Evolving interpretable plasticity for spiking networks

Continuous adaptation allows survival in an ever-changing world. Adjustments in the synaptic coupling strength between neurons are essential for this capability, setting us apart from simpler, hard-wired organisms. How these changes can be mathematically described at the phenomenological level, as so-called ‘plasticity rules’, is essential both for understanding biological information processing and for developing cognitively performant artificial systems. We suggest an automated approach for discovering biophysically plausible plasticity rules based on the definition of task families, associated performance measures and biophysical constraints. By evolving compact symbolic expressions, we ensure the discovered plasticity rules are amenable to intuitive understanding, fundamental for successful communication and human-guided generalization. We successfully apply our approach to typical learning scenarios and discover previously unknown mechanisms for learning efficiently from rewards, recover efficient gradient-descent methods for learning from target signals, and uncover various functionally equivalent STDP-like rules with tuned homeostatic mechanisms.

Sensory perception in bacterial cyclic diguanylate signal transduction

Cyclic diguanylate (c-di-GMP) signal transduction systems provide bacteria the ability to sense changing cell status or environmental conditions and then execute suitable physiological and social behaviours in response. In this review, we provide a comprehensive census of the stimuli and receptors that are linked to modulation of intracellular c-di-GMP. Emerging evidence indicates that c-di-GMP networks sense light, surfaces, energy, redox potential, respiratory electron acceptors, temperature, and structurally diverse biotic and abiotic chemicals. Bioinformatic analysis of sensory domains in diguanylate cyclases and c-di-GMP-specific phosphodiesterases as well as the receptor complexes associated with them reveals that these functions are linked to a diverse repertoire of protein domain families. We describe the principles of stimulus perception learned from studying these modular sensory devices, illustrate how they are assembled in varied combinations with output domains, and summarize a system for classifying these sensor proteins based on their complexity. Biological information-processing via c-di-GMP signal transduction is not only fundamental to bacterial survival in dynamic environments, but also is being used to engineer gene expression circuitry and synthetic proteins with à la carte biochemical functionalities.

Deep Learning Networks and Visual Perception

Deep learning is an approach to artificial intelligence (AI) centered on the training of deep artificial neural networks to perform complex tasks. Since the early 21st century, this approach has led to record-breaking advances in AI, allowing computers to solve complex board games, video games, natural language-processing tasks, and vision problems. Neuroscientists and psychologists have also utilized these networks as models of biological information processing to understand language, motor control, cognition, audition, and—most commonly—vision. Specifically, early feedforward network architectures were inspired by visual neuroscience and are used to model neural activity and human behavior. They also provide useful representations of the perceptual space of images. The extent to which these models match data, however, depends on the methods used to characterize and compare them. The limitations of these feedforward neural networks to account for, for example, simple visual reasoning tasks, suggests that feedback mechanisms may be necessary to solve visual recognition tasks beyond image categorization.

ANDOR and beyond: dynamically switchable logic gates as modules for flexible information processing

Understanding how complex (bio-)chemical pathways and regulatory networks may be capable of processing information in efficient, flexible, and robust ways is a key question with implications touching fields across biology, systems biology, biochemistry, synthetic biology, dynamical systems theory, and network science. Considerable effort has been focused on the identification and characterization of structural motifs in these signaling networks, and companion efforts have instead sought to cast their operation as controlled by dynamical modules that appear out of dynamical correlations during information processing. While both these approaches have been successful in many examples of biological information processing, cases in which the signaling or regulatory network exhibits multi-functionality or context dependence remain problematic. We here propose a small set of higher-order effective modules that simultaneously incorporate both network structure and the attendant dynamical landscape. In so doing, we render effective computational units that can perform different logical operations based purely on the basin of attraction in which the network dynamics resides or is steered to. These dynamically switchable biochemical logic gates require fewer chemical components or gene products overall than their traditional analogs where static, separate gates are used for each desired function. We demonstrate the applicability and limits of these flexible gates by determining a robust range of parameters over which they correctly operate and further characterize the resilience of their function against intrinsic noise of the constituent reactions using the theory of large deviations. We also show the capability of this framework for general computations by designing a binary adder/subtractor circuit composed of only six components.

The structure dilemma in biological and artificial neural networks

Brain research up to date has revealed that structure and function are highly related. Thus, for example, studies have repeatedly shown that the brains of patients suffering from schizophrenia or other diseases have a different connectome compared to healthy people. Apart from stochastic processes, however, an inherent logic describing how neurons connect to each other has not yet been identified. We revisited this structural dilemma by comparing and analyzing artificial and biological-based neural networks. Namely, we used feed-forward and recurrent artificial neural networks as well as networks based on the structure of the micro-connectome of C. elegans and of the human macro-connectome. We trained these diverse networks, which markedly differ in their architecture, initialization and pruning technique, and we found remarkable parallels between biological-based and artificial neural networks, as we were additionally able to show that the dilemma is also present in artificial neural networks. Our findings show that structure contains all the information, but that this structure is not exclusive. Indeed, the same structure was able to solve completely different problems with only minimal adjustments. We particularly put interest on the influence of weights and the neuron offset value, as they show a different adaption behaviour. Our findings open up new questions in the fields of artificial and biological information processing research.

Plant behaviour in response to the environment: information processing in the solid state

Information processing and storage underpins many biological processes of vital importance to organism survival. Like animals, plants also acquire, store and process environmental information relevant to their fitness, and this is particularly evident in their decision-making. The control of plant organ growth and timing of their developmental transitions are carefully orchestrated by the collective action of many connected computing agents, the cells, in what could be addressed as distributed computation. Here, we discuss some examples of biological information processing in plants, with special interest in the connection to formal computational models drawn from theoretical frameworks. Research into biological processes with a computational perspective may yield new insights and provide a general framework for information processing across different substrates.This article is part of the theme issue ‘Liquid brains, solid brains: How distributed cognitive architectures process information’.

Phase separation provides a mechanism to reduce noise in cells

A central problem in cellular control is how cells cope with the inherent noise in gene expression. Although transcriptional and posttranscriptional feedback mechanisms can suppress noise, they are often slow, and cannot explain how cells buffer acute fluctuations. Here, by using a physical model that links fluctuations in protein concentration to the theory of phase separation, we show that liquid droplets can act as fast and effective buffers for gene expression noise. We confirm our theory experimentally using an engineered phase separating protein that forms liquid-like compartments in mammalian cells. These data suggest a novel role of phase separation in biological information processing.

Perceptual suboptimality: Bug or feature?

Rahnev & Denison (R&D) argue that whether people are “optimal” or “suboptimal” is not a well-posed question. We agree. However, we argue that the critical question is why humans make suboptimal perceptual decisions in the first place. We suggest that perceptual distortions have a normative explanation – that they promote efficient coding and computation in biological information processing systems.