Search

Natural odorant scenes are complex landscapes comprising mixtures of volatile compounds. It was previously proposed that the Antennal Lobe circuit recovers the odorant identity in a concentration-invariant manner via divisive normalization of Local Neurons. It remains unclear, however, how identities of odorant components in a mixture is represented or recovered in the fruit fly early olfactory pathway. In the current work, we take a different approach from the traditional steady-state analyses that classify odorant mixture encoding into configural vs. elemental schemes. Instead, we focus on the spatio-temporal responses of the early olfactory pathway at the levels of the Antennal Lobe and the Mushroom Body, and formulate the odorant demixing problem as a blind source separation problem - where the identities of each individual odorant component and their corresponding concentration waveforms are recovered from the spatio-temporal PSTH of Olfactory Sensory Neurons (OSNs), Projection Neurons (PNs), and Kenyon Cells (KCs) respectively. Building upon previous models of the Antenna and the Antennal Lobe, we advanced a feedback divisive normalization architecture of the Mushroom Body Calyx circuit comprised of PN, KC and the giant Anterior Paired Lateral (APL) neuron. We demonstrate that the PN-KC-APL circuit produces a high dimensional representation of odorant mixture with robust sparsity, and results in greater odorant demixing performance than the PN responses.

Divisive normalization is a canonical neural computation employed for sensory adaptation in vision, olfaction and attention modulation. While Divisive Normalization has been proposed to be an efficient coding algorithm, it remains unclear whether such transformation results in information loss in a dynamic setting. Leveraging a previously proposed general mathematical framework called the Divisive Normalization Processor (DNP), we first show that the DNP circuit describes a wide class of neural circuits as well as phenomenological models including the Linear-Nonlinear cascade model. We then demonstrate both theoretically and computationally that the DNP is an invertible operator that could faithfully represent input information given sufficient output samples.

Recent progress on connectomics has resulted in huge datasets of brain structures in single-synapse scale. This calls for the modeling of executable circuits to discover the functional logic of these neural circuits in different scales. To provide an approach to standardize analyses of neuropils with multiple-input-multi-output channels, we put forward a programmable model focusing on the antennal lobe (AL), a circuit at the periphery of the olfactory system with nonlinear computation. We present an approach for constructing and analyzing antennal lobe circuits using models of glomeruli and local feedback loops. We establish the composability of the connectivity of glomeruli with local neuron feedback loops by combining pairs of glomeruli, and then by combining all glomeruli together to characterize the I/O of the entire AL. We thus provide a methodology for the quantitative characterization of the I/O of the AL as a function of feedback loop motifs.

We advance a model of the Mushroom Body of the Drosophila brain based on the recently published fruit fly brain connectome. We quantify the effect of the connectivity from antennal lobe projection neurons to Kenyon Cells (KCs) and the effect of feedback between KCs and the anterior paired lateral (APL) neuron on the representation of odorants at the KC level. We then characterize odorant representation in the mushroom body output neurons (MBONs) as a function of semantics represented by the dopaminergic neurons (DANs). Finally, we evaluate the performance of associative learning in the MB that endows pure odorants with DAN semantics.

The circadian clock orchestrates daily changes in physiology and behavior to ensure internal temporal order and optimal timing across the day. In animals, a central brain clock coordinates circadian rhythms throughout the body and is characterized by a remarkable robustness that depends on synaptic connections between constituent neurons. The clock neuron network of Drosophila, which shares network motifs with clock networks in the mammalian brain yet is built of many fewer neurons, offers a powerful model for understanding the network properties of circadian timekeeping. Here, we report an assessment of synaptic connectivity within a clock network, focusing on the critical lateral neuron (LN) clock neuron classes within the Janelia hemibrain dataset. Our results reveal that previously identified anatomical and functional subclasses of LNs represent distinct connectomic types. Moreover, we identify a small number of non-clock cell subtypes representing highly synaptically coupled nodes within the clock neuron network. This suggests that neurons lacking molecular timekeeping likely play integral roles within the circadian timekeeping network. To our knowledge, this represents the first comprehensive connectomic analysis of a circadian neuronal network.

