Language processing with dynamic fields

We construct a mapping from complex recursive linguistic data structures to spherical wave functions using Smolensky’s filler/role bindings and tensor product representations. Syntactic language processing is then described by the transient evolution of these spherical patterns whose amplitudes are governed by nonlinear order parameter equations. Implications of the model in terms of brain wave dynamics are indicated. Electronic supplementary material  The online version of this article (doi: ) contains supplementary material, which is available to authorized users.

Insights from classifying visual concepts with multiple kernel learning

Combining information from various image features has become a standard technique in concept recognition tasks. However, the optimal way of fusing the resulting kernel functions is usually unknown in practical applications. Multiple kernel learning (MKL) techniques allow to determine an optimal linear combination of such similarity matrices. Classical approaches to MKL promote sparse mixtures. Unfortunately, 1-norm regularized MKL variants are often observed to be outperformed by an unweighted sum kernel. The main contributions of this paper are the following: we apply a recently developed non-sparse MKL variant to state-of-the-art concept recognition tasks from the application domain of computer vision. We provide insights on benefits and limits of non-sparse MKL and compare it against its direct competitors, the sum-kernel SVM and sparse MKL. We report empirical results for the PASCAL VOC 2009 Classification and ImageCLEF2010 Photo Annotation challenge data sets. Data sets (kernel matrices) as well as further information are available at http://doc.ml.tu-berlin.de/image_mkl/(Accessed 2012 Jun 25).     

Reduced multiple empirical kernel learning machine

Multiple kernel learning (MKL) is demonstrated to be flexible and effective in depicting heterogeneous data sources since MKL can introduce multiple kernels rather than a single fixed kernel into applications. However, MKL would get a high time and space complexity in contrast to single kernel learning, which is not expected in real-world applications. Meanwhile, it is known that the kernel mapping ways of MKL generally have two forms including implicit kernel mapping and empirical kernel mapping (EKM), where the latter is less attracted. In this paper, we focus on the MKL with the EKM, and propose a reduced multiple empirical kernel learning machine named RMEKLM for short. To the best of our knowledge, it is the first to reduce both time and space complexity of the MKL with EKM. Different from the existing MKL, the proposed RMEKLM adopts the Gauss Elimination technique to extract a set of feature vectors, which is validated that doing so does not lose much information of the original feature space. Then RMEKLM adopts the extracted feature vectors to span a reduced orthonormal subspace of the feature space, which is visualized in terms of the geometry structure. It can be demonstrated that the spanned subspace is isomorphic to the original feature space, which means that the dot product of two vectors in the original feature space is equal to that of the two corresponding vectors in the generated orthonormal subspace. More importantly, the proposed RMEKLM brings a simpler computation and meanwhile needs a less storage space, especially in the processing of testing. Finally, the experimental results show that RMEKLM owns a much efficient and effective performance in terms of both complexity and classification. The contributions of this paper can be given as follows: (1) by mapping the input space into an orthonormal subspace, the geometry of the generated subspace is visualized; (2) this paper first reduces both the time and space complexity of the EKM-based MKL; (3) this paper adopts the Gauss Elimination, one of the on-the-shelf techniques, to generate a basis of the original feature space, which is stable and efficient.     

Calpain cleavage prediction using multiple kernel learning

Calpain, an intracellular -dependent cysteine protease, is known to play a role in a wide range of metabolic pathways through limited proteolysis of its substrates. However, only a limited number of these substrates are currently known, with the exact mechanism of substrate recognition and cleavage by calpain still largely unknown. While previous research has successfully applied standard machine-learning algorithms to accurately predict substrate cleavage by other similar types of proteases, their approach does not extend well to calpain, possibly due to its particular mode of proteolytic action and limited amount of experimental data. Through the use of Multiple Kernel Learning, a recent extension to the classic Support Vector Machine framework, we were able to train complex models based on rich, heterogeneous feature sets, leading to significantly improved prediction quality (6% over highest AUC score produced by state-of-the-art methods). In addition to producing a stronger machine-learning model for the prediction of calpain cleavage, we were able to highlight the importance and role of each feature of substrate sequences in defining specificity: primary sequence, secondary structure and solvent accessibility. Most notably, we showed there existed significant specificity differences across calpain sub-types, despite previous assumption to the contrary. Prediction accuracy was further successfully validated using, as an unbiased test set, mutated sequences of calpastatin (endogenous inhibitor of calpain) modified to no longer block calpain''s proteolytic action. An online implementation of our prediction tool is available at http://calpain.org.     

A flow-through tissue culture system with fast dynamic response

A flow-through system in which monolayer cells, growing on a 50-cm² glass surface, are in contact with a film of medium with a thickness of 0.14 mm, is described. For murine B16 melanoma cells, the loading capacity is 8.10⁶ cells. The flow-through principle permits frequent off-line or on-line detection of medium constituents for a period on the order of days. The system has a fast dynamic response. With off-line radiochemical detection, the system was applied to the uptake of uridine and excretion of uracil over a period of 45 h. With on-line fluorescence detection, the interaction between the cells and two anthracycline analogs was monitored. The cells can be easily observed with a light microscope.     

