Cognitive Graph Research Articles

A Cognitive Method for Automatically Retrieving Complex Information on a Large Scale

Modern retrieval systems tend to deteriorate because of their large output of useless and even misleading information, especially for complex search requests on a large scale. Complex information retrieval (IR) tasks requiring multi-hop reasoning need to fuse multiple scattered text across two or more documents. However, there are two challenges for multi-hop retrieval. To be specific, the first challenge is that since some important supporting facts have little lexical or semantic relationship with the retrieval query, the retriever often omits them; the second challenge is that once a retriever chooses misinformation related to the query as the entities of cognitive graphs, the retriever will fail. In this study, in order to improve the performance of retrievers in complex tasks, an intelligent sensor technique was proposed based on a sub-scope with cognitive reasoning (2SCR-IR), a novel method of retrieving reasoning paths over the cognitive graph to provide users with verified multi-hop reasoning chains. Inspired by the users’ process of step-by-step searching online, 2SCR-IR includes a dynamic fusion layer that starts from the entities mentioned in the given query, explores the cognitive graph dynamically built from the query and contexts, gradually finds relevant supporting entities mentioned in the given documents, and verifies the rationality of the retrieval facts. Our experimental results show that 2SCR-IR achieves competitive results on the HotpotQA full wiki and distractor settings, and outperforms the previous state-of-the-art methods by a more than two points absolute gain on the full wiki setting.

Probing the invariant structure of spatial knowledge: Support for the cognitive graph hypothesis

We tested four hypotheses about the structure of spatial knowledge used for navigation: (1) the Euclidean hypothesis, a geometrically consistent map; (2) the Neighborhood hypothesis, adjacency relations between spatial regions, based on visible boundaries; (3) the Cognitive Graph hypothesis, a network of paths between places, labeled with approximate local distances and angles; and (4) the Constancy hypothesis, whatever geometric properties are invariant during learning. In two experiments, different groups of participants learned three virtual hedge mazes, which varied specific geometric properties (Euclidean Control Maze, Elastic Maze with stretching paths, Swap Maze with alternating paths to the same place). Spatial knowledge was then tested using three navigation tasks (metric shortcuts on empty ground plane, neighborhood shortcuts with visible boundaries, route task in corridors). They yielded the following results: (a) Metric shortcuts were insensitive to detectable shifts in target location, inconsistent with the Euclidean hypothesis. (b) Neighborhood shortcuts were constrained by visible boundaries in the Elastic Maze, but not in the Swap Maze, contrary to the Neighborhood and Constancy hypotheses. (c) The route task indicated that a graph of the maze was acquired in all environments, including knowledge of local path lengths. We conclude that primary spatial knowledge is consistent with the Cognitive Graph hypothesis. Neighborhoods are derived from the graph, and local distance and angle information is not embedded in a geometrically consistent map.

Waypoint Path Planning With Synaptic-Dependent Spike Latency

Path planning, critical to mobile robot autonomy, is an essential component of long-range navigation that seeks to determine the sequence of steps required to reach a goal based on prior encoding of the environment. Our current neurobiological understanding of such a pathfinding process comes from the research on hippocampal place cells. We use an interconnected network of spiking neurons to model place cells as a cognitive graph. Stimulation of the place cell that represents the goal location causes activation of place cells through spike propagation on the network. We describe neural networks that could be used to decode this spiking activity to select the direction for movement. Two models of synaptic-dependent spike latency are compared and their influence on path planning is illustrated using examples implemented with a neuromorphic VLSI circuit.

Novel QoS optimization paradigm for IoT systems with fuzzy logic and visual information mining integration

The Internet of Things is a new round of information technology revolution after computers, the Internet and mobile communications. Internet of Things technology is an important means to improve the level of social information, which will have a profound impact on economic development and social life. IoT can stimulate the economy, increase employment, improve efficiency and make people’s lives and work more convenient. Since fuzzy control can make good use of expert fuzzy information and effectively deal with the complex process of modeling, fuzzy control has received extensive attention once it has been proposed. Fuzzy logic system has become a research hotspot in academic and application fields due to its wide application. Fuzzy system identification includes structure identification and parameter identification. Fuzzy cognitive graph is a kind of soft computing method. It has stronger semantics than neural network because of its intuitive expression ability and powerful reasoning ability. Due to the widespread popularity of visual data acquisition devices, people can use the device to capture a large number of videos and images and spread them over the network in daily learning, production, life, work and entertainment. Computer science and technology, information computing technology, automated detection technology and Internet of Things technology contribute to the research of visual information data. In this paper, we conduct research on the novel QoS optimization paradigm for the IoT systems based on fuzzy logic and visual information mining integration. The experimental results show that the proposed optimization scheme has higher robustness.