Recent advances in molecular transduction of odorants in the Olfactory Sensory Neurons (OSNs) of the Drosophila Antenna have shown that the odorant object identity is multiplicatively coupled with the odorant concentration waveform. The resulting combinatorial neural code is a confounding representation of odorant semantic information (identity) and syntactic information (concentration). To distill the functional logic of odor information processing in the Antennal Lobe (AL) a number of challenges need to be addressed including 1) how is the odorant semantic information decoupled from the syntactic information at the level of the AL, 2) how are these two information streams processed by the diverse AL Local Neurons (LNs) and 3) what is the end-to-end functional logic of the AL? By analyzing single-channel physiology recordings at the output of the AL, we found that the Projection Neuron responses can be decomposed into a concentration-invariant component, and two transient components boosting the positive/negative concentration contrast that indicate onset/offset timing information of the odorant object. We hypothesized that the concentration-invariant component, in the multi-channel context, is the recovered odorant identity vector presented between onset/offset timing events. We developed a model of LN pathways in the Antennal Lobe termed the differential Divisive Normalization Processors (DNPs), which robustly extract the semantics (the identity of the odorant object) and the ON/OFF semantic timing events indicating the presence/absence of an odorant object. For real-time processing with spiking PN models, we showed that the phase-space of the biological spike generator of the PN offers an intuit perspective for the representation of recovered odorant semantics and examined the dynamics induced by the odorant semantic timing events. Finally, we provided theoretical and computational evidence for the functional logic of the AL as a robust ON-OFF odorant object identity recovery processor across odorant identities, concentration amplitudes and waveform profiles.

The Drosophila brain has only a fraction of the number of neurons of higher organisms such as mice and humans. Yet the sheer complexity of its neural circuits recently revealed by large connectomics datasets suggests that computationally modeling the function of fruit fly brain circuits at this scale posits significant challenges.To address these challenges, we present here a programmable ontology that expands the scope of the current Drosophila brain anatomy ontologies to encompass the functional logic of the fly brain. The programmable ontology provides a language not only for modeling circuit motifs but also for programmatically exploring their functional logic. To achieve this goal, we tightly integrated the programmable ontology with the workflow of the interactive FlyBrainLab computing platform. As part of the programmable ontology, we developed NeuroNLP++, a web application that supports free-form English queries for constructing functional brain circuits fully anchored on the available connectome/synaptome datasets, and the published worldwide literature.In addition, we present a methodology for including a model of the space of odorants into the programmable ontology, and for modeling olfactory sensory circuits of the antenna of the fruit fly brain that detect odorant sources. Furthermore, we describe a methodology for modeling the functional logic of the antennal lobe circuit consisting of massive local feedback loops, a characteristic feature observed across Drosophila brain regions. Finally, using a circuit library, we demonstrate the power of our methodology for interactively exploring the functional logic of the massive number of feedback loops in the antennal lobe.

The early olfactory system of the fruit fly, while sensing a complex odorant landscape, encodes the odorant object identity (semantic information) and the odorant concentration waveform (syntactic information) into a combinatorial neural code. Single-channel physiology recordings at the output of the Antenna Lobe (AL) exhibit concentration-invariance and contrast-boosting properties, indicating a decoupling of the odorant object identity from the concentration waveform in steady-state while responding strongly to odorant concentration onset and offset in transient states.Through exhaustive computational explorations of the AL circuits, we show that the steady-state and transient response features of the AL are, respectively, due to presynaptic and postsynaptic Local Neurons (LNs). Theoretical analysis reveals that the LN pathways can be modeled as parallel differential Divisive Normalization Processors (DNPs). Differential DNPs robustly extract odorant identity (semantic information) and ON/OFF odorant event-timing (syntactic information), thereby providing for the AL the functional logic of ON-OFF odorant identity recovery.

