Deep Learning of Orthographic Representations in Baboons

What is the origin of our ability to learn orthographic knowledge? We use deep convolutional networks to emulate the primate''s ventral visual stream and explore the recent finding that baboons can be trained to discriminate English words from nonwords [1]. The networks were exposed to the exact same sequence of stimuli and reinforcement signals as the baboons in the experiment, and learned to map real visual inputs (pixels) of letter strings onto binary word/nonword responses. We show that the networks'' highest levels of representations were indeed sensitive to letter combinations as postulated in our previous research. The model also captured the key empirical findings, such as generalization to novel words, along with some intriguing inter-individual differences. The present work shows the merits of deep learning networks that can simulate the whole processing chain all the way from the visual input to the response while allowing researchers to analyze the complex representations that emerge during the learning process.     

Structural Analysis to Determine the Core of Hypoxia Response Network

The advent of sophisticated molecular biology techniques allows to deduce the structure of complex biological networks. However, networks tend to be huge and impose computational challenges on traditional mathematical analysis due to their high dimension and lack of reliable kinetic data. To overcome this problem, complex biological networks are decomposed into modules that are assumed to capture essential aspects of the full network''s dynamics. The question that begs for an answer is how to identify the core that is representative of a network''s dynamics, its function and robustness. One of the powerful methods to probe into the structure of a network is Petri net analysis. Petri nets support network visualization and execution. They are also equipped with sound mathematical and formal reasoning based on which a network can be decomposed into modules. The structural analysis provides insight into the robustness and facilitates the identification of fragile nodes. The application of these techniques to a previously proposed hypoxia control network reveals three functional modules responsible for degrading the hypoxia-inducible factor (HIF). Interestingly, the structural analysis identifies superfluous network parts and suggests that the reversibility of the reactions are not important for the essential functionality. The core network is determined to be the union of the three reduced individual modules. The structural analysis results are confirmed by numerical integration of the differential equations induced by the individual modules as well as their composition. The structural analysis leads also to a coarse network structure highlighting the structural principles inherent in the three functional modules. Importantly, our analysis identifies the fragile node in this robust network without which the switch-like behavior is shown to be completely absent.     

A Family of Algorithms for Computing Consensus about Node State from Network Data

Biological and social networks are composed of heterogeneous nodes that contribute differentially to network structure and function. A number of algorithms have been developed to measure this variation. These algorithms have proven useful for applications that require assigning scores to individual nodes–from ranking websites to determining critical species in ecosystems–yet the mechanistic basis for why they produce good rankings remains poorly understood. We show that a unifying property of these algorithms is that they quantify consensus in the network about a node''s state or capacity to perform a function. The algorithms capture consensus by either taking into account the number of a target node''s direct connections, and, when the edges are weighted, the uniformity of its weighted in-degree distribution (breadth), or by measuring net flow into a target node (depth). Using data from communication, social, and biological networks we find that that how an algorithm measures consensus–through breadth or depth– impacts its ability to correctly score nodes. We also observe variation in sensitivity to source biases in interaction/adjacency matrices: errors arising from systematic error at the node level or direct manipulation of network connectivity by nodes. Our results indicate that the breadth algorithms, which are derived from information theory, correctly score nodes (assessed using independent data) and are robust to errors. However, in cases where nodes “form opinions” about other nodes using indirect information, like reputation, depth algorithms, like Eigenvector Centrality, are required. One caveat is that Eigenvector Centrality is not robust to error unless the network is transitive or assortative. In these cases the network structure allows the depth algorithms to effectively capture breadth as well as depth. Finally, we discuss the algorithms'' cognitive and computational demands. This is an important consideration in systems in which individuals use the collective opinions of others to make decisions.     