Non-Euclidean navigation.

A basic set of navigation strategies supports navigational tasks ranging from homing to novel detours and shortcuts. To perform these last two tasks, it is generally thought that humans, mammals and perhaps some insects possess Euclidean cognitive maps, constructed on the basis of input from the path integration system. In this article, I review the rationale and behavioral evidence for this metric cognitive map hypothesis, and find it unpersuasive: in practice, there is little evidence for truly novel shortcuts in animals, and human performance is highly unreliable and biased by environmental features. I develop the alternative hypothesis that spatial knowledge is better characterized as a labeled graph: a network of paths between places augmented with local metric information. What distinguishes such a cognitive graph from a metric cognitive map is that this local information is not embedded in a global coordinate system, so spatial knowledge is often geometrically inconsistent. Human path integration appears to be better suited to piecewise measurements of path lengths and turn angles than to building a consistent map. In a series of experiments in immersive virtual reality, we tested human navigation in non-Euclidean environments and found that shortcuts manifest large violations of the metric postulates. The results are contrary to the Euclidean map hypothesis and support the cognitive graph hypothesis. Apparently Euclidean behavior, such as taking novel detours and approximate shortcuts, can be explained by the adaptive use of non-Euclidean strategies.

A large-group emergency risk decision method based on data mining of public attribute preferences

Aiming at the lack of public concern about the attributes of large-group emergency decision making in major emergencies, a large-group emergency risk decision method based on data mining of public attribute preferences is proposed. First, according to historical data from a similar emergency, an attribute-keyword lexicon is established using text mining and latent semantic analysis. Meanwhile, fuzzy association rule mining and a fuzzy cognitive graph are applied to obtain information regarding public opinion about the attributes. Then, a risk measurement model of degree indicators based on information entropy is presented, and according to the linguistic preference for information given by large-group members, the decision risk of large-group members is measured. On this basis, a group member clustering algorithm is constructed to cluster large groups, and then the best alternative is acquired by comparing score function and precise function of the interval intuitionistic fuzzy number. Finally, a case application using historical data is conducted to verify the method’s rationality and feasibility.

УПРАВЛІННЯ ПРОФЕСІЙНИМИ КОМПЕТЕНТНОСТЯМИ ІТ ФАХІВЦІВ ВІДПОВІДНО ДО ВИМОГ ІНДУСТРІЇ НА ОСНОВІ КОГНІТИВНИХ КАРТ

The article analyzes and summarizes competences of the IT professions defined by the European Competence Framework and competencies declared by higher IT education standards. The authors carried out a comparative analysis of competencies identified by modern educational IT standards in order to determine the commonalities and differences in the content of IT specialists training in Ukraine with international standards. With the help of cognitive cards, a simulation of the development of professional IT competencies was carried out within the framework of the cognitive modeling methodology. As designations of concepts at the vertices of the cognitive graph, the notions of competences defined in the notation of the European competence framework are taken. Building cognitive cards to manage the competencies of IT professionals and imitation on their basis of managerial influences allow exploring the subject area and developing strategies close to optimal for organizing the learning process in order to improve the quality of training specialists for the IT industry.

Wormholes in virtual space: From cognitive maps to cognitive graphs

Humans and other animals build up spatial knowledge of the environment on the basis of visual information and path integration. We compare three hypotheses about the geometry of this knowledge of navigation space: (a) ‘cognitive map’ with metric Euclidean structure and a consistent coordinate system, (b) ‘topological graph’ or network of paths between places, and (c) ‘labelled graph’ incorporating local metric information about path lengths and junction angles. In two experiments, participants walked in a non-Euclidean environment, a virtual hedge maze containing two ‘wormholes’ that visually rotated and teleported them between locations. During training, they learned the metric locations of eight target objects from a ‘home’ location, which were visible individually. During testing, shorter wormhole routes to a target were preferred, and novel shortcuts were directional, contrary to the topological hypothesis. Shortcuts were strongly biased by the wormholes, with mean constant errors of 37° and 41° (45° expected), revealing violations of the metric postulates in spatial knowledge. In addition, shortcuts to targets near wormholes shifted relative to flanking targets, revealing ‘rips’ (86% of cases), 'folds' (91%), and ordinal reversals (66%) in spatial knowledge. Moreover, participants were completely unaware of these geometric inconsistencies, reflecting a surprising insensitivity to Euclidean structure. The probability of the shortcut data under the Euclidean map model and labelled graph model indicated decisive support for the latter (BFGM>100). We conclude that knowledge of navigation space is best characterized by a labelled graph, in which local metric information is approximate, geometrically inconsistent, and not embedded in a common coordinate system. This class of ‘cognitive graph’ models supports route finding, novel detours, and rough shortcuts, and has the potential to unify a range of data on spatial navigation.