In mobile and edge devices, always-on keyword spotting (KWS) is an essential function to detect wake-up words. Recent works achieved extremely low power dissipation down to $\sim500$ nW [1]. However, most of them adopt noise-dependent training, i.e. training for a specific signal-to-noise ratio (SNR) and noise type [1], and therefore their accuracies degrade for different SNR levels and noise types that are not targeted in the training (Fig. 9.9.1, top left). To improve robustness, so-called noise-independent training can be considered, which is to use the training data that includes all the possible SNR levels and noise types [2]. But, this approach is challenging for an ultra-low-power device since it demands a large neural network to learn all the possible features. A neural network of a fixed size has its own memory capacity limit and reaches a plateau in accuracy if it has to learn more than its limit (Fig. 9.9.1, top right). On the other hand, it is known that biological acoustic systems employ a simpler process, called divisive energy normalization (DN), to maintain accuracy even in varying noise conditions [3]. In this work, therefore, by adopting such a DN, we prototype a normalized acoustic feature extractor chip (NAFE) in 65nm. The NAFE can take an acoustic signal from a microphone and produce spike-rate coded features. We pair NAFE with a spiking neural network (SNN) classifier chip [4], creating the end-to-end KWS system. The proposed system achieves 89-to-94% accuracy across -5 to 20dB SNRs and four different noise types on HeySnips [5], while the baseline without DN achieves a much lower accuracy of 71-87%. NAFE consumes up to 109nW and the KWS system 570nW.

The term “SDN” represents a significant evolution in networking technology and draws attention and support from network operators, vendors, researchers and industry regulators. Contrary to popular belief that SDN is a recent invention, or is related only to IP technologies, the concept of network programmability and control/data plane separation has its roots in the 1960s and 1970s when the telephone network started its transition to digital. Many iterations followed, leading to applications in packet networks and today’s Internet. In this article, we review a number of early works on network programmability that illustrate how several features of SDN emerged progressively over several decades. Our review starts from the early concepts of network control in the telephone network and continues to examine a prolific period of research advancements in the 1990s and early 2000s that led to a number of startup companies that followed, IEEE’s own efforts in standardizing network programmability, and finally the arrival of the OpenFlow standard. We study the importance of this architectural transformation and its influences on modern cloud computing and next-generation networking.

In recent years, a wealth of Drosophila neuroscience data have become available including cell type and connectome/synaptome datasets for both the larva and adult fly. To facilitate integration across data modalities and to accelerate the understanding of the functional logic of the fruit fly brain, we have developed FlyBrainLab, a unique open-source computing platform that integrates 3D exploration and visualization of diverse datasets with interactive exploration of the functional logic of modeled executable brain circuits. FlyBrainLab’s User Interface, Utilities Libraries and Circuit Libraries bring together neuroanatomical, neurogenetic and electrophysiological datasets with computational models of different researchers for validation and comparison within the same platform. Seeking to transcend the limitations of the connectome/synaptome, FlyBrainLab also provides libraries for molecular transduction arising in sensory coding in vision/olfaction. Together with sensory neuron activity data, these libraries serve as entry points for the exploration, analysis, comparison, and evaluation of circuit functions of the fruit fly brain.

In recent years, a wealth of Drosophila neuroscience data have become available. These include cell type, connectome and synaptome datasets for both the larva and adult fly. To facilitate integration across data modalities and to accelerate the understanding of the functional logic of the fly brain, we developed an interactive computing environment called FlyBrainLab.FlyBrainLab is uniquely positioned to accelerate the discovery of the functional logic of the Drosophila brain. Its interactive open source architecture seamlessly integrates and brings together computational models with neuroanatomical, neurogenetic and electrophysiological data, changing the organization of neuroscientific fly brain data from a group of seemingly disparate databases, arrays and tables, to a well structured data and executable circuit repository.The FlyBrainLab User Interface supports a highly intuitive and automated work-flow that streamlines the 3D exploration and visualization of fly brain circuits, and the interactive exploration of the functional logic of executable circuits created directly from the explored and visualized fly brain data. Furthermore, efficient comparisons of circuit models are supported, across models developed by different researchers, across different developmental stages of the fruit fly and across different datasets.The FlyBrainLab Utility Libraries help untangle the graph structure of neural circuits from raw connectome and synaptome data. The Circuit Libraries facilitate the exploration of neural circuits of the neuropils of the central complex and, the development and implementation of models of the adult and larva fruit fly early olfactory systems.Seeking to transcend the limitations of the connectome, FlyBrainLab provides additional libraries for molecular transduction arising in sensory coding in vision and olfaction. Together with sensory neuron activity data, these libraries serve as entry points for discovering circuit function in the sensory systems of the fruit fly brain. They also enable the biological validation of developed executable circuits within the same platform.