Multi-Scale Clustering by Building a Robust and Self Correcting Ultrametric Topology on Data Points

The advent of high-throughput technologies and the concurrent advances in information sciences have led to an explosion in size and complexity of the data sets collected in biological sciences. The biggest challenge today is to assimilate this wealth of information into a conceptual framework that will help us decipher biological functions. A large and complex collection of data, usually called a data cloud, naturally embeds multi-scale characteristics and features, generically termed geometry. Understanding this geometry is the foundation for extracting knowledge from data. We have developed a new methodology, called data cloud geometry-tree (DCG-tree), to resolve this challenge. This new procedure has two main features that are keys to its success. Firstly, it derives from the empirical similarity measurements a hierarchy of clustering configurations that captures the geometric structure of the data. This hierarchy is then transformed into an ultrametric space, which is then represented via an ultrametric tree or a Parisi matrix. Secondly, it has a built-in mechanism for self-correcting clustering membership across different tree levels. We have compared the trees generated with this new algorithm to equivalent trees derived with the standard Hierarchical Clustering method on simulated as well as real data clouds from fMRI brain connectivity studies, cancer genomics, giraffe social networks, and Lewis Carroll''s Doublets network. In each of these cases, we have shown that the DCG trees are more robust and less sensitive to measurement errors, and that they provide a better quantification of the multi-scale geometric structures of the data. As such, DCG-tree is an effective tool for analyzing complex biological data sets.     

Cursive word recognition based on interactive activation and early visual processing models

We present an off-line cursive word recognition system based completely on neural networks: reading models and models of early visual processing. The first stage (normalization) preprocesses the input image in order to reduce letter position uncertainty; the second stage (feature extraction) is based on the feedforward model of orientation selectivity; the third stage (letter pre-recognition) is based on a convolutional neural network, and the last stage (word recognition) is based on the interactive activation model.     

The knowledge used in vision and where it comes from.

Knowledge is often thought to be something brought from outside to act upon the visual messages received from the eye in a ''top-down'' fashion, but this is a misleadingly narrow view. First, the visual system is a multilevel heterarchy with connections acting in all directions so it has no ''top''; and second, knowledge is provided through innately determined structure and by analysis of the redundancy in sensory messages themselves, as well as from outside. This paper gives evidence about mechanisms analysing sensory redundancy in biological vision. Automatic gain controls for luminance and contrast depend upon feedback from the input, and there are strong indications that the autocorrelation function, and other associations between input variables, affect the contrast sensitivity function and our subjective experience of the world. The associative structure of sensory message can provide much knowledge about the world we live in, and neural mechanisms that discount established associative structure in the input messages by recoding them can improve survival by making new structure more easily detectable. These mechanisms may be responsible for illusions, such as those produced by a concave face-mask, that are classically attributed to top-down influences.     

Are combined orthopaedic and rheumatology clinics worth while?

The practice of holding combined orthopaedic and rheumatology clinics is widespread, but no attempt has been made to assess their usefulness. In a one year prospective study patients were randomised either to a combined clinic with a rheumatologist and an orthopaedic surgeon or to a clinic with an orthopaedic surgeon alone. The details of the patients'' disease, the interview, the surgeons'' and rheumatologists'' responses, and the patients'' opinions were recorded on questionnaires. Interviews were appreciably longer in the combined clinic, and more referrals for surgery were made. The rheumatologist correctly predicted the need for surgery in 95% of cases, and his presence in the clinic was considered desirable by the surgeon, principally when the referral letter was inadequate. It is concluded that most patients with rheumatic diseases can be seen more efficiently in routine orthopaedic clinics provided a good quality letter of referral is sent.     

Orangutan information broadcast via consonant-like and vowel-like calls breaches mathematical models of linguistic evolution

The origin of language is one of the most significant evolutionary milestones of life on Earth, but one of the most persevering scientific unknowns. Two decades ago, game theorists and mathematicians predicted that the first words and grammar emerged as a response to transmission errors and information loss in language''s precursor system, however, empirical proof is lacking. Here, we assessed information loss in proto-consonants and proto-vowels in human pre-linguistic ancestors as proxied by orangutan consonant-like and vowel-like calls that compose syllable-like combinations. We played back and re-recorded calls at increasing distances across a structurally complex habitat (i.e. adverse to sound transmission). Consonant-like and vowel-like calls degraded acoustically over distance, but no information loss was detected regarding three distinct classes of information (viz. individual ID, context and population ID). Our results refute prevailing mathematical predictions and herald a turning point in language evolution theory and heuristics. Namely, explaining how the vocal–verbal continuum was crossed in the hominid family will benefit from future mathematical and computational models that, in order to enjoy empirical validity and superior explanatory power, will be informed by great ape behaviour and repertoire.     

