Detecting Information Relays in Deep Neural Networks.

Deep learning of artificial neural networks (ANNs) is creating highly functional processes that are, unfortunately, nearly as hard to interpret as their biological counterparts. Identification of functional modules in natural brains plays an important role in cognitive and neuroscience alike, and can be carried out using a wide range of technologies such as fMRI, EEG/ERP, MEG, or calcium imaging. However, we do not have such robust methods at our disposal when it comes to understanding functional modules in artificial neural networks. Ideally, understanding which parts of an artificial neural network perform what function might help us to address a number of vexing problems in ANN research, such as catastrophic forgetting and overfitting. Furthermore, revealing a network's modularity could improve our trust in them by making these black boxes more transparent. Here, we introduce a new information-theoretic concept that proves useful in understanding and analyzing a network's functional modularity: the relay information IR. The relay information measures how much information groups of neurons that participate in a particular function (modules) relay from inputs to outputs. Combined with a greedy search algorithm, relay information can be used to identify computational modules in neural networks. We also show that the functionality of modules correlates with the amount of relay information they carry.

Robustness of Sparsely Distributed Representations to Adversarial Attacks in Deep Neural Networks.

Deep learning models have achieved an impressive performance in a variety of tasks, but they often suffer from overfitting and are vulnerable to adversarial attacks. Previous research has shown that dropout regularization is an effective technique that can improve model generalization and robustness. In this study, we investigate the impact of dropout regularization on the ability of neural networks to withstand adversarial attacks, as well as the degree of "functional smearing" between individual neurons in the network. Functional smearing in this context describes the phenomenon that a neuron or hidden state is involved in multiple functions at the same time. Our findings confirm that dropout regularization can enhance a network's resistance to adversarial attacks, and this effect is only observable within a specific range of dropout probabilities. Furthermore, our study reveals that dropout regularization significantly increases the distribution of functional smearing across a wide range of dropout rates. However, it is the fraction of networks with lower levels of functional smearing that exhibit greater resilience against adversarial attacks. This suggests that, even though dropout improves robustness to fooling, one should instead try to decrease functional smearing.

Understanding Memories of the Past in the Context of Different Complex Neural Network Architectures.

Deep learning (primarily using backpropagation) and neuroevolution are the preeminent methods of optimizing artificial neural networks. However, they often create black boxes that are as hard to understand as the natural brains they seek to mimic. Previous work has identified an information-theoretic tool, referred to as R, which allows us to quantify and identify mental representations in artificial cognitive systems. The use of such measures has allowed us to make previous black boxes more transparent. Here we extend R to not only identify where complex computational systems store memory about their environment but also to differentiate between different time points in the past. We show how this extended measure can identify the location of memory related to past experiences in neural networks optimized by deep learning as well as a genetic algorithm.

Identifying patient-specific behaviors to understand illness trajectories and predict relapses in bipolar disorder using passive sensing and deep anomaly detection: protocol for a contactless cohort study.

BACKGROUND: Predictive models for mental disorders or behaviors (e.g., suicide) have been successfully developed at the level of populations, yet current demographic and clinical variables are neither sensitive nor specific enough for making individual clinical predictions. Forecasting episodes of illness is particularly relevant in bipolar disorder (BD), a mood disorder with high recurrence, disability, and suicide rates. Thus, to understand the dynamic changes involved in episode generation in BD, we propose to extract and interpret individual illness trajectories and patterns suggestive of relapse using passive sensing, nonlinear techniques, and deep anomaly detection. Here we describe the study we have designed to test this hypothesis and the rationale for its design. METHOD: This is a protocol for a contactless cohort study in 200 adult BD patients. Participants will be followed for up to 2 years during which they will be monitored continuously using passive sensing, a wearable that collects multimodal physiological (heart rate variability) and objective (sleep, activity) data. Participants will complete (i) a comprehensive baseline assessment; (ii) weekly assessments; (iii) daily assessments using electronic rating scales. Data will be analyzed using nonlinear techniques and deep anomaly detection to forecast episodes of illness. DISCUSSION: This proposed contactless, large cohort study aims to obtain and combine high-dimensional, multimodal physiological, objective, and subjective data. Our work, by conceptualizing mood as a dynamic property of biological systems, will demonstrate the feasibility of incorporating individual variability in a model informing clinical trajectories and predicting relapse in BD.

Open-Endedness for the Sake of Open-Endedness.