Over the past two decades, substantial amount of work has been conducted to characterize different odorant receptors, neuroanatomy and odorant response properties of the early olfactory system of Drosophila melanogaster. Yet many odorant receptors remain only partially characterized, and the odorant transduction process and the axon hillock spiking mechanism of the olfactory sensory neurons (OSNs) have yet to be fully determined. Identity and concentration, two key characteristics of the space of odorants, are encoded by the odorant transduction process. Detailed molecular models of the odorant transduction process are, however, scarce for fruit flies. To address these challenges we advance a comprehensive model of fruit fly OSNs as a cascade consisting of an odorant transduction process (OTP) and a biophysical spike generator (BSG). We model odorant identity and concentration using an odorant-receptor binding rate tensor, modulated by the odorant concentration profile, and an odorant-receptor dissociation rate tensor, and quantitatively describe the mechanics of the molecular ligand binding/dissociation of the OTP. We model the BSG as a Connor-Stevens point neuron. The resulting spatio-temporal encoding model of the Drosophila antenna provides a theoretical foundation for understanding the neural code of both odorant identity and odorant concentration and advances the state-of-the-art in a number of ways. First, it quantifies on the molecular level the spatio-temporal level of complexity of the transformation taking place in the antennae. The concentration-dependent spatio-temporal code at the output of the antenna circuits determines the level of complexity of olfactory processing in the downstream neuropils, such as odorant recognition and olfactory associative learning. Second, the model is biologically validated using multiple electrophysiological recordings. Third, the model demonstrates that the currently available data for odorant-receptor responses only enable the estimation of the affinity of the odorant-receptor pairs. The odorant-dissociation rate is only available for a few odorant-receptor pairs. Finally, our model calls for new experiments for massively identifying the odorant-receptor dissociation rates of relevance to flies., Author summary Identity and concentration, intrinsically embedded in the odorant space, are two key characteristics of olfactory coding that define the level of complexity of neural processing throughout the olfactory system in the fruit fly. In this paper we advance a theoretical foundation for understanding these two characteristics by quantifying mathematically the odorant space and devising a biophysical model of the olfactory sensory neurons (OSNs). To validate our modeling approach, we propose and apply an algorithm to estimate the affinity value and the dissociation rate, the two characteristics that define odorant identity, of multiple odorant-receptor pairs. We then evaluate our model with a multitude of odorant waveforms and demonstrate that the model output reproduces the temporal responses of OSNs obtained from in vivo electrophysiology recordings. Furthermore, we evaluate the model at the OSN population level and quantify on the molecular level the spatio-temporal level of complexity of the transformation taking place between the odorant space and the OSNs. The resulting concentration-dependent spatio-temporal code determines the level of complexity of the input space driving olfactory processing in the downstream neuropils. Lastly, our model demonstrates that the currently available data for OSN responses only enables estimation of affinity value. This calls for new experiments for massively identifying the odorant-receptor dissociation rates of relevance to flies.

Functional identification is a key methodology for uncovering the logic of neuroinformation processing of brain circuits. For neural circuits modeling sensory systems as dynamical systems, the complexity of the identification algorithm largely depends on the number of stimuli used. Neurons in these circuits consist of dendritic stimulus processors modeling the signal processing in the dendritric tree and biological spike generators modeling the spiking mechanism at the axon hillock. Here, we review the identification of multi-sensory spatio-temporal dendritic stimulus processors that arise in the encoding of auditory scenes, color visual fields, and the mixing of auditory scenes and natural visual fields. We demonstrate the fundamental duality between the identification of the dendritic stimulus processor of a single neuron and the decoding of stimuli encoded by a population of neurons with a bank of dendritic stimulus processors. The duality enables us to reconstruct the originally encoded stimuli from all the generated spikes by using the identified neural circuit. The reconstruction leads to a simple and intuitive evaluation of the identified dendritic stimulus processors in the space of stimuli.