基于复杂网络的长三角城市对外服务群落结构研究

城市群已成为国家参与全球竞争与国际分工的重要空间载体,其空间组织形式正从个体城市集聚、等级性的中心地结构向多中心、嵌套式的群落结构演变。为刻画长三角地区城市在不同要素层面下形成的多层次群落结构及其状态,借助复杂网络分析工具,从城市群落的节点特征、垂直和水平结构、不同群落结构间的相互关联三方面实证分析长三角地区3类(生产性、生活性、公共性)对外服务流的网络结构特征。研究显示:1)长三角城市节点服务功能分化,节点层级性分异显著,生产和生活性服务网络的节点规模呈"长尾"分布,公共性服务网络的节点规模相对均衡;2)垂直结构上,3类对外服务网络的网络密度、网络效率、流量占比和空间分布各不相同;水平结构上,初步形成对外服务网络的专业化分工格局,部分城市突破区域界线,呈跨地域集聚组团的态势。3)较强的结构关联性存在于生产性和生活性对外服务网络之间,两者在中低度值的城市节点上具有一致性,呈联动发展格局;公共性对外服务网络与前两者的节点度值分异较大,促进了整体群落服务功能结构的丰富和完善。基于复杂网络的群落研究可以从多维结构的分析中寻求城市群落的分工协作和共生,为当前多核心、网络化的城市空间组织与规划提供科学参考。     

On the Rotation of Canonical Solutions

Some points of CLIFF /KRUS 's important rotation procedure are criticized. This result in the definition of simple structure of canonical solutions, the use of STEWART /LOVE 's redundancy index as a measure of common variance, and two new rotation procedures (HAKSTIAN 's modified varimax rotation, separate rotation of both sets). The objects of rotation should be the (intraset) loadings.     

Language evolution: syntax before phonology?

Phonology and syntax represent two layers of sound combination central to language''s expressive power. Comparative animal studies represent one approach to understand the origins of these combinatorial layers. Traditionally, phonology, where meaningless sounds form words, has been considered a simpler combination than syntax, and thus should be more common in animals. A linguistically informed review of animal call sequences demonstrates that phonology in animal vocal systems is rare, whereas syntax is more widespread. In the light of this and the absence of phonology in some languages, we hypothesize that syntax, present in all languages, evolved before phonology.     

Data Mechanics and Coupling Geometry on Binary Bipartite Networks

We quantify the notion of pattern and formalize the process of pattern discovery under the framework of binary bipartite networks. Patterns of particular focus are interrelated global interactions between clusters on its row and column axes. A binary bipartite network is built into a thermodynamic system embracing all up-and-down spin configurations defined by product-permutations on rows and columns. This system is equipped with its ferromagnetic energy ground state under Ising model potential. Such a ground state, also called a macrostate, is postulated to congregate all patterns of interest embedded within the network data in a multiscale fashion. A new computing paradigm for indirect searching for such a macrostate, called Data Mechanics, is devised by iteratively building a surrogate geometric system with a pair of nearly optimal marginal ultrametrics on row and column spaces. The coupling measure minimizing the Gromov-Wasserstein distance of these two marginal geometries is also seen to be in the vicinity of the macrostate. This resultant coupling geometry reveals multiscale block pattern information that characterizes multiple layers of interacting relationships between clusters on row and on column axes. It is the nonparametric information content of a binary bipartite network. This coupling geometry is then demonstrated to shed new light and bring resolution to interaction issues in community ecology and in gene-content-based phylogenetics. Its implied global inferences are expected to have high potential in many scientific areas.     