Natural evolution keeps inventing new complex and intricate forms and behaviors. Digital evolution and genetic algorithms fail to create the same kind of complexity, not just because we still lack the computational resources to rival nature, but because (it has been argued) we have not understood in principle how to create open-ended evolving systems. Much effort has been made to define such open-endedness so as to create forms of increasing complexity indefinitely. Here, however, a simple evolving computational system that satisfies all such requirements is presented. Doing so reveals a shortcoming in the definitions for open-ended evolution. The goal to create models that rival biological complexity remains. This work suggests that our current definitions allow for even simple models to pass as open-ended, and that our definitions of complexity and diversity are more important for the quest of open-ended evolution than the fact that something runs indefinitely.

A genome wide dosage suppressor network reveals genomic robustness.

Genomic robustness is the extent to which an organism has evolved to withstand the effects of deleterious mutations. We explored the extent of genomic robustness in budding yeast by genome wide dosage suppressor analysis of 53 conditional lethal mutations in cell division cycle and RNA synthesis related genes, revealing 660 suppressor interactions of which 642 are novel. This collection has several distinctive features, including high co-occurrence of mutant-suppressor pairs within protein modules, highly correlated functions between the pairs and higher diversity of functions among the co-suppressors than previously observed. Dosage suppression of essential genes encoding RNA polymerase subunits and chromosome cohesion complex suggests a surprising degree of functional plasticity of macromolecular complexes, and the existence of numerous degenerate pathways for circumventing the effects of potentially lethal mutations. These results imply that organisms and cancer are likely able to exploit the genomic robustness properties, due the persistence of cryptic gene and pathway functions, to generate variation and adapt to selective pressures.

The Evolution of Neuroplasticity and the Effect on Integrated Information.

Information integration theory has been developed to quantify consciousness. Since conscious thought requires the integration of information, the degree of this integration can be used as a neural correlate (Φ) with the intent to measure degree of consciousness. Previous research has shown that the ability to integrate information can be improved by Darwinian evolution. The value Φ can change over many generations, and complex tasks require systems with at least a minimum Φ . This work was done using simple animats that were able to remember previous sensory inputs, but were incapable of fundamental change during their lifetime: actions were predetermined or instinctual. Here, we are interested in changes to Φ due to lifetime learning (also known as neuroplasticity). During lifetime learning, the system adapts to perform a task and necessitates a functional change, which in turn could change Φ . One can find arguments to expect one of three possible outcomes: Φ might remain constant, increase, or decrease due to learning. To resolve this, we need to observe systems that learn, but also improve their ability to learn over the many generations that Darwinian evolution requires. Quantifying Φ over the course of evolution, and over the course of their lifetimes, allows us to investigate how the ability to integrate information changes. To measure Φ , the internal states of the system must be experimentally observable. However, these states are notoriously difficult to observe in a natural system. Therefore, we use a computational model that not only evolves virtual agents (animats), but evolves animats to learn during their lifetime. We use this approach to show that a system that improves its performance due to feedback learning increases its ability to integrate information. In addition, we show that a system's ability to increase Φ correlates with its ability to increase in performance. This suggests that systems that are very plastic regarding Φ learn better than those that are not.

Episode forecasting in bipolar disorder: Is energy better than mood?

OBJECTIVE: Bipolar disorder is a severe mood disorder characterized by alternating episodes of mania and depression. Several interventions have been developed to decrease high admission rates and high suicides rates associated with the illness, including psychoeducation and early episode detection, with mixed results. More recently, machine learning approaches have been used to aid clinical diagnosis or to detect a particular clinical state; however, contradictory results arise from confusion around which of the several automatically generated data are the most contributory and useful to detect a particular clinical state. Our aim for this study was to apply machine learning techniques and nonlinear analyses to a physiological time series dataset in order to find the best predictor for forecasting episodes in mood disorders. METHODS: We employed three different techniques: entropy calculations and two different machine learning approaches (genetic programming and Markov Brains as classifiers) to determine whether mood, energy or sleep was the best predictor to forecast a mood episode in a physiological time series. RESULTS: Evening energy was the best predictor for both manic and depressive episodes in each of the three aforementioned techniques. This suggests that energy might be a better predictor than mood for forecasting mood episodes in bipolar disorder and that these particular machine learning approaches are valuable tools to be used clinically. CONCLUSIONS: Energy should be considered as an important factor for episode prediction. Machine learning approaches provide better tools to forecast episodes and to increase our understanding of the processes that underlie mood regulation.

Origin of life in a digital microcosm.