Development of artificial cognitive systems based on models of the brain of living organisms

Despite the continuous growth of our knowledge of the functioning of nervous systems and sophisticated features of the neuroanatomy and neurophysiology of various animal species, the basic mechanisms that provide for such properties as the ability to learn, use memory, recognize patterns, and learn about the world are poorly understood. In this paper, we present an overview of artificial devices that model the brain and solve such cognitive tasks as navigation, pattern recognition, routing, and target site finding. We discuss both hybrid systems (hybrots), in which living neural networks control an artificial body, and systems in which such an artificial body is controlled by computer programs based on different models of the brain and its regions (animats). Two basic types of hybrid systems are considered: those in which the robot is connected to the brain of a living body, such as a rat, and those in which information is received from neurons taken from the body or neurons cultured on a microelectrode array detecting their electrical potentials. Among the computational approaches that simulate nervous systems of living organisms, we can mark out the Darwin family of devices based on the theory of neuronal group selection (TNGS). In addition, we consider papers in which animats solve navigation tasks using different models of the rat hippocampus, based on such modeling methods as cognitive graph, view cells, place cells, and experience cells. The approaches under consideration provide researchers with new tools to analyze basic principles of neuron interaction between each other and with the outside world, the principles that provide higher brain functions.

The Spectrum Distribution Algorithm on the Cognitive Radio System

In order to reduce complexity for gaining spectrum from the spectrum sharing pool, The thesis researches the idea of combination with using the graph theory and game theory, first spectrum allocation uses graph theory, and then second distribution uses game theory based on water-filling, to a certain extent, graph theory algorithm is intuitive, it easy to understand, in cognitive radio network graph theory model complete firstly the spectrum distribution, then between cognitive users game by using waterfilling algorithm and optimize the channel to complete the redistribution of channels. Finally, simulation results show that the spectrum resources can be fully utilized, under transmission power constraining, each cognitive user optimizes their throughput, allocates reasonable power, the simulation can verify that the algorithm of combination with graph theory and game theory has convergence and effectiveness.

Bi-linear adaptive estimation of Fuzzy Cognitive Networks

Fuzzy Cognitive Networks (FCNs) have been introduced by the authors as an operational extension of Fuzzy Cognitive Maps (FCMs), initially introduced by Kosko to model complex behavioral systems in various scientific areas. FCNs rely on the admission that the underlying cognitive graph reaches a certain equilibrium point after an initial perturbation. Weight conditions for reaching equilibrium points have been recently derived in [54] along with an algorithm for weight estimation. In this paper, the conditions are extended to take into account not only the weights of the map but also the inclination parameters of the involved sigmoid functions, increasing the structural flexibility of the network. This in turn gives rise to the development of a new adaptive bilinear weight and sigmoid parameter estimation algorithm, which employs appropriate weight projection criteria to assure that the equilibrium is always achieved.

