We report an improvement to the conventional Echo State Network (ESN) across three benchmark chaotic time-series prediction tasks using fruit fly connectome data alone. We also investigate the impact of key connectome-derived structural features on prediction performance – uniquely bridging neurobiological structure and machine learning function; and find that both increasing the global average clustering coefficient and modifying the position of weights – by permuting their synapse-synapse partners – can lead to increased model variance and (in some cases) degraded performance. In all we consider four topological point modifications to a connectome-derived ESN reservoir (null model): namely, we alter the network sparsity, re-draw nonzero weights from a uniform distribution, permute nonzero weight positions, and increase the network global average clustering coefficient. We compare the four resulting ESN model classes – and the null model – with a conventional ESN by conducting time-series prediction experiments on size-variants of the Mackey-Glass 17 (MG-17), Lorenz, and Rossler chaotic time series; denoting each model's performance and variance across train-validate trials.

Multifunctionality describes the capacity for a neural network to perform multiple mutually exclusive tasks without altering its network connections; and is an emerging area of interest in the reservoir computing machine learning paradigm. Multifunctionality has been observed in the brains of humans and other animals: particularly, in the lateral horn of the fruit fly. In this work, we transplant the connectome of the fruit fly lateral horn to a reservoir computer (RC), and investigate the extent to which this ‘fruit fly RC’ (FFRC) exhibits multifunctionality using the ‘seeing double’ problem as a benchmark test. We furthermore explore the dynamics of how this FFRC achieves multifunctionality while varying the network's spectral radius. Compared to the widely-used Erdös-Renyi Reservoir Computer (ERRC), we report that the FFRC exhibits a greater capacity for multifunctionality; is multifunctional across a broader hyperparameter range; and solves the seeing double problem far beyond the previously observed spectral radius limit, wherein the ERRC's dynamics become chaotic.

Can connectome-derived constraints inform computation? In this paper we investigate the contribution of a fruit fly connectome's topology on the performance of an Echo State Network (ESN) - a subset of Reservoir Computing which is competitive in chaotic time series prediction. Specifically, we replace the reservoir layer of a classical ESN - normally a fixed, random graph represented as a 2-d matrix - with a particular (female) fruit fly connectome-derived connectivity matrix. We refer to this experimental class of models (with connectome-derived reservoirs) as “Fruit Fly ESNs” (FFESNs). We train and validate the FFESN on a chaotic time series prediction task; here we consider four sets of trials with different training input sizes (small, large) and train-validate splits (two variants). We compare the validation performance (Mean-Squared Error) of all of the best FFESN models to a class of control model ESNs (simply referred to as “ESNs”). Overall, for all four sets of trials we find that the FFESN either significantly outperforms (and has lower variance than) the ESN; or simply has lower variance than the ESN.

The fruit fly (Drosophila Melanogaster) is well-studied; the organism has served scientists for decades in all manner of biological research - most notably, perhaps, in genetics. However, much of the neuronal "middleware" of the fruit fly is unknown: for instance, how its neural architecture gives rise to functionalities such as odor categorization. Moreover, there is potential for the fruit fly neural network (FFNN) architecture in modelling Artificial Neural Networks (ANNs) - the former having been "crafted" over time by generations of evolutionary adaptation. In this work we hope to gain some insight with regards to both problem domains: firstly, with regards to understanding the "middleware" of the fruit fly neural network; secondly, with regards to constructing FFNN-derived ANNs. In particular, we recognize that there is a new opportunity to explore these problem domains in light of recent work on the EM (Electron Microscopy) "hemibrain" - the most comprehensive (to date) EM-derived digital reconstruction of the fruit fly brain, comprising 25,000 neurons (with labels for all neurons and synapses) [1]. Using the hemibrain, we look to the fruit fly olfactory system for the purposes of both exploring its neural architecture and creating an odor classifier. Our FFNN-derived ANN is - at present - fully-connected and uses the 800 most prevalent neurons in the fruit fly olfactory circuit (Antenna Lobe, Mushroom Body Calyx, and Lateral Horn [2]); it also has weight values assigned based on the number of synapses between neurons (an assumption made by the hemibrain authors [1]). Our initial dataset for odor classification includes 33 samples; each with 16 input components (individual resistance values from an array of 16 metal oxide sensors) and 4 output classes (air, ethanol, acetone, or mixed). We augment the dataset to size 33,033 with input noise based on a Gaussian normal distribution. Our current prototype yields greater-than-random test accuracy (37.5%) with 100 epochs of training.

Smart devices are becoming more universally adopted and can be used to track and model user activity and monitor for abnormalities. Deviations from what is expected may indicate that a fall is imminent or that an injury has been sustained. Healthcare practitioners can use descriptive models of human kinematics as a tool to monitor patient recovery. This work extends previous work which generated descriptive nonlinear symbolic models of human kinematics with genetic programming. Previously, linear models were developed and compared to the nonlinear models. Although the linear models fit the data well, they were significantly worse than the nonlinear models. In this phase of the project, ensembles of nonlinear models were created to more accurately fit and classify data. Different model selection strategies for the ensembles were investigated. As one would expect, ensembles of models were significantly better than a single model classifier. It was also observed that, although more models in the ensemble yielded better results, only 2 models were required to obtain significantly better results. It was also observed that a random model selection strategy for the ensembles produced competitive results when compared to a more rigorous model selection strategy.

