André Grüning

Supervised learning in multilayer spiking neural networks

We introduce a supervised learning algorithm for multilayer spiking neural networks. The algorithm overcomes a limitation of existing learning algorithms: it can be applied to neurons firing multiple spikes in artificial neural networks with hidden layers. It can also, in principle, be used with any linearizable neuron model and allows different coding schemes of spike train patterns. The algorithm is applied successfully to classic linearly nonseparable benchmarks such as the XOR problem and the Iris data set, as well as to more complex classification and mapping problems. The algorithm has been successfully tested in the presence of noise, requires smaller networks than reservoir computing, and results in faster convergence than existing algorithms for similar tasks such as SpikeProp.

Sex‐specific demography and generalization of the Trivers–Willard theory

The Trivers–Willard theory proposes that the sex ratio of offspring should vary with maternal condition when it has sex‐specific influences on offspring fitness. In particular, mothers in good condition in polygynous and dimorphic species are predicted to produce an excess of sons, whereas mothers in poor condition should do the opposite. Despite the elegance of the theory, support for it has been limited. Here we extend and generalize the Trivers–Willard theory to explain the disparity between predictions and observations of offspring sex ratio. In polygynous species, males typically have higher mortality rates, different age‐specific reproductive schedules and more risk‐prone life history tactics than females; however, these differences are not currently incorporated into the Trivers–Willard theory. Using two‐sex models parameterized with data from free‐living mammal populations with contrasting levels of sex differences in demography, we demonstrate how sex differences in life history traits over the entire lifespan can lead to a wide range of sex allocation tactics, and show that correlations between maternal condition and offspring sex ratio alone are insufficient to conclude that mothers adaptively adjust offspring sex ratio.

Anodic microbial community diversity as a predictor of the power output of microbial fuel cells.

The relationship between the diversity of mixed-species microbial consortia and their electrogenic potential in the anodes of microbial fuel cells was examined using different diversity measures as predictors. Identical microbial fuel cells were sampled at multiple time-points. Biofilm and suspension communities were analysed by denaturing gradient gel electrophoresis to calculate the number and relative abundance of species. Shannon and Simpson indices and richness were examined for association with power using bivariate and multiple linear regression, with biofilm DNA as an additional variable. In simple bivariate regressions, the correlation of Shannon diversity of the biofilm and power is stronger (r=0.65, p=0.001) than between power and richness (r=0.39, p=0.076), or between power and the Simpson index (r=0.5, p=0.018). Using Shannon diversity and biofilm DNA as predictors of power, a regression model can be constructed (r=0.73, p<0.001). Ecological parameters such as the Shannon index are predictive of the electrogenic potential of microbial communities.

Segregation of the Anodic Microbial Communities in a Microbial Fuel Cell Cascade

Metabolic interactions within microbial communities are essential for the efficient degradation of complex organic compounds, and underpin natural phenomena driven by microorganisms, such as the recycling of carbon-, nitrogen-, and sulfur-containing molecules. These metabolic interactions ultimately determine the function, activity and stability of the community, and therefore their understanding would be essential to steer processes where microbial communities are involved. This is exploited in the design of microbial fuel cells (MFCs), bioelectrochemical devices that convert the chemical energy present in substrates into electrical energy through the metabolic activity of microorganisms, either single species or communities. In this work, we analyzed the evolution of the microbial community structure in a cascade of MFCs inoculated with an anaerobic microbial community and continuously fed with a complex medium. The analysis of the composition of the anodic communities revealed the establishment of different communities in the anodes of the hydraulically connected MFCs, with a decrease in the abundance of fermentative taxa and a concurrent increase in respiratory taxa along the cascade. The analysis of the metabolites in the anodic suspension showed a metabolic shift between the first and last MFC, confirming the segregation of the anodic communities. Those results suggest a metabolic interaction mechanism between the predominant fermentative bacteria at the first stages of the cascade and the anaerobic respiratory electrogenic population in the latter stages, which is reflected in the observed increase in power output. We show that our experimental system represents an ideal platform for optimization of processes where the degradation of complex substrates is involved, as well as a potential tool for the study of metabolic interactions in complex microbial communities.