The fruit fly’s natural visual environment is often characterized by light intensities ranging across several orders of magnitude and by rapidly varying contrast across space and time. Fruit fly photoreceptors robustly transduce and, in conjunction with amacrine cells, process visual scenes and provide the resulting signal to downstream targets. Here, we model the first step of visual processing in the photoreceptor-amacrine cell layer. We propose a novel divisive normalization processor (DNP) for modeling the computation taking place in the photoreceptor-amacrine cell layer. The DNP explicitly models the photoreceptor feedforward and temporal feedback processing paths and the spatio-temporal feedback path of the amacrine cells. We then formally characterize the contrast gain control of the DNP and provide sparse identification algorithms that can efficiently identify each the feedforward and feedback DNP components. The algorithms presented here are the first demonstration of tractable and robust identification of the components of a divisive normalization processor. The sparse identification algorithms can be readily employed in experimental settings, and their effectiveness is demonstrated with several examples.

Computing technology has been a backbone of our society. Its importance is hard to overemphasize. Today, we again confirm its extreme importance with recent advances in deep neural networks. Those emerging workloads impose an unprecedented amount of arithmetic complexity and data access beyond our existing computing systems can barely handle. Particularly, mobile and embedded computing systems will face a major challenge in achieving energy-efficient computing for truly enabling intelligent systems. In this talk, we will discuss the emerging analog and mixed-signal circuit techniques to improve energy efficiency. We will discuss two recent cases using such techniques, one on the speech recognition processor in hybrid analog and digital circuits and the other on the embedded SRAM circuits that support analog-mixed-signal in-memory (in-bitcell) computing for convolutional and deep neural networks.

The Fruit Fly Brain Observatory (FFBO) is a collaborative effort between experimentalists, theorists and computational neuroscientists at Columbia University, National Tsing Hua University and Sheffield University with the goal to (i) create an open platform for the emulation and biological validation of fruit fly brain models in health and disease, (ii) standardize tools and methods for graphical rendering, representation and manipulation of brain circuits, (iii) standardize tools for representation of fruit fly brain data and its abstractions and support for natural language queries, (iv) create a focus for the neuroscience community with interests in the fruit fly brain and encourage the sharing of fruit fly brain structural data and executable code worldwide. NeuroNLP and NeuroGFX, two key FFBO applications, aim to address two major challenges, respectively: i) seamlessly integrate structural and genetic data from multiple sources that can be intuitively queried, effectively visualized and extensively manipulated, ii) devise executable brain circuit models anchored in structural data for understanding and developing novel hypotheses about brain function. NeuroNLP enables researchers to use plain English (or other languages) to probe biological data that are integrated into a novel database system, called NeuroArch, that we developed for integrating biological and abstract data models of the fruit fly brain. With powerful 3D graphical visualization, NeuroNLP presents a highly accessible portal for the fruit fly brain data. NeuroGFX provides users highly intuitive tools to execute neural circuit models with Neurokernel, an open-source platform for emulating the fruit fly brain, with full data support from the NeuroArch database and visualization support from an interactive graphical interface. Brain circuits can be configured with high flexibility and investigated on multiple levels, e.g., whole brain, neuropil, and local circuit levels. The FFBO is publicly available and accessible at http://fruitflybrain.org from any modern web browsers, including those running on smartphones.

Claude Shannon, born on April 30, 1916, pioneered the mathematical theory of communication in his 1948 paper in the Bell System Technical Journal . Information theory has since provided the foundation for the digital revolution in communications technology. In addition, it has provided a powerful framework for investigating the fundamental limitations of naturally occurring communications, particularly in biological systems. This two-part special issue of the IEEE Transactions on Molecular, Biological, and Multi-Scale Communications honors Shannon’s centennial by highlighting recent progress in biological and bio-inspired information theory. Part I includes papers on applications of multivariate mutual information to data clustering, statistical thermodynamics of nonequilibrium chemical processes, information processing in neuronal systems, biochemical signal transduction systems, DNA sequencing, and the swarming of insects.

