Schedule
Complexity From Cells to Consciousness:
Free Energy, Integrated Information, and Epsilon Machines


Thessaloniki, Greece  Thursday, September 27, 2018


08:40  09:00

Introductory Remarks

Brennan Klein  Conor Heins

I would not be surprised

09:00  09:45

Keynote Talk

Karl Friston

I am therefore I think

09:45  10:30

Keynote Talk

Jessica Flack

Collective computation: Towards a “statistical mechanics” for
information processing systems

10:30  11:00  Coffee Break and Poster Viewing  
11:00  11:30

Invited Talk

Rosalyn Moran

The Free Energy Principle plays Doom: a comparison with
rewardbased decision making in artificial intelligence environments

11:30  12:00

Invited Talk

Martin Biehl

The intrinsic motivation in active inference and possible alternatives 
12:00  12:20

Contributed Talk

Kai Ueltzhöffer

Deep active inference

12:20  12:40

Contributed Talk

Christopher Lynn

Structure from noise: Mental errors yield abstract
representations of events

12:40  13:00

Contributed Talk

Thomas Parr

Frontoparietal connections and active inference

13:00  14:30

Lunch Break


14:30  15:00

Invited Talk

William Marshall

Integrated information: From consciousness to cells

15:00  15:30

Invited Talk

Erik Hoel

Quantifying emergence and reduction in complex systems

15:30  16:00

Invited Talk

Jayne Thompson

Causal asymmetry in a quantum world

16:00  16:30

Coffee Break and Poster Viewing


16:30  17:00

Invited Talk

Felix Pollock

Interference and inference: Quantum stochastic processes
and the Free Energy Principle

17:00  17:30

Invited Talk

Mile Gu

Quantum Simplicity: How quantum agents can witness
simpler reality

17:40  18:30  Panel Discussion 
From Cells to Consciousness:
Karl Friston, Jessica Flack, Mile Gu, & Rosalyn Moran, led by Jakob Hohwy