While all organisms on Earth share a common descent, there is no consensus on whether the origin of the ancestral self-replicator was a one-off event or whether it only represented the final survivor of multiple origins. Here, we use the digital evolution system Avida to study the origin of self-replicating computer programs. By using a computational system, we avoid many of the uncertainties inherent in any biochemical system of self-replicators (while running the risk of ignoring a fundamental aspect of biochemistry). We generated the exhaustive set of minimal-genome self-replicators and analysed the network structure of this fitness landscape. We further examined the evolvability of these self-replicators and found that the evolvability of a self-replicator is dependent on its genomic architecture. We also studied the differential ability of replicators to take over the population when competed against each other, akin to a primordial-soup model of biogenesis, and found that the probability of a self-replicator outcompeting the others is not uniform. Instead, progenitor (most-recent common ancestor) genotypes are clustered in a small region of the replicator space. Our results demonstrate how computational systems can be used as test systems for hypotheses concerning the origin of life.This article is part of the themed issue 'Reconceptualizing the origins of life'.

Punishment in public goods games leads to meta-stable phase transitions and hysteresis.

The evolution of cooperation has been a perennial problem in evolutionary biology because cooperation can be undermined by selfish cheaters who gain an advantage in the short run, while compromising the long-term viability of the population. Evolutionary game theory has shown that under certain conditions, cooperation nonetheless evolves stably, for example if players have the opportunity to punish cheaters that benefit from a public good yet refuse to pay into the common pool. However, punishment has remained enigmatic because it is costly and difficult to maintain. On the other hand, cooperation emerges naturally in the public goods game if the synergy of the public good (the factor multiplying the public good investment) is sufficiently high. In terms of this synergy parameter, the transition from defection to cooperation can be viewed as a phase transition with the synergy as the critical parameter. We show here that punishment reduces the critical value at which cooperation occurs, but also creates the possibility of meta-stable phase transitions, where populations can 'tunnel' into the cooperating phase below the critical value. At the same time, cooperating populations are unstable even above the critical value, because a group of defectors that are large enough can 'nucleate' such a transition. We study the mean-field theoretical predictions via agent-based simulations of finite populations using an evolutionary approach where the decisions to cooperate or to punish are encoded genetically in terms of evolvable probabilities. We recover the theoretical predictions and demonstrate that the population shows hysteresis, as expected in systems that exhibit super-heating and super-cooling. We conclude that punishment can stabilize populations of cooperators below the critical point, but it is a two-edged sword: it can also stabilize defectors above the critical point.

Evolution of integrated causal structures in animats exposed to environments of increasing complexity.

Natural selection favors the evolution of brains that can capture fitness-relevant features of the environment's causal structure. We investigated the evolution of small, adaptive logic-gate networks ("animats") in task environments where falling blocks of different sizes have to be caught or avoided in a 'Tetris-like' game. Solving these tasks requires the integration of sensor inputs and memory. Evolved networks were evaluated using measures of information integration, including the number of evolved concepts and the total amount of integrated conceptual information. The results show that, over the course of the animats' adaptation, i) the number of concepts grows; ii) integrated conceptual information increases; iii) this increase depends on the complexity of the environment, especially on the requirement for sequential memory. These results suggest that the need to capture the causal structure of a rich environment, given limited sensors and internal mechanisms, is an important driving force for organisms to develop highly integrated networks ("brains") with many concepts, leading to an increase in their internal complexity.

Scaling metagenome sequence assembly with probabilistic de Bruijn graphs.

Deep sequencing has enabled the investigation of a wide range of environmental microbial ecosystems, but the high memory requirements for de novo assembly of short-read shotgun sequencing data from these complex populations are an increasingly large practical barrier. Here we introduce a memory-efficient graph representation with which we can analyze the k-mer connectivity of metagenomic samples. The graph representation is based on a probabilistic data structure, a Bloom filter, that allows us to efficiently store assembly graphs in as little as 4 bits per k-mer, albeit inexactly. We show that this data structure accurately represents DNA assembly graphs in low memory. We apply this data structure to the problem of partitioning assembly graphs into components as a prelude to assembly, and show that this reduces the overall memory requirements for de novo assembly of metagenomes. On one soil metagenome assembly, this approach achieves a nearly 40-fold decrease in the maximum memory requirements for assembly. This probabilistic graph representation is a significant theoretical advance in storing assembly graphs and also yields immediate leverage on metagenomic assembly.

Sequence dependence of isothermal DNA amplification via EXPAR.