A computational study of action planning in syntax evolution

When we think about what makes humans distinct from other species, the use of language and the use of tools typically come to mind. However, we lack an adequate theory to account for the evolutionary origin and material substrate of these abilities [1]. In particular, we lack a bridging theory, that is, an explanatory framework that relates phenomena at one level (cognition) with another level (the brain). I hypothesize (a) that action planning [2] may form a key bridge linking linguistic and non-linguistic cognition; (b) that tool use and linguistic abilities may have co-evolved from simpler motor cognition; and (c) that this co-evolution may be understood in terms of the increasing topological complexity of brain network dynamical systems over evolutionary time. In order to test these hypotheses for plausibility, I have been developing a simple computer model which implements some of these ideas in the context of a highly idealized predator-prey interaction. The computational model consists of two non-linguistic cognitive agents, herbivore H and predator P, interacting with one another and with an environment E. Both H and P move through E, selecting and elaborating actions based on their goals and the perception of their environment. H has two goals: to find food and to avoid P. P has only one goal: to catch H. E an unbounded 2D plane, with a sparse set of points that have special properties, either as food or shelter points. Agents move through he environment according to a specified set of rules. A basic assumption of the model is the necessity to account for the entire cognitive context, not simply the action planning proto-grammar. Each agent has a cognitive structure that is represented as a graph. Each node of the cognitive graph represents an hypothesized neurocognitive subsystem, implemented as a finite state machine. Each node is itself an autonomous agent, interacting with a subset of other nodes. The nodes are connected by edges, through which information may be exchanged between connected nodes (generally bidirectional). The outermost node set is the sensorimotor layer. On the sensory side, the orientation and distance of environmental landmarks (food, shelter, other agents) are represented with respect to the current position of the agent. On the motor side, the currently invoked motor program is represented as the orientation, speed, and stopping condition of the current move. The representation layer associates properties (such as food or shelter) or actions (e.g., move towards a target) with the sensorimotor field. The representation layer embodies action planning, combining sensorimotor representations with goals, as determined by the executive and integration layers. There is also an emotion layer, which assigns approach/avoidance valences to target locations. The states of the action planning node implement a simplified proto-grammar, represented as a graph. The action representation graph has terminal nodes representing actions and their objects (or targets). The action representation dynamics are biased by inputs from the integration layer. The proto-grammar consists of two sorts of tokens (actions and objects) along with two operations (select and join). The selection operator selects an action, dependent on the states of other nodes. The join operator associates the selected action with a suitable object. The join operator therefore may be seen as an evolutionary precursor to the postulated linguistic merge operator [3]. Unlike merge, however, join is not recursive, in the current implementation of the model. In addition, operations that are conventionally segregated into semantic, pragmatic, and syntactic functions are mingled in the proto-grammar. In its present form, the model accounts only for productive (as opposed to receptive) aspects of action planning. The model is still in an early of development. Results will be presented that correlate the internal model states (including sensitivity to the choice of model parameters) with the overt behavior of the agents, visualized as trajectories in the simulated 2D environment.

Rough center mining algorithm of rough cognitive map

For the large-scale systems with complex causal connection in practical problems, fusion rough cognitive maps and graph mining methods, designed a rough central area mining algorithm of rough cognitive maps to complete the pruning problem of rough cognitive map, simplified the structure of rough cognitive map to facilitate the further relations. First of all, from the perspective of the arc, give another definition of rough cognitive map, introduce the concept of rough centrality R accuracy; second, define certainty -possible class road, according to the given threshold to determine the rough central area. Finally, design an effective rough central area mining algorithm for further analysis and forecast the rough cognitive map.

Comparaison paléocognitive des niveaux moustérien et magdalénien de la grotte Tournal (Bize, Aude)

We analyse the functional processing chain of large mammals acquired at the mousterian and magdalenian levels of the Tournal cave (Bize, Aude). Palethnologic results are analysed through a method from experimental psychology of goal oriented activities. The elements are the objects used by prehistoric man and their properties. Objects and properties are linked by binary relationships. These relationships are processed by software, SIMBOL, developed at the IPH, based on an algorithm initially published by the Laboratoire de psychologie expérimentale of Paris-8. SIMBOL returns the cognitive graph of the task representation for the mousterian and the magdalenian layer, allowing a rigorous comparison of the cognitive activities on both periods. The main result is that the logical structure is preserved when passing from one level to the other. This could be interpreted as a due to the noticeable cognitive similarity of the men from the different periods, along with the fact that cognitive complexity seems to be the same on both graphs. Cognitive fluidity improves from mousterian to magdalenian suggesting that cognitive mechanisms such as sense sliding and analogies could be easier at the magdalenian level. Could we interpret this result as a due to a greater creativity at the magdalenian level than at the mousterian one? Further investigations are needed to state this, as well as enhanced data volume, potentially allowed by this entirely new method.

Animat navigation using a cognitive graph.

This article describes a computational model of the hippocampus that makes it possible for a simulated rat to navigate in a continuous environment containing obstacles. This model views the hippocampus as a "cognitive graph", that is, a hetero-associative network that learns temporal sequences of visited places and stores a topological representation of the environment. Calling upon place cells, head direction cells, and "goal cells", it suggests a biologically plausible way of exploiting such a spatial representation for navigation that does not require complicated graph-search algorithms. Moreover, it permits "latent learning" during exploration, that is, the building of a spatial representation without the need of any reinforcement. When the rat occasionally discovers some rewarding place it may wish to rejoin subsequently, it simply records within its cognitive graph, through a series of goal and sub-goal cells, the direction in which to move from any given start place. Accordingly, the model implements a simple "place-recognition-triggered response" navigation strategy. Two implementations of place cell management are studied in parallel. The first one associates place cells with place fields that are given a priori and that are uniformly distributed in the environment. The second one dynamically recruits place cells as exploration proceeds and adjusts the density of such cells to the local complexity of the environment. Both implementations lead to identical results. The article ends with a few predictions about results to be expected in experiments involving simultaneous recordings of multiple cells in the rat hippocampus.