18:30  19:30  Poster Presentations  Cocktail hour, discussion, and closing remarks 
Speakers and Talks
Introductory Remarks: Brennan Klein  Conor Heins  I would not be surprised (8:40am  9:00am)
Northeastern University  Max Planck Institute for Dynamics & SelfOrganization
Keynote Talk: Karl Friston  I am therefore I think (9:00am  9:45am)
University College London
This overview of the free energy principle offers an account of embodied exchange with the world that associates neuronal operations with actively inferring the causes of our sensations. Its agenda is to link formal (mathematical) descriptions of dynamical systems to a description of perception in terms of beliefs and goals. The argument has two parts: the first calls on the lawful dynamics of any (weakly mixing) system — from a single cell organism to a human brain. These lawful dynamics suggest that (internal) states can be interpreted as modelling or predicting the (external) causes of sensory fluctuations. In other words, if a system exists, its internal states must encode probabilistic beliefs about external states. Heuristically, this means that if I exist (am) then I must have beliefs (think). The second part of the argument is that the only tenable beliefs I can entertain about myself are that I exist. This may seem rather obvious; however, it transpires that this is equivalent to believing that the world — and the way it is sampled — will resolve uncertainty about the causes of sensations. We will consider the implications for functional anatomy, in terms of predictive coding and hierarchical architectures in the brain. We will conclude by looking at the epistemic behaviour that emerges using simulations of active inference.
Keynote Talk: Jessica Flack  Collective computation: Towards a “statistical mechanics” for information processing systems (9:45am  10:30am)
Santa Fe Institute
Physics produces order though the minimization of energy. Adaptive systems produce order through the addition of information processing. Why do adaptive systems have this extra step and does it make them fundamentally subjective, uncharacterizable by laws and unamenable to prediction? A natural point of entry into this debate is to ask how nature overcomes subjectivity to produce ordered states. We propose adaptive systems overcome subjectivity by collectively computing slowly changing coarsegrained microstates that reduce uncertainty about the future. I will discuss these issues, introduce a framework for studying collective computation and micro to maps in information processing systems, propose some principles, and pose open questions including what the relationship is between the theory of collective computation and other theories for the origins of scale.
Invited Talk: Rosalyn Moran  The Free Energy Principle plays Doom: a comparison with reward based decision making in artificial intelligence environments (11:00am  11:30am)
King's College London
Under Active Inference (Friston 2009), a decision — such as that to move one’s eyes — is driven by the imperative to minimise a bound on surprise known as the Free Energy. In the context of partially observable Markov decision processes (POMDPs), a modelbased framework in which we can cast naturalistic decisionmaking tasks, the Free Energy of a policy (a sequence of actions) can be understood as a drive to both minimise cost (maximise the likelihood of achieving a goal) while maximising the information return from a given set of actions. This scheme has been used to model decision making in tasks such as ‘the urn task’ and also in reading. In my talk I will introduce the technical framework of Free Energy minimisation in the context of online gaming environments (designed to test artificial intelligence algorithms) and present data from decisionmaking simulations. Specifically I will present the game ‘Doom’ and compare agents trained under Active Inference to agents trained to maximise reward. Linking these simulations to putative neurobiological substrates I will describe the potential links from brain to computation.
Invited Talk: Martin Biehl  The intrinsic motivation in active inference and possible alternatives (11:30am  12:00pm)
Araya Inc.
Active inference as proposed by Karl Friston combines model updating due to experience and action selection according to the predicted consequences of actions in an optimization of a single functional. In the original formulation the consequences of actions are evaluated by the "expected free energy". This expected free energy satisfies the conditions of an intrinsic motivation. On the one hand, this means that it can be used in a reinforcement learning setup like other intrinsic motivations. On the other hand we find that other intrinsic motivations can also be used in active inference. In this talk I will present a formulation of active inference and show how the expected free energy can be replaced by other intrinsic motivation functions from the literature while keeping the rest of the active inference framework intact.
Contributed Talk: Kai Ueltzhöffer  Deep active inference (12:00pm  12:20pm)
University of Heidelberg
Contributed Talk: Christopher Lynn  Structure from noise: Mental errors yield abstract representations of events (12:20pm  12:40pm)
University of Pennsylvania
Contributed Talk: Thomas Parr  Frontoparietal connections and active inference (12:40pm  13:00pm)
University College London
Invited Talk: William Marshall  Integrated information: From consciousness to cells (14:30pm  15:00pm)
University of Wisconsin
Integrated information theory starts from the phenomenology and identifies five fundamental properties of every experience (axioms). It then postulates that there must be a reason for these properties and translates the axioms into a set of postulates about the physical substrate of consciousness. I will review two of the core mathematical ideas the derive from the postulates — integrated information and causeeffect structures. I will then outline how integrated information and causeeffect structures can be used as a measure of the complexity of a system from its own intrinsic perspective. As a demonstration, I will present results from applying the integrated information framework to a Boolean network model of the fission yeast (Schizosaccharomyces pombe) cellcycle. Finally, I will discussion potential connections with control and artificial life.
Invited Talk: Erik Hoel  Quantifying emergence and reduction in complex systems (15:00pm  15:30pm)
Tufts University
Many physical systems can be coherently described in terms of their function and causal structure at multiple different levels. How we can reconcile these seemingly disparate levels of description? This is especially problematic because the lower scales at first appear more fundamental in three ways: in terms of their causal work, in terms of the amount of information they contain, and their theoretical superiority in terms of model choice. However, recent research bringing information theory and causal analysis to bear on modeling systems at different scales significantly reframes the issue, revealing that higher scales can be "causal codes" that allow for the generation of additional information through error correction. This result has significant implications for causal model choice in science and engineering. The findings indicate how emergence and reduction can be identified, measured, and used to design optimally informative experiments.