The brain is a nonlinear computational system; however, most methods employed in finding functional connectivity models with functional magnetic resonance imaging (fMRI) data produce strictly linear models - models incapable of truly describing the underlying system.Genetic programming is used to develop nonlinear models of functional connectivity from fMRI data. The study builds on previous work and observes that nonlinear models contain relationships not found by traditional linear methods. When compared to linear models, the nonlinear models contained fewer regions of interest and were never significantly worse when applied to data the models were fit to. Nonlinear models could generalize to unseen data from the same subject better than traditional linear models (intrasubject). Nonlinear models could not generalize to unseen data recorded from other subjects (intersubject) as well as the linear models, and reasons for this are discussed. This study presents the problem that many, manifestly different models in both operators and features, can effectively describe the system with acceptable metrics.

Smart devices are becoming an extension of ourselves that contain sensitive information and are often targeted for theft. The development of an intelligent and reliable means of user identification and authentication is critical. Not only can the development of user models performing tasks be used for user and task identification, but systems can also notify individuals if there is a potential health concern. The construction of an idealized model of human locomotion may give medical care providers a better understanding of individual differences and guide therapy and treatment. Data was gathered from a smartwatch worn by six subjects performing five different tasks and Genetic Programming was used to perform symbolic regression - a model free, nonlinear type of regression analysis. Symbolic regression was applied to smartwatch data and a collection of nonlinear closed form symbolic mathematical models were generated. Not only did these models fit the data well, but they provided insight into the underlying system. With only 5 seconds of unseen data, the models could classify which subjects were performing which task with 83.9% accuracy when chance was only 3.33%.

In contrast to conventional, univariate analysis, various types of multivariate analysis have been applied to functional magnetic resonance imaging (fMRI) data. In this paper, we compare two contemporary approaches for multivariate regression on task-based fMRI data: linear regression with ridge regularization and non-linear symbolic regression using genetic programming. The data for this project is representative of a contemporary fMRI experimental design for visual stimuli. Linear and non-linear models were generated for 10 subjects, with another 4 withheld for validation. Model quality is evaluated by comparing R scores (Pearson product-moment correlation) in various contexts, including single run self-fit, within-subject generalization, and between-subject generalization. Propensity for modelling strategies to overfit is estimated using a separate resting state scan. Results suggest that neither method is objectively or inherently better than the other.

Patients who suffered a traumatic brain injury (TBI) require special care, and physicians often monitor intercranial pressure (ICP) as it can greatly aid in management. Although monitoring ICP can be critical, it requires neurosurgery, which presents additional significant risk. Monitoring ICP also aids in clinical situations beyond TBI, however the risk of neurosurgery can prevent physicians from gathering the data. The need for surgery may be eliminated if ICP could be accurately inferred using noninvasive physiological measures. Genetic programming (GP) and linear regression were used to develop nonlinear and linear mathematical models describing the relationships between intercranial pressure and a collection of physiological measurements from noninvasive instruments. Nonlinear models of ICP were generated that not only fit the subjects they were trained on, but generalized well across other subjects. The nonlinear models were analysed and provided insight into the studied underlying system which led to the creation of additional models. The new models were developed with a refined search, and were more accurate and general. It was also found that the relations between the features could be explained effectively with a simple linear model after GP refined the search.

Despite significant effort, there is currently no formal or de facto standard framework or format for constructing, representing, or manipulating general neural networks. In computational neuroscience, there have been some attempts to formalize connectionist notations and generative operations for neural networks, including Connection Set Algebra, but none are truly formal or general. In computational intelligence (CI), though the use of linear algebra and tensor-based models are widespread, graph-based frameworks are also popular and there is a lack of tools supporting the transfer of information between systems. To address these gaps, we exploited existing results about the connection between linear and relation algebras to define a concise, formal algebraic framework that generalizes graph and tensor-based neural networks. For simplicity and compatibility, this framework is purposefully defined as a minimal extension to linear algebra. We demonstrate the merits of this approach first by defining new operations for network composition along with proofs of their most important properties. An implementation of the algebraic framework is presented and applied to create an instance of an artificial neural network that is compatible with both graph and tensor based CI frameworks. The result is an algebraic framework for neural networks that generalizes the formats used in at least two systems, together with an example implementation.

The vast majority of methods employed in the analysis of functional Magnetic Resonance Imaging (fMRI) produce exclusively linear models; however, it is clear that linear models cannot fully describe a system with the observed behavioral complexity of the human brain --- an intrinsically nonlinear system. By using tools embracing the possibility of modeling the underlying nonlinear system we may uncover meaningful undiscovered relationships which further our understanding of the brain.

SUMMARY: Copy number variants (CNVs) are a major source of genetic variation. Comparing CNVs between samples is important in elucidating their potential effects in a wide variety of biological contexts. HD-CNV (hotspot detector for copy number variants) is a tool for downstream analysis of previously identified CNV regions from multiple samples, and it detects recurrent regions by finding cliques in an interval graph generated from the input. It creates a unique graphical representation of the data, as well as summary spreadsheets and UCSC (University of California, Santa Cruz) Genome Browser track files. The interval graph, when viewed with other software or by automated graph analysis, is useful in identifying genomic regions of interest for further study. AVAILABILITY AND IMPLEMENTATION: HD-CNV is an open source Java code and is freely available, with tutorials and sample data from http://daleylab.org. CONTACT: [email protected]

This paper presents the design of a new web browser, the Tree Trailblazer, which allows users to browse the web while maintaining a visual record of their exploration path, or trail, through the information space. This design enhances the backtracking aspects of web browsing over current designs by providing visual cues regarding the pages related to the page being viewed, providing users with an understanding of their position in the trail. This design also helps users blaze new trails off a page by allowing them to open previews of pages off of the currently viewed page. The scenario based design process that was used to construct the browser is discussed in conjunction with the initial prototype implementation. A formative user evaluation of this prototype showed this browser design to be very easy to learn and highly usable, with particular attention being paid to aspects of the tree visualization.