Bilingualism Accentuates Children's Conversational Understanding

Background

Although bilingualism is prevalent throughout the world, little is known about the extent to which it influences children''s conversational understanding. Our investigation involved children aged 3–6 years exposed to one or more of four major languages: English, German, Italian, and Japanese. In two experiments, we examined the children''s ability to identify responses to questions as violations of conversational maxims (to be informative and avoid redundancy, to speak the truth, be relevant, and be polite).

Principal Findings

In Experiment 1, with increasing age, children showed greater sensitivity to maxim violations. Children in Italy who were bilingual in German and Italian (with German as the dominant language L1) significantly outperformed Italian monolinguals. In Experiment 2, children in England who were bilingual in English and Japanese (with English as L1) significantly outperformed Japanese monolinguals in Japan with vocabulary age partialled out.

Conclusions

As the monolingual and bilingual groups had a similar family SES background (Experiment 1) and similar family cultural identity (Experiment 2), these results point to a specific role for early bilingualism in accentuating children''s developing ability to appreciate effective communicative responses.     

Colored motifs reveal computational building blocks in the C. elegans brain

Background

Complex networks can often be decomposed into less complex sub-networks whose structures can give hints about the functional organization of the network as a whole. However, these structural motifs can only tell one part of the functional story because in this analysis each node and edge is treated on an equal footing. In real networks, two motifs that are topologically identical but whose nodes perform very different functions will play very different roles in the network.

Methodology/Principal Findings

Here, we combine structural information derived from the topology of the neuronal network of the nematode C. elegans with information about the biological function of these nodes, thus coloring nodes by function. We discover that particular colorations of motifs are significantly more abundant in the worm brain than expected by chance, and have particular computational functions that emphasize the feed-forward structure of information processing in the network, while evading feedback loops. Interneurons are strongly over-represented among the common motifs, supporting the notion that these motifs process and transduce the information from the sensor neurons towards the muscles. Some of the most common motifs identified in the search for significant colored motifs play a crucial role in the system of neurons controlling the worm''s locomotion.

Conclusions/Significance

The analysis of complex networks in terms of colored motifs combines two independent data sets to generate insight about these networks that cannot be obtained with either data set alone. The method is general and should allow a decomposition of any complex networks into its functional (rather than topological) motifs as long as both wiring and functional information is available.     

Functional partitioning of yeast co-expression networks after genome duplication

Several species of yeast, including the baker's yeast Saccharomyces cerevisiae, underwent a genome duplication roughly 100 million years ago. We analyze genetic networks whose members were involved in this duplication. Many networks show detectable redundancy and strong asymmetry in their interactions. For networks of co-expressed genes, we find evidence for network partitioning whereby the paralogs appear to have formed two relatively independent subnetworks from the ancestral network. We simulate the degeneration of networks after duplication and find that a model wherein the rate of interaction loss depends on the “neighborliness” of the interacting genes produces networks with parameters similar to those seen in the real partitioned networks. We propose that the rationalization of network structure through the loss of pair-wise gene interactions after genome duplication provides a mechanism for the creation of semi-independent daughter networks through the division of ancestral functions between these daughter networks.     

北京东灵山森林植物多样性的网络结构特征

一个群落可以看作是由物种相互连接的复杂系统,刻画其网络结构有助于深入揭示系统性质以及结构与功能之间的关系。在生态学中,复杂网络理论已被成功应用于食物网与互利网络的结构研究,但尚未检验其刻画生物多样性格局的能力。采用复杂网络理论研究了北京东灵山森林乔木层、灌木层、草本层植物关联关系的网络结构特征及其差异。结果表明,植物物种的共同出现是非随机的,并表现出一定的小世界模式;乔木层、灌木层、草本层植物共同出现的网络在结构特征上存在明显差异,草本层网络比乔木层和灌木层网络更松散且平均路径更长,灌木层和草本层网络的聚类系数高于乔木层且存在度的幂律分布。皆表明复杂网络理论具备反映不同层植物多样性格局差异的能力。