Supervised Learning in Spiking Neural Networks for Precise Temporal Encoding

Precise spike timing as a means to encode information in neural networks is biologically supported, and is advantageous over frequency-based codes by processing input features on a much shorter time-scale. For these reasons, much recent attention has been focused on the development of supervised learning rules for spiking neural networks that utilise a temporal coding scheme. However, despite significant progress in this area, there still lack rules that have a theoretical basis, and yet can be considered biologically relevant. Here we examine the general conditions under which synaptic plasticity most effectively takes place to support the supervised learning of a precise temporal code. As part of our analysis we examine two spike-based learning methods: one of which relies on an instantaneous error signal to modify synaptic weights in a network (INST rule), and the other one relying on a filtered error signal for smoother synaptic weight modifications (FILT rule). We test the accuracy of the solutions provided by each rule with respect to their temporal encoding precision, and then measure the maximum number of input patterns they can learn to memorise using the precise timings of individual spikes as an indication of their storage capacity. Our results demonstrate the high performance of the FILT rule in most cases, underpinned by the rule’s error-filtering mechanism, which is predicted to provide smooth convergence towards a desired solution during learning. We also find the FILT rule to be most efficient at performing input pattern memorisations, and most noticeably when patterns are identified using spikes with sub-millisecond temporal precision. In comparison with existing work, we determine the performance of the FILT rule to be consistent with that of the highly efficient E-learning Chronotron rule, but with the distinct advantage that our FILT rule is also implementable as an online method for increased biological realism.

Improved spike-timed mappings using a tri-phasic spike timing-dependent plasticity rule

Reservoir computing and the liquid state machine model have received much attention in the literature in recent years. In this paper we investigate using a reservoir composed of a network of spiking neurons, with synaptic delays, whose synapses are allowed to evolve using a tri-phasic spike timing-dependent plasticity (STDP) rule. The networks are trained to produce specific spike trains in response to spatio-temporal input patterns. The results of using a tri-phasic STDP rule on the network properties are compared to those found using the more common exponential form of the rule. It is found that each rule causes the synaptic weights to evolve in significantly different fashions giving rise to different network dynamics. It is also found that the networks evolved with the tri-phasic rule are more capable of mapping input spatio-temporal patterns to the output spike trains.

The emergence of polychronous groups under varying input patterns, plasticity rules and network connectivities

Polychronous groups are unique temporal patterns of neural activity that exist implicitly within non-linear, recurrently connected networks. Through Hebbian based learning these groups can be strengthened to give rise to larger chains of spatiotemporal activity. Compared to other structures such as Synfire chains, they have demonstrated the potential of a much larger capacity for memory or computation within spiking neural networks. Polychronous groups are believed to relate to the input signals under which they emerge. Here we investigate the quantity of groups that emerge from increasing numbers of repeating input patterns, whilst also comparing the differences between two plasticity rules and two network connectivities. We find - perhaps counter-intuitively - that fewer groups are formed as the number of repeating input patterns increases. Furthermore, we find that a tri-phasic learning rule gives rise to fewer groups than the `classical double decaying exponential STDP plasticity window. It is also found that a scale-free network structure produces a similar quantity, but generally smaller groups than a randomly connected Erdös-Rényi structure.

Supervised Learning of Logical Operations in Layered Spiking Neural Networks with Spike Train Encoding

Few algorithms for supervised training of spiking neural networks exist that can deal with patterns of multiple spikes, and their computational properties are largely unexplored. We demonstrate in a set of simulations that the ReSuMe learning algorithm can successfully be applied to layered neural networks. Input and output patterns are encoded as spike trains of multiple precisely timed spikes, and the network learns to transform the input trains into target output trains. This is done by combining the ReSuMe learning algorithm with multiplicative scaling of the connections of downstream neurons. We show in particular that layered networks with one hidden layer can learn the basic logical operations, including Exclusive-Or, while networks without hidden layer cannot, mirroring an analogous result for layered networks of rate neurons. While supervised learning in spiking neural networks is not yet fit for technical purposes, exploring computational properties of spiking neural networks advances our understanding of how computations can be done with spike trains.

Low-Potential Respirators Support Electricity Production in Microbial Fuel Cells

In this paper, we analyse how electric power production in microbial fuel cells (MFCs) depends on the composition of the anodic biofilm in terms of metabolic capabilities of identified sets of species. MFCs are a promising technology for organic waste treatment and sustainable bioelectricity production. Inoculated with natural communities, they present a complex microbial ecosystem with syntrophic interactions between microbes with different metabolic capabilities. Our results demonstrate that low-potential anaerobic respirators—that is those that are able to use terminal electron acceptors with a low redox potential—are important for good power production. Our results also confirm that community metabolism in MFCs with natural inoculum and fermentable feedstock is a two-stage system with fermentation followed by anode respiration.

Stack-like and queue-like dynamics in recurrent neural networks

What dynamics do simple recurrent networks (SRNs) develop to represent stack-like and queue-like memories? SRNs have been widely used as models in cognitive science. However, they are interesting in their own right as non-symbolic computing devices from the viewpoints of analogue computing and dynamical systems theory. In this paper, SRNs are trained on two prototypical formal languages with recursive structures that need stack-like or queue-like memories for processing, respectively. The evolved dynamics are analysed, then interpreted in terms of simple dynamical systems, and the different ease with which SRNs aquire them is related to the properties of these simple dynamical systems. Within the dynamical systems framework, it is concluded that the stack-like language is simpler than the queue-like language, without making use of arguments from symbolic computation theory.