Isothermal nucleic acid amplification is becoming increasingly important for molecular diagnostics. Therefore, new computational tools are needed to facilitate assay design. In the isothermal EXPonential Amplification Reaction (EXPAR), template sequences with similar thermodynamic characteristics perform very differently. To understand what causes this variability, we characterized the performance of 384 template sequences, and used this data to develop two computational methods to predict EXPAR template performance based on sequence: a position weight matrix approach with support vector machine classifier, and RELIEF attribute evaluation with Naïve Bayes classification. The methods identified well and poorly performing EXPAR templates with 67-70% sensitivity and 77-80% specificity. We combined these methods into a computational tool that can accelerate new assay design by ruling out likely poor performers. Furthermore, our data suggest that variability in template performance is linked to specific sequence motifs. Cytidine, a pyrimidine base, is over-represented in certain positions of well-performing templates. Guanosine and adenosine, both purine bases, are over-represented in similar regions of poorly performing templates, frequently as GA or AG dimers. Since polymerases have a higher affinity for purine oligonucleotides, polymerase binding to GA-rich regions of a single-stranded DNA template may promote non-specific amplification in EXPAR and other nucleic acid amplification reactions.

Reducing Epistasis and Pleiotropy Can Avoid the Survival of the Flattest Tragedy.

This study investigates whether reducing epistasis and pleiotropy enhances mutational robustness in evolutionary adaptation, utilizing an indirect encoded model within the "survival of the flattest" (SoF) fitness landscape. By simulating genetic variations and their phenotypic consequences, we explore organisms' adaptive mechanisms to maintain positions on higher, narrower evolutionary peaks amidst environmental and genetic pressures. Our results reveal that organisms can indeed sustain their advantageous positions by minimizing the complexity of genetic interactions-specifically, by reducing the levels of epistasis and pleiotropy. This finding suggests a counterintuitive strategy for evolutionary stability: simpler genetic architectures, characterized by fewer gene interactions and multifunctional genes, confer a survival advantage by enhancing mutational robustness. This study contributes to our understanding of the genetic underpinnings of adaptability and robustness, challenging traditional views that equate complexity with fitness in dynamic environments.

Not missing at random: Missing data are associated with clinical status and trajectories in an electronic monitoring longitudinal study of bipolar disorder.

There is limited information on the association between participants' clinical status or trajectories and missing data in electronic monitoring studies of bipolar disorder (BD). We collected self-ratings scales and sensor data in 145 adults with BD. Using a new metric, Missing Data Ratio (MDR), we assessed missing self-rating data and sensor data monitoring activity and sleep. Missing data were lowest for participants in the midst of a depressive episode, intermediate for participants with subsyndromal symptoms, and highest for participants who were euthymic. Over a mean ± SD follow-up of 246 ± 181 days, missing data remained unchanged for participants whose clinical status did not change throughout the study (i.e., those who entered the study in a depressive episode and did not improve, or those who entered the study euthymic and remained euthymic). Conversely, when participants' clinical status changed during the study (e.g., those who entered the study euthymic and experienced the occurrence of a depressive episode), missing data for self-rating scales increased, but not for sensor data. Overall missing data were associated with participants' clinical status and its changes, suggesting that these are not missing at random.

The evolution of representation in simple cognitive networks.

Representations are internal models of the environment that can provide guidance to a behaving agent, even in the absence of sensory information. It is not clear how representations are developed and whether they are necessary or even essential for intelligent behavior. We argue here that the ability to represent relevant features of the environment is the expected consequence of an adaptive process, give a formal definition of representation based on information theory, and quantify it with a measure R. To measure how R changes over time, we evolve two types of networks--an artificial neural network and a network of hidden Markov gates--to solve a categorization task using a genetic algorithm. We find that the capacity to represent increases during evolutionary adaptation and that agents form representations of their environment during their lifetime. This ability allows the agents to act on sensorial inputs in the context of their acquired representations and enables complex and context-dependent behavior. We examine which concepts (features of the environment) our networks are representing, how the representations are logically encoded in the networks, and how they form as an agent behaves to solve a task. We conclude that R should be able to quantify the representations within any cognitive system and should be predictive of an agent's long-term adaptive success.

Impact of epistasis and pleiotropy on evolutionary adaptation.