We investigate the sparse functional identification of complex cells and the decoding of spatio-temporal visual stimuli encoded by an ensemble of complex cells. The reconstruction algorithm is formulated as a rank minimization problem that significantly reduces the number of sampling measurements (spikes) required for decoding. We also establish the duality between sparse decoding and functional identification and provide algorithms for identification of low-rank dendritic stimulus processors. The duality enables us to efficiently evaluate our functional identification algorithms by reconstructing novel stimuli in the input space. Finally, we demonstrate that our identification algorithms substantially outperform the generalized quadratic model, the nonlinear input model, and the widely used spike-triggered covariance algorithm. Electronic Supplementary Material The online version of this article (10.1186/s13408-017-0057-1) contains supplementary material.

SummaryRecently, multiple focused efforts have resulted in substantial increase in the availability of connectome data in the fruit fly brain. Elucidating neural circuit function from such structural data calls for a scalable computational modeling methodology. We propose such a methodology that includes i) a brain emulation engine, with an architecture that can tackle the complexity of whole brain modeling, ii) a database that supports tight integration of biological and modeling data along with support for domain specific queries and circuit transformations, and iii) a graphical interface that allows for total flexibility in configuring neural circuits and visualizing run-time results, both anchored on model abstractions closely reflecting biological structure. Towards the realization of such a methodology, we have developed NeuroGFX and integrated it into the architecture of the Fruit Fly Brain Observatory (http://fruitflybrain.org). The computational infrastructure in NeuroGFX is provided by Neurokernel, an open source platform for the emulation of the fruit fly brain, and NeuroArch, a database for querying and executing fruit fly brain circuits. The integration of the two enables the algorithmic construction/manipulation/revision of executable circuits on multiple levels of abstraction of the same model organism. The power of this computational infrastructure can be leveraged through an intuitive graphical interface that allows visualizing execution results in the context of biological structure. This provides an environment where computational researchers can present configurable, executable neural circuits, and experimental scientists can easily explore circuit structure and function ultimately leading to biological validation. With these capabilities, NeuroGFX enables the exploration of function from circuit structure at whole brain, neuropil, and local circuit level of abstraction. By allowing for independently developed models to be integrated at the architectural level, NeuroGFX provides an open plug and play, collaborative environment for whole brain computational modeling of the fruit fly.

SummaryNeuroNLP, is a key application on the Fruit Fly Brain Observatory platform (FFBO, http://fruitflybrain.org), that provides a modern web-based portal for navigating fruit fly brain circuit data. Increases in the availability and scale of fruit fly connectome data, demand new, scalable and accessible methods to facilitate investigation into the functions of the latest complex circuits being uncovered. NeuroNLP enables in-depth exploration and investigation of the structure of brain circuits, using intuitive natural language queries that are capable of revealing the latent structure and information, obscured due to expansive yet independent data sources. NeuroNLP is built on top of a database system call NeuroArch that codifies knowledge about the fruit fly brain circuits, spanning multiple sources. Users can probe biological circuits in the NeuroArch database with plain English queries, such as “show glutamatergic local neurons in the left antennal lobe” and “show neurons with dendrites in the left mushroom body and axons in the fan-shaped body”. This simple yet powerful interface replaces the usual, cumbersome checkboxes and dropdown menus prevalent in today’s neurobiological databases. Equipped with powerful 3D visualization, NeuroNLP standardizes tools and methods for graphical rendering, representation, and manipulation of brain circuits, while integrating with existing databases such as the FlyCircuit. The userfriendly graphical user interface complements the natural language queries with additional controls for exploring the connectivity of neurons and neural circuits. Designed with an open-source, modular structure, it is highly scalable/flexible/extensible to additional databases or to switch between databases and supports the creation of additional parsers for other languages. By supporting access through a web browser from any modern laptop or smartphone, NeuroNLP significantly increases the accessibility of fruit fly brain data and improves the impact of the data in both scientific and educational exploration.