Invited Talk: Jayne Thompson  Causal asymmetry in a quantum world (15:30pm  16:00pm)
National University of Singapore
How can we observe an asymmetry in the temporal order of events when physics at the quantum level is time symmetric? The source of time’s barbed arrow is a longstanding puzzle in foundational science. Causal asymmetry offers a provocative perspective. It asks how Occam’s razor — the principle of assuming no more causes of natural things than are both true and sufficient to explain their appearances — can privilege one particular temporal direction over another. That is, if we want to model a process causally — such that the model makes statistically correct future predictions based only on information from the past  what is the minimum past information we must store? Are we forced to store more data if we model events in one particular temporal order over the other? Surprisingly most stochastic processes display nonzero causal asymmetry — implying that there is a privileged temporal direction when seeking the simplest causal model capable of explaining these events. Models running in the opposite temporal direction are penalized with an unavoidable memory overhead. This has been cited as a potential source of time's barbed arrow in complex processes. Here we examine what happens to this causal asymmetry in the quantum domain.
Invited Talk: Felix Pollock  Interference and inference: Quantum stochastic processes and the Free Energy Principle (16:30pm  17:00pm)
Monash University
Friston's free energy principle follows as a direct consequence of the stochastic evolution of any system with a Markov blanket (under some very loose assumptions). However, to the best of our knowledge, it is quantum mechanics that fundamentally underpins the behaviour of all physical systems, with the deterministic Schrödinger equation governing evolution. Using a new framework for describing quantum stochastic processes, I will show that the notion of a Markov blanket naturally emerges in composite quantum systems, before exploring how this could lead to a more general free energy principle that emerges from deterministic quantum physics.
Invited Talk: Mile Gu  Quantum Simplicity: How quantum agents can witness simpler reality (17:00pm  17:30pm)
Nanyang Technological University
To thrive in the complex environments, intelligent agents must be capable for anticipating future events, based on observations and actions they made in the past. The more complex this environment, the more memory an agent must devote to tracking the past, to generate statistically correct future predictions. In this presentation, I explore the question: Could an agent capable of harnessing quantum information processing have an operational advantage over classical counterparts? I outline our recent works showing how one can construct quantum agents, that are capable of exhibit the same degree of complex adaptive behaviour as classical counterparts, while using less memory than classical counterparts. I then discuss how these results challenging current views of what is complex, and highlight scenarios where quantum agents could exhibit extreme operational advantage.
Closing Panel Discussion: The Complex Systems future of the free energy principle, integrated information theory, and epsilon machines (17:40pm  18:30pm)
with Karl Friston, Jessica Flack, Mile Gu, Rosalyn Moran, moderated by Jakob Hohwy (Monash University).
Poster Presentations: Cocktail hour, discussion, and closing remarks (18:30pm)
Poster Presentations
Thijs van de Laar
ForneyLab: A toolbox for biologically plausible free energy minimization in dynamic neural models
Shervin Safavi
From optimal efficient coding to criticality
Tim Verbelen
Deep active inference for state estimation by learning from demonstration
Sergio Rubin
Does Gaia minimize free energy?
Genji Kawakita
The impact of network structures on the dynamics of decisionmaking processes
Dobromir Dotov (presented by Carlos Gershenson)
What is the causal depth of generative models in learning of complex dynamics?
Call for submissions
The Organizing Committee of Complexity from Cells to Consciousness is pleased to announce the Call for Submissions for this satellite at this year’s Conference on Complex Systems. We have a packed schedule of invited speakers and will try to highlight as many interdisciplinary submissions as possible. We are now accepting submissions for Poster Presentations and a limited number of Contributed Talks.
The goal of this workshop is to bring together researchers studying fundamental questions around information, complexity, emergence and scale. In this fullday satellite, we will hear talks about unifying frameworks that aim to account for the structure and dynamics of complex systems across scales and domains. In particular, we are emphasizing the ability of frameworks like the Free Energy Principle and Integrated Information Theory to explain the emergent teleology of complex systems. In addition, we will hear talks exploring the application of new modeling tools from information theory and complexity, such as Epsilon Machines.
Particular attention will be given to the following topics:
• Connections between statistical physics and causality, prediction, consciousness, and control
• Artificial intelligence, artificial life, exploration/exploitation, and agentbased models
• Emergent behavior in complex networks, largescale social/political systems, or crowds
• Nonlinear dynamics, statistical physics, climate science
• Philosophy of science, falsifiablility, epistemics
The satellite format will include frequent breaks for conversation and discussion around the various poster presentations. We encourage submissions that describe novel applications or interpretations of these informationthermodynamical frameworks. We are excited to hear about the ways these principles might manifest in different domains (like yours!). We are therefore happy to invite interested researchers from any discipline of Complex Systems research to submit.
Important Details
• Abstract Submission Deadline: June 30, 2018
• Abstract Submission Guidelines: PDF format, max. 500 words, 1 figure, submitted via email to:
 Brennan Klein (klein.br *at* northeastern.edu) & Conor Heins (conor.heins *at* ds.mpg.de). Please do not hesitate to reach out if you have questions.
Organizing Committee
Brennan Klein, Northeastern University  Conor Heins, Max Planck Institute for Dynamics & SelfOrganization  Rosalyn Moran, King’s College London  Timothy Bayne, Monash University  Jakob Hohwy, Monash University  Kavan Modi, Monash University  Naotsugu Tsuchiya, Monash University
This event is possible with support from the Monash University Network of Excellence for Causation & Complexity in the Conscious Brain.