Evolutionary adaptation is often likened to climbing a hill or peak. While this process is simple for fitness landscapes where mutations are independent, the interaction between mutations (epistasis) as well as mutations at loci that affect more than one trait (pleiotropy) are crucial in complex and realistic fitness landscapes. We investigate the impact of epistasis and pleiotropy on adaptive evolution by studying the evolution of a population of asexual haploid organisms (haplotypes) in a model of N interacting loci, where each locus interacts with K other loci. We use a quantitative measure of the magnitude of epistatic interactions between substitutions, and find that it is an increasing function of K. When haplotypes adapt at high mutation rates, more epistatic pairs of substitutions are observed on the line of descent than expected. The highest fitness is attained in landscapes with an intermediate amount of ruggedness that balance the higher fitness potential of interacting genes with their concomitant decreased evolvability. Our findings imply that the synergism between loci that interact epistatically is crucial for evolving genetic modules with high fitness, while too much ruggedness stalls the adaptive process.

Integrated information increases with fitness in the evolution of animats.

One of the hallmarks of biological organisms is their ability to integrate disparate information sources to optimize their behavior in complex environments. How this capability can be quantified and related to the functional complexity of an organism remains a challenging problem, in particular since organismal functional complexity is not well-defined. We present here several candidate measures that quantify information and integration, and study their dependence on fitness as an artificial agent ("animat") evolves over thousands of generations to solve a navigation task in a simple, simulated environment. We compare the ability of these measures to predict high fitness with more conventional information-theoretic processing measures. As the animat adapts by increasing its "fit" to the world, information integration and processing increase commensurately along the evolutionary line of descent. We suggest that the correlation of fitness with information integration and with processing measures implies that high fitness requires both information processing as well as integration, but that information integration may be a better measure when the task requires memory. A correlation of measures of information integration (but also information processing) and fitness strongly suggests that these measures reflect the functional complexity of the animat, and that such measures can be used to quantify functional complexity even in the absence of fitness data.

Critical dynamics in the evolution of stochastic strategies for the iterated prisoner's dilemma.

The observed cooperation on the level of genes, cells, tissues, and individuals has been the object of intense study by evolutionary biologists, mainly because cooperation often flourishes in biological systems in apparent contradiction to the selfish goal of survival inherent in Darwinian evolution. In order to resolve this paradox, evolutionary game theory has focused on the Prisoner's Dilemma (PD), which incorporates the essence of this conflict. Here, we encode strategies for the iterated Prisoner's Dilemma (IPD) in terms of conditional probabilities that represent the response of decision pathways given previous plays. We find that if these stochastic strategies are encoded as genes that undergo Darwinian evolution, the environmental conditions that the strategies are adapting to determine the fixed point of the evolutionary trajectory, which could be either cooperation or defection. A transition between cooperative and defective attractors occurs as a function of different parameters such as mutation rate, replacement rate, and memory, all of which affect a player's ability to predict an opponent's behavior. These results imply that in populations of players that can use previous decisions to plan future ones, cooperation depends critically on whether the players can rely on facing the same strategies that they have adapted to. Defection, on the other hand, is the optimal adaptive response in environments that change so quickly that the information gathered from previous plays cannot usefully be integrated for a response.

Comparative Transcriptomics Reveals Distinct Patterns of Gene Expression Conservation through Vertebrate Embryogenesis.

Despite life's diversity, studies of variation often remind us of our shared evolutionary past. Abundant genome sequencing and analyses of gene regulatory networks illustrate that genes and entire pathways are conserved, reused, and elaborated in the evolution of diversity. Predating these discoveries, 19th-century embryologists observed that though morphology at birth varies tremendously, certain stages of vertebrate embryogenesis appear remarkably similar across vertebrates. In the mid to late 20th century, anatomical variability of early and late-stage embryos and conservation of mid-stages embryos (the "phylotypic" stage) was named the hourglass model of diversification. This model has found mixed support in recent analyses comparing gene expression across species possibly owing to differences in species, embryonic stages, and gene sets compared. We compare 186 microarray and RNA-seq data sets covering embryogenesis in six vertebrate species. We use an unbiased clustering approach to group stages of embryogenesis by transcriptomic similarity and ask whether gene expression similarity of clustered embryonic stages deviates from a null expectation. We characterize expression conservation patterns of each gene at each evolutionary node after correcting for phylogenetic nonindependence. We find significant enrichment of genes exhibiting early conservation, hourglass, late conservation patterns in both microarray and RNA-seq data sets. Enrichment of genes showing patterned conservation through embryogenesis indicates diversification of embryogenesis may be temporally constrained. However, the circumstances under which each pattern emerges remain unknown and require both broad evolutionary sampling and systematic examination of embryogenesis across species.