, we showed how social diversity can introduce stability and better network long-term network performance.
Our work with music improvisation received plenty of media coverage, being featured in The Times and BBC. Additionally, we made two concert+talk activities: one at the Imperial's Late Festival (06-dec-2018), and one in the Guildhall School of Music (01-may-2019). We have scheduled a new concert+talk that will take in the Exhibition Road Festival (30-june-2019). We also disseminated our results in the workshop ""The power of musical networks"", co-organised with the Orpheus Institute, which took place in Gent, Belgium (21/22-feb-2018).
The results of the project was presented at multiple international conferences and meetings, including CompNet2017, CCS2017 and 2018, NetSci2018, and Worshops at Oxford, Amsterdamn, Hong-Kong, Enschede, Gent, Valparaiso and Cuernavaca."
"Within complexity science, this project lead a new trend to go beyond pairwise networks and look for high-order relationships. While many graph-theoretic researchers are advancing in tools for studying so-called hyper-graphs (i.e. graphs whose links can connect more than two edges), our project has developed a unique method of deciding how to actually build such hypergraphs from data. At Imperial College, in association with the Departments of Mathematics, Department of Medicine, and the Data Science Institute, we are actively working on exploiting these tools to investigate these not-well explored aspects of data science, which might hold rich information to better exploit neuroimaging data.
With this project we have been able to develop a strong European network of collaborators in the area of human engagement through arts. Currently we are working with researchers from UK and Denmark to better understand the dynamics of various biometric indicators in collective activities, including the attendance of artistic performances (dance and music), collaborative work, adrenaline experiences, and others. It is our hope that the further development of these research efforts might help us to understand better the nature of human-to-human engagement, and hence appreciate what make us humans.
In the context of data privacy, we have been able to leverage high-order interactions to propose a novel scheme to guarantee data privacy. This scheme could eventually provide an alternative to ""differential privacy approach"", which despite it popularity doesn't have solid theoretical guarantees."
With this project we have been able to develop a strong European network of collaborators in the area of human engagement through arts. Currently we are working with researchers from UK and Denmark to better understand the dynamics of various biometric indicators in collective activities, including the attendance of artistic performances (dance and music), collaborative work, adrenaline experiences, and others. It is our hope that the further development of these research efforts might help us to understand better the nature of human-to-human engagement, and hence appreciate what make us humans.
In the context of data privacy, we have been able to leverage high-order interactions to propose a novel scheme to guarantee data privacy. This scheme could eventually provide an alternative to ""differential privacy approach"", which despite it popularity doesn't have solid theoretical guarantees."