SummaryThe Fruit Fly Brain Observatory (FFBO) is a collaborative effort between experimentalists, theorists and computational neuroscientists at Columbia University, National Tsing Hua University and Sheffield University with the goal to (i) create an open platform for the emulation and biological validation of fruit fly brain models in health and disease, (ii) standardize tools and methods for graphical rendering, representation and manipulation of brain circuits, (iii) standardize tools for representation of fruit fly brain data and its abstractions and support for natural language queries, (iv) create a focus for the neuroscience community with interests in the fruit fly brain and encourage the sharing of fruit fly brain structural data and executable code worldwide. NeuroNLP and NeuroGFX, two key FFBO applications, aim to address two major challenges, respectively: i) seamlessly integrate structural and genetic data from multiple sources that can be intuitively queried, effectively visualized and extensively manipulated, ii) devise executable brain circuit models anchored in structural data for understanding and developing novel hypotheses about brain function. NeuroNLP enables researchers to use plain English (or other languages) to probe biological data that are integrated into a novel database system, called NeuroArch, that we developed for integrating biological and abstract data models of the fruit fly brain. With powerful 3D graphical visualization, NeuroNLP presents a highly accessible portal for the fruit fly brain data. NeuroGFX provides users highly intuitive tools to execute neural circuit models with Neurokernel, an open-source platform for emulating the fruit fly brain, with full data support from the NeuroArch database and visualization support from an interactive graphical interface. Brain circuits can be configured with high flexibility and investigated on multiple levels, e.g., whole brain, neuropil, and local circuit levels. The FFBO is publicly available and accessible at http://fruitflybrain.org from any modern web browsers, including those running on smartphones.

During the research period we: (i) devised path breaking algorithms for the functional identification and evaluation of non-linear dendritic processing, and (ii) released a groundbreaking open source platform for the isolated and integrated emulation of the fruit fly brain on multiple GPUs (Neurokernel). We established that identifying a single dendritic stimulus processor (DSP) is mathematically dual to decoding of stimuli encoded by a population of neurons with a bank of DSPs. Building on this key duality property, we (i) demonstrated that the evaluation of the functional identification methodology can be effectively and intuitively performed in the stimulus space, and (ii) characterized the effect of noise parameters on the precision of the functional identification of feed forward, feedback and cross-feedback neural circuits with DSP/biological spike generator neuron models. We demonstrated the power of Neurokernel's model integration by combining independently developed models of the retina and lamina neuropils in the fly's visual system and by demonstrating their neuroinformation processing capability.

We have developed an open software platform called Neurokernel for collaborative development of comprehensive models of the brain of the fruit fly Drosophila melanogaster and their execution and testing on multiple Graphics Processing Units (GPUs). Neurokernel provides a programming model that capitalizes upon the structural organization of the fly brain into a fixed number of functional modules to distinguish between these modules' local information processing capabilities and the connectivity patterns that link them. By defining mandatory communication interfaces that specify how data is transmitted between models of each of these modules regardless of their internal design, Neurokernel explicitly enables multiple researchers to collaboratively model the fly's entire brain by integration of their independently developed models of its constituent processing units. We demonstrate the power of Neurokernel's model integration by combining independently developed models of the retina and lamina neuropils in the fly's visual system and by demonstrating their neuroinformation processing capability. We also illustrate Neurokernel's ability to take advantage of direct GPU-to-GPU data transfers with benchmarks that demonstrate scaling of Neurokernel's communication performance both over the number of interface ports exposed by an emulation's constituent modules and the total number of modules comprised by an emulation.

Previous research demonstrated thatglobalphase alone can be used to faithfully represent visual scenes. Here we provide a reconstruction algorithm by using onlylocalphase information. We also demonstrate that local phase alone can be effectively used to detect local motion. The local phase-based motion detector is akin to models employed to detect motion in biological vision, for example, the Reichardt detector. The local phase-based motion detection algorithm introduced here consists of two building blocks. The first building block measures/evaluates the temporal change of the local phase. The temporal derivative of the local phase is shown to exhibit the structure of a second order Volterra kernel with two normalized inputs. We provide an efficient, FFT-based algorithm for implementing the change of the local phase. The second processing building block implements the detector; it compares the maximum of the Radon transform of the local phase derivative with a chosen threshold. We demonstrate examples of applying the local phase-based motion detection algorithm on several video sequences. We also show how the locally detected motion can be used for segmenting moving objects in video scenes and compare our local phase-based algorithm to segmentation achieved with a widely used optic flow algorithm.

