Dimitrije Markovic

Power laws and self-organized criticality in theory and nature

Abstract Power laws and distributions with heavy tails are common features of many complex systems. Examples are the distribution of earthquake magnitudes, solar flare intensities and the sizes of neuronal avalanches. Previously, researchers surmised that a single general concept may act as an underlying generative mechanism, with the theory of self organized criticality being a weighty contender. The power-law scaling observed in the primary statistical analysis is an important, but by far not the only feature characterizing experimental data. The scaling function, the distribution of energy fluctuations, the distribution of inter-event waiting times, and other higher order spatial and temporal correlations, have seen increased consideration over the last years. Leading to realization that basic models, like the original sandpile model, are often insufficient to adequately describe the complexity of real-world systems with power-law distribution. Consequently, a substantial amount of effort has gone into developing new and extended models and, hitherto, three classes of models have emerged. The first line of models is based on a separation between the time scales of an external drive and an internal dissipation, and includes the original sandpile model and its extensions, like the dissipative earthquake model. Within this approach the steady state is close to criticality in terms of an absorbing phase transition. The second line of models is based on external drives and internal dynamics competing on similar time scales and includes the coherent noise model, which has a non-critical steady state characterized by heavy-tailed distributions. The third line of models proposes a non-critical self-organizing state, being guided by an optimization principle, such as the concept of highly optimized tolerance. We present a comparative overview regarding distinct modeling approaches together with a discussion of their potential relevance as underlying generative models for real-world phenomena. The complexity of physical and biological scaling phenomena has been found to transcend the explanatory power of individual paradigmal concepts. The interaction between theoretical development and experimental observations has been very fruitful, leading to a series of novel concepts and insights.

Intrinsic adaptation in autonomous recurrent neural networks

A massively recurrent neural network responds on one side to input stimuli and is autonomously active, on the other side, in the absence of sensory inputs. Stimuli and information processing depend crucially on the qualia of the autonomous-state dynamics of the ongoing neural activity. This default neural activity may be dynamically structured in time and space, showing regular, synchronized, bursting, or chaotic activity patterns. We study the influence of nonsynaptic plasticity on the default dynamical state of recurrent neural networks. The nonsynaptic adaption considered acts on intrinsic neural parameters, such as the threshold and the gain, and is driven by the optimization of the information entropy. We observe, in the presence of the intrinsic adaptation processes, three distinct and globally attracting dynamical regimes: a regular synchronized, an overall chaotic, and an intermittent bursting regime. The intermittent bursting regime is characterized by intervals of regular flows, which are quite insensitive to external stimuli, interceded by chaotic bursts that respond sensitively to input signals. We discuss these findings in the context of self-organized information processing and critical brain dynamics.

Self-organized chaos through polyhomeostatic optimization.

The goal of polyhomeostatic control is to achieve a certain target distribution of behaviors, in contrast to homeostatic regulation, which aims at stabilizing a steady-state dynamical state. We consider polyhomeostasis for individual and networks of firing-rate neurons, adapting to achieve target distributions of firing rates maximizing information entropy. We show that any finite polyhomeostatic adaption rate destroys all attractors in Hopfield-like network setups, leading to intermittently bursting behavior and self-organized chaos. The importance of polyhomeostasis to adapting behavior in general is discussed.

Neuropsychological constraints to human data production on a global scale

Abstract Which are the factors underlying human information production on a global level? In order to gain an insight into this question we study a corpus of 252–633 mil. publicly available data files on the Internet corresponding to an overall storage volume of 284–675 Terabytes. Analyzing the file size distribution for several distinct data types we find indications that the neuropsychological capacity of the human brain to process and record information may constitute the dominant limiting factor for the overall growth of globally stored information, with real-world economic constraints having only a negligible influence. This supposition draws support from the observation that the files size distributions follow a power law for data without a time component, like images, and a log-normal distribution for multimedia files, for which time is a defining qualia.

Inferring Neuronal Dynamics from Calcium Imaging Data Using Biophysical Models and Bayesian Inference

Calcium imaging has been used as a promising technique to monitor the dynamic activity of neuronal populations. However, the calcium trace is temporally smeared which restricts the extraction of quantities of interest such as spike trains of individual neurons. To address this issue, spike reconstruction algorithms have been introduced. One limitation of such reconstructions is that the underlying models are not informed about the biophysics of spike and burst generations. Such existing prior knowledge might be useful for constraining the possible solutions of spikes. Here we describe, in a novel Bayesian approach, how principled knowledge about neuronal dynamics can be employed to infer biophysical variables and parameters from fluorescence traces. By using both synthetic and in vitro recorded fluorescence traces, we demonstrate that the new approach is able to reconstruct different repetitive spiking and/or bursting patterns with accurate single spike resolution. Furthermore, we show that the high inference precision of the new approach is preserved even if the fluorescence trace is rather noisy or if the fluorescence transients show slow rise kinetics lasting several hundred milliseconds, and inhomogeneous rise and decay times. In addition, we discuss the use of the new approach for inferring parameter changes, e.g. due to a pharmacological intervention, as well as for inferring complex characteristics of immature neuronal circuits.

Vertex routing models

A class of models describing the flow of information within networks via routing processes is proposed and investigated, concentrating on the effects of memory traces on the global properties. The long-term flow of information is governed by cyclic attractors, allowing to define a measure for the information centrality of a vertex given by the number of attractors passing through this vertex. We find the number of vertices having a nonzero information centrality to be extensive/subextensive for models with/without a memory trace in the thermodynamic limit. We evaluate the distribution of the number of cycles, of the cycle length and of the maximal basins of attraction, finding a complete scaling collapse in the thermodynamic limit for the latter. Possible implications of our results for the information flow in social networks are discussed.