Improved prediction of malaria degradomes by supervised learning with SVM and profile kernel

The spread of drug resistance through malaria parasite populations calls for the development of new therapeutic strategies. However, the seemingly promising genomics-driven target identification paradigm is hampered by the weak annotation coverage. To identify potentially important yet uncharacterized proteins, we apply support vector machines using profile kernels, a supervised discriminative machine learning technique for remote homology detection, as a complement to the traditional alignment based algorithms. In this study, we focus on the prediction of proteases, which have long been considered attractive drug targets because of their indispensable roles in parasite development and infection. Our analysis demonstrates that an abundant and complex repertoire is conserved in five Plasmodium parasite species. Several putative proteases may be important components in networks that mediate cellular processes, including hemoglobin digestion, invasion, trafficking, cell cycle fate, and signal transduction. This catalog of proteases provides a short list of targets for functional characterization and rational inhibitor design. Electronic supplementary material  The online version of this article (doi:) contains supplementary material, which is available to authorized users. Rui Kuang and Jianying Gu have contributed equally to this work. An erratum to this article can be found at     

Tracking population densities using dynamic neural fields with moderately strong inhibition

We discuss the ability of dynamic neural fields to track noisy population codes in an online fashion when signals are constantly applied to the recurrent network. To report on the quantitative performance of such networks we perform population decoding of the ‘orientation’ embedded in the noisy signal and determine which inhibition strength in the network provides the best decoding performance. We also study the performance of decoding on time-varying signals. Simulations of the system show good performance even in the very noisy case and also show that noise is beneficial to decoding time-varying signals.     

Adaptive diffusion kernel learning from biological networks for protein function prediction

Background  

Machine-learning tools have gained considerable attention during the last few years for analyzing biological networks for protein function prediction. Kernel methods are suitable for learning from graph-based data such as biological networks, as they only require the abstraction of the similarities between objects into the kernel matrix. One key issue in kernel methods is the selection of a good kernel function. Diffusion kernels, the discretization of the familiar Gaussian kernel of Euclidean space, are commonly used for graph-based data.     

Nonparametric evaluation of dynamic disease risk: a spatio-temporal kernel approach

Quantifying the distributions of disease risk in space and time jointly is a key element for understanding spatio-temporal phenomena while also having the potential to enhance our understanding of epidemiologic trajectories. However, most studies to date have neglected time dimension and focus instead on the "average" spatial pattern of disease risk, thereby masking time trajectories of disease risk. In this study we propose a new idea titled "spatio-temporal kernel density estimation (stKDE)" that employs hybrid kernel (i.e., weight) functions to evaluate the spatio-temporal disease risks. This approach not only can make full use of sample data but also "borrows" information in a particular manner from neighboring points both in space and time via appropriate choice of kernel functions. Monte Carlo simulations show that the proposed method performs substantially better than the traditional (i.e., frequency-based) kernel density estimation (trKDE) which has been used in applied settings while two illustrative examples demonstrate that the proposed approach can yield superior results compared to the popular trKDE approach. In addition, there exist various possibilities for improving and extending this method.     

Neural mechanisms of social learning and decision-making

One of the hallmarks of human society is the ubiquitous interactions among individuals. Indeed, a significant portion of human daily routine decision making is socially related. Normative economic theory, namely game theory, has prescribed the canonical decision strategy when "rational" social agents have full information about the decision environment. In reality, however, social decision is often influenced by the trait and state parameters of selves and others. Therefore, understanding the cognitive and neural processes of inferring the decision parameters is pivotal for social decision making. Recently, both correlational and causal non-invasive neuroimaging studies have started to reveal the critical neural computations underlying social learning and decision-making, and highlighted the unique roles of "social" brain structures such as temporal-parietal junction(TPJ) and dorsomedial prefrontal cortex(dmPFC). Here we review recent advances in social decision neuroscience and maintain the focus on how the inference about others is dynamically acquired during social learning, as well as how the prosocial(altruistic)behavior results from orchestrated interactions of different brain regions specified under the social utility framework. We conclude by emphasizing the importance of combining computational decision theory with the identification of neural mechanisms that represent, evaluate and integrate value related social information and generate decision variables guiding behavioral output in the complex social environment.     

Neural cytoskeleton capabilities for learning and memory

This paper proposes a physical model involving the key structures within the neural cytoskeleton as major players in molecular-level processing of information required for learning and memory storage. In particular, actin filaments and microtubules are macromolecules having highly charged surfaces that enable them to conduct electric signals. The biophysical properties of these filaments relevant to the conduction of ionic current include a condensation of counterions on the filament surface and a nonlinear complex physical structure conducive to the generation of modulated waves. Cytoskeletal filaments are often directly connected with both ionotropic and metabotropic types of membrane-embedded receptors, thereby linking synaptic inputs to intracellular functions. Possible roles for cable-like, conductive filaments in neurons include intracellular information processing, regulating developmental plasticity, and mediating transport. The cytoskeletal proteins form a complex network capable of emergent information processing, and they stand to intervene between inputs to and outputs from neurons. In this manner, the cytoskeletal matrix is proposed to work with neuronal membrane and its intrinsic components (e.g., ion channels, scaffolding proteins, and adaptor proteins), especially at sites of synaptic contacts and spines. An information processing model based on cytoskeletal networks is proposed that may underlie certain types of learning and memory.     