The compound eyes of the fruit flies consist of the retina and 4 neuropils. The retina comprises about 700~800 ommatidia, each of which houses 8 photoreceptors, where the phototransduction takes place. Although seemingly simple, each photoreceptor can perform complex computation on their own [1]. Nevertheless, it is not yet clear how individual photoreceptors contribute to an overall spatiotemporal processing of visual scenes. Towards this goal, we constructed a full-scale simulation of the retina on a cluster of GPUs. The geometry of the eye was taken into account by building a hexagonal array of 721 ommatidia on a hemisphere and assigning a proper optic axis and an acceptance angle to each of the photoreceptors. The visual stimulus was presented on either a hemisphere or a cylinder surrounding the eye, as in many experimental settings (see Figure ​Figure11). Figure 1 A. Stimulus presented on a hemisphere screen. B. Inputs (Number of Photons) to R1 photoreceptors in all ommatidia. C. Geometric relation between eye and screen. D. Gamma corrected screen intensity. E. Gamma corrected inputs to R1 photoreceptors. F. Voltage ... The phototransduction process is based on a biochemical model of 30,000 microvilli followed by a conductance based biophysical model of the cell membrane [1]. Here we only consider monochromatic photoreceptors R1-R6 in each ommatidium. Each photoreceptor is described by ~ 450, 000 equations; the total number of equations simulated for the entire retina amounted to about 1.95 billion. All the simulations were performed on 4 Tesla K20X GPUs; it took approx. 7 minutes to finish 1 second of simulation. After constructing the detailed simulation model, we tested the processing of visual stimuli by the retina circuit. Furthermore, to test the effect of feedback originating in the lamina on the photoreceptors, we linked the retina model with a lamina model [2] in the Neurokernel environment [3]. An example of response of the circuit to natural stimuli is shown in Figure ​Figure1.1. The simulations reveal that composition rules in the lamina circuit affect the spatiotemporal processing carried out by the photoreceptors under the model parameters considered here. Finally, we scaled up the retina model of the fruit fly to the size of the blow fly to investigate the effect of visual acuity and noise.

Neural circuits built with complex cells play a key role in modeling the primary visual cortex. The encoding capability of an ensemble of complex cells has not been systematically studied, however. Can visual scenes be reconstructed using the spike times generated by an ensemble of complex cells? Can the processing taking place in complex cells be identified with high accuracy? Processing by complex cells has the complexity of Volterra models [1]. General Volterra based models call, among others, for efficient functional identification and decoding algorithms. We demonstrate that complex cells exhibit Volterra dendritic stimulus processors (Volterra DSPs) that are analytically and computationally tractable. Decoding and identification problems arising in neural circuits built with complex cells can be efficiently solved as rank minimization problems [2]. We provide (i) an algorithm that reconstructs the visual stimuli based on the spike times generated by circuits with widely employed complex cells models (Complex Cell Time Decoding Machines), and (ii) propose a mechanistic algorithm for functionally identifying the processing in complex cells using the spike times they generate (Complex Cell Channel Identification Machines). These algorithms are based on the key observation that the functional identification of processing in a single complex cell is dual to the problem of decoding stimuli encoded by an ensemble of complex cells. In addition, we show that the number of spikes needed for perfect reconstruction of a band-limited stimulus is proportional to the dimension of the stimulus space rather than the square of its dimension, thereby reducing the required number of neurons/measurements to a physiologically plausible range. This result demonstrates that visual stimuli can be efficiently reconstructed from the amplitude information carried in the complex cells alone. Similar results obtained for identification establish the computational tractability of higher order Volterra DSPs. We provide examples of perfect reconstruction of space-time stimuli (Figure ​(Figure1A)1A) and examples of identification of complex cell DSPs (Figure ​(Figure1B).1B). We demonstrate that our identification algorithms substantially outperform algorithms based on spike-triggered covariance (STC) (Figure ​(Figure1C).1C). Finally, we evaluate our identification algorithms by reconstructing novel stimuli in the input space using identified Volterra DSPs (Figure ​(Figure1D)1D) [3]. Figure 1 A. SNR of reconstruction when varying the number of measurements/spikes (dimension of the input space: 117). B. Mean SNR of identified complex cell DSP when varying the number of input trials used in identification. C. Comparison of identification by ...