Modeling the Evolution of Beliefs Using an Attentional Focus Mechanism

For making decisions in everyday life we often have first to infer the set of environmental features that are relevant for the current task. Here we investigated the computational mechanisms underlying the evolution of beliefs about the relevance of environmental features in a dynamical and noisy environment. For this purpose we designed a probabilistic Wisconsin card sorting task (WCST) with belief solicitation, in which subjects were presented with stimuli composed of multiple visual features. At each moment in time a particular feature was relevant for obtaining reward, and participants had to infer which feature was relevant and report their beliefs accordingly. To test the hypothesis that attentional focus modulates the belief update process, we derived and fitted several probabilistic and non-probabilistic behavioral models, which either incorporate a dynamical model of attentional focus, in the form of a hierarchical winner-take-all neuronal network, or a diffusive model, without attention-like features. We used Bayesian model selection to identify the most likely generative model of subjects’ behavior and found that attention-like features in the behavioral model are essential for explaining subjects’ responses. Furthermore, we demonstrate a method for integrating both connectionist and Bayesian models of decision making within a single framework that allowed us to infer hidden belief processes of human subjects.

Comparative Analysis of Behavioral Models for Adaptive Learning in Changing Environments

Probabilistic models of decision making under various forms of uncertainty have been applied in recent years to numerous behavioral and model-based fMRI studies. These studies were highly successful in enabling a better understanding of behavior and delineating the functional properties of brain areas involved in decision making under uncertainty. However, as different studies considered different models of decision making under uncertainty, it is unclear which of these computational models provides the best account of the observed behavioral and neuroimaging data. This is an important issue, as not performing model comparison may tempt researchers to over-interpret results based on a single model. Here we describe how in practice one can compare different behavioral models and test the accuracy of model comparison and parameter estimation of Bayesian and maximum-likelihood based methods. We focus our analysis on two well-established hierarchical probabilistic models that aim at capturing the evolution of beliefs in changing environments: Hierarchical Gaussian Filters and Change Point Models. To our knowledge, these two, well-established models have never been compared on the same data. We demonstrate, using simulated behavioral experiments, that one can accurately disambiguate between these two models, and accurately infer free model parameters and hidden belief trajectories (e.g., posterior expectations, posterior uncertainties, and prediction errors) even when using noisy and highly correlated behavioral measurements. Importantly, we found several advantages of Bayesian inference and Bayesian model comparison compared to often-used Maximum-Likelihood schemes combined with the Bayesian Information Criterion. These results stress the relevance of Bayesian data analysis for model-based neuroimaging studies that investigate human decision making under uncertainty.

Adaptive Interface for Mapping Body Movements to Sounds.

Contemporary digital musical instruments allow an abundance of means to generate sound. Although superior to traditional instruments in terms of producing a unique audio-visual act, there is still an unmet need for digital instruments that allow performers to generate sounds through movements in an intuitive manner. One of the key factors for an authentic digital music act is a low latency between movements (user commands) and corresponding sounds. Here we present such a low-latency interface that maps the user’s kinematic actions into sound samples. The interface relies on wireless sensor nodes equipped with inertial measurement units and a real-time algorithm dedicated to the early detection and classification of a variety of movements/gestures performed by a user. The core algorithm is based on the approximate inference of a hierarchical generative model with piecewise-linear dynamical components. Importantly, the model’s structure is derived from a set of motion gestures. The performance of the Bayesian algorithm was compared against the k-nearest neighbors (k-NN) algorithm, which showed the highest classification accuracy, in a pre-testing phase, among several existing state-of-the-art algorithms. In almost all of the evaluation metrics the proposed probabilistic algorithm outperformed the k-NN algorithm.

Deterministic response strategies in trial-and-error learning

Trial-and-error learning is a universal strategy for establishing which actions are beneficial or harmful in new environments. However, learning stimulus-response associations solely via trial-and-error is often suboptimal, as in many settings dependencies among stimuli and responses can be exploited to increase learning efficiency. Previous studies have shown that in settings featuring such dependencies, humans typically engage high-level cognitive processes and employ advanced learning strategies to improve their learning efficiency. Here we analyze in detail the initial learning phase of a sample of human subjects (N = 85) performing a trial-and-error learning task with deterministic feedback and hidden stimulus-response dependencies. Using computational modeling, we find that the standard Q-learning model cannot sufficiently explain human learning strategies in this setting. Instead, newly introduced deterministic response models, which are theoretically optimal and transform stimulus sequences unambiguously into response sequences, provide the best explanation for 50.6% of the subjects. Most of the remaining subjects either show a tendency towards generic optimal learning (21.2%) or at least partially exploit stimulus-response dependencies (22.3%), while a few subjects (5.9%) show no clear preference for any of the employed models. After the initial learning phase, asymptotic learning performance during the subsequent practice phase is best explained by the standard Q-learning model. Our results show that human learning strategies in trial-and-error learning go beyond merely associating stimuli and responses via incremental reinforcement. Specifically during initial learning, high-level cognitive processes support sophisticated learning strategies that increase learning efficiency while keeping memory demands and computational efforts bounded. The good asymptotic fit of the Q-learning model indicates that these cognitive processes are successively replaced by the formation of stimulus-response associations over the course of learning.