Reception and learning of electric fields in bees

Honeybees, like other insects, accumulate electric charge in flight, and when their body parts are moved or rubbed together. We report that bees emit constant and modulated electric fields when flying, landing, walking and during the waggle dance. The electric fields emitted by dancing bees consist of low- and high-frequency components. Both components induce passive antennal movements in stationary bees according to Coulomb''s law. Bees learn both the constant and the modulated electric field components in the context of appetitive proboscis extension response conditioning. Using this paradigm, we identify mechanoreceptors in both joints of the antennae as sensors. Other mechanoreceptors on the bee body are potentially involved but are less sensitive. Using laser vibrometry, we show that the electrically charged flagellum is moved by constant and modulated electric fields and more strongly so if sound and electric fields interact. Recordings from axons of the Johnston organ document its sensitivity to electric field stimuli. Our analyses identify electric fields emanating from the surface charge of bees as stimuli for mechanoreceptors, and as biologically relevant stimuli, which may play a role in social communication.     

Spectacle: fast chromatin state annotation using spectral learning

Epigenomic data from ENCODE can be used to associate specific combinations of chromatin marks with regulatory elements in the human genome. Hidden Markov models and the expectation-maximization (EM) algorithm are often used to analyze epigenomic data. However, the EM algorithm can have overfitting problems in data sets where the chromatin states show high class-imbalance and it is often slow to converge. Here we use spectral learning instead of EM and find that our software Spectacle overcame these problems. Furthermore, Spectacle is able to find enhancer subtypes not found by ChromHMM but strongly enriched in GWAS SNPs. Spectacle is available at https://github.com/jiminsong/Spectacle.

Electronic supplementary material

The online version of this article (doi:10.1186/s13059-015-0598-0) contains supplementary material, which is available to authorized users.     

Neural networks learning with sliding mode control: the sliding mode backpropagation algorithm

Based on the classical backpropagation weight update equations, sliding mode control theory is introduced as a technique to adapt weights of a multi-layer perceptron. As will be demonstrated, the introduction of sliding mode has resulted in a much faster version of the standard backpropagation. The results show also that the proposed algorithm presents some important features of sliding mode control, which are robustness and high speed of learning. In addition to that, this paper shows also how control theory can be applied to train neural networks.     

Conditions of activity bubble uniqueness in dynamic neural fields

Dynamic neural fields (DNFs) offer a rich spectrum of dynamic properties like hysteresis, spatiotemporal information integration, and coexistence of multiple attractors. These properties make DNFs more and more popular in implementations of sensorimotor loops for autonomous systems. Applications often imply that DNFs should have only one compact region of firing neurons (activity bubble), whereas the rest of the field should not fire (e.g., if the field represents motor commands). In this article we prove the conditions of activity bubble uniqueness in the case of locally symmetric input bubbles. The qualitative condition on inhomogeneous inputs used in earlier work on DNFs is transfered to a quantitative condition of a balance between the internal dynamics and the input. The mathematical analysis is carried out for the two-dimensional case with methods that can be extended to more than two dimensions. The article concludes with an example of how our theoretical results facilitate the practical use of DNFs.     

Stimulus sampling as an exploration mechanism for fast reinforcement learning

Reinforcement learning in neural networks requires a mechanism for exploring new network states in response to a single, nonspecific reward signal. Existing models have introduced synaptic or neuronal noise to drive this exploration. However, those types of noise tend to almost average out—precluding or significantly hindering learning —when coding in neuronal populations or by mean firing rates is considered. Furthermore, careful tuning is required to find the elusive balance between the often conflicting demands of speed and reliability of learning. Here we show that there is in fact no need to rely on intrinsic noise. Instead, ongoing synaptic plasticity triggered by the naturally occurring online sampling of a stimulus out of an entire stimulus set produces enough fluctuations in the synaptic efficacies for successful learning. By combining stimulus sampling with reward attenuation, we demonstrate that a simple Hebbian-like learning rule yields the performance that is very close to that of primates on visuomotor association tasks. In contrast, learning rules based on intrinsic noise (node and weight perturbation) are markedly slower. Furthermore, the performance advantage of our approach persists for more complex tasks and network architectures. We suggest that stimulus sampling and reward attenuation are two key components of a framework by which any single-cell supervised learning rule can be converted into a reinforcement learning rule for networks without requiring any intrinsic noise source. This work was supported by the Swiss National Science Foundation grant K-32K0-118084.     

Adaptive computational models of fast learning of motion direction discrimination

In a previous study, we found that subjects' performance in a task of direction discrimination in stochastic motion stimuli shows fast improvement in the absence of feedback and the learned ability is retained over a period of time. We model this learning using two unsupervised approaches: a clustering model that learns to accommodate the motion noise, and an averaging model that learns to ignore the noise. Extensive simulations with the models show performance similar to psychophysical results.