The hippocampus as a cognitive graph.

A theory of cognitive mapping is developed that depends only on accepted properties of hippocampal function, namely, long-term potentiation, the place cell phenomenon, and the associative or recurrent connections made among CA3 pyramidal cells. It is proposed that the distance between the firing fields of connected pairs of CA3 place cells is encoded as synaptic resistance (reciprocal synaptic strength). The encoding occurs because pairs of cells with coincident or overlapping fields will tend to fire together in time, thereby causing a decrease in synaptic resistance via long-term potentiation; in contrast, cells with widely separated fields will tend never to fire together, causing no change or perhaps (via long-term depression) an increase in synaptic resistance. A network whose connection pattern mimics that of CA3 and whose connection weights are proportional to synaptic resistance can be formally treated as a weighted, directed graph. In such a graph, a "node" is assigned to each CA3 cell and two nodes are connected by a "directed edge" if and only if the two corresponding cells are connected by a synapse. Weighted, directed graphs can be searched for an optimal path between any pair of nodes with standard algorithms. Here, we are interested in finding the path along which the sum of the synaptic resistances from one cell to another is minimal. Since each cell is a place cell, such a path also corresponds to a path in two-dimensional space. Our basic finding is that minimizing the sum of the synaptic resistances along a path in neural space yields the shortest (optimal) path in unobstructed two-dimensional space, so long as the connectivity of the network is great enough. In addition to being able to find geodesics in unobstructed space, the same network enables solutions to the "detour" and "shortcut" problems, in which it is necessary to find an optimal path around a newly introduced barrier and to take a shorter path through a hole opened up in a preexisting barrier, respectively. We argue that the ability to solve such problems qualifies the proposed hippocampal object as a cognitive map. Graph theory thus provides a sort of existence proof demonstrating that the hippocampus contains the necessary information to function as a map, in the sense postulated by others (O'Keefe, J., and L. Nadel. 1978. The Hippocampus as a Cognitive Map. Clarendon Press, Oxford, UK). It is also possible that the cognitive mapping functions of the hippocampus are carried out by parallel graph searching algorithms implemented as neural processes. This possibility has the great attraction that the hippocampus could then operate in much the same way to find paths in general problem space; it would only be necessary for pyramidal cells to exhibit a strong nonpositional firing correlate.

Cognitive graphs, resistive grids, and the hippocampal representation of space.

The hippocampal formation appears to play an important role in the ability of rats and other mammals to perform sophisticated spatial tasks, beyond simply following a familiar route or approaching a visible cue. Evidence for this role came from the observation of selective spatial deficits after hippocampal lesions (O'Keefe et al., 1975; O'Keefe and Nadel, 1978; Morris et al., 1982; Sutherland et al., 1982) and from the spatially selective firing of hippocampal complex spike cells (O'Keefe and Dostrovsky, 1971; McNaughton et al., 1983; Muller et al., 1987) (place cells), and led to the suggestion that the hippocampus was the locus of a map in the mammalian brain (O'Keefe and Nadel, 1978). The hippocampus is thought to have a broader mnemonic function in higher mammals like monkeys and humans, (Scoville and Milner, 1957), such as episodic memory (Kinsbourne and Wood, 1975; O'Keefe and Nadel, 1978; Tulving, 1983). The extensive electrophysiological data available from studies in freely moving rats raise the possibility that the functional properties of the neural machinery of the hippocampus might be unders tood in the spatial domain, which could also shed light on other putative hippocampal functions for which there are fewer neuronal data. In this issue of The Journal of General Physiology, Muller et al. (1996) present a model in which the synaptic efficacies of the recurrent collaterals in the CA3 region of the hippocampus form a cognitive graph. They initially demonstrate that CA3 has properties adequate to support a directed connected graph, and they go on to outline their model. In this model, the net resistance of the synaptic connect ion between two place cells increases monotonically with the time taken to move between the corresponding firing fields (place fields), and, hence, on their separation. The graph is built as the animal moves around the environment and longterm potentiat ion (LTP) causes synapses between place cells with overlapping place fields to strengthen much more than synapses between cells with well separated fields, because of the increased chance of both cells firing within the LTP-permissive time window. Once the graph is formed, such that the connections between cells within it effectively code for the distance between their place fields, it can be used as follows. If one place cell represents a start location and another represents the goal location, then the chain of place cells between them that involves the strongest synapses (i.e., path of minimum resistance through the network) corresponds to a good route between the start and goal locations (traced out by the place fields along the chain). How this route might be read by the rest of the brain is not explicitly simulated. This simple model has many attractive features. For example, rats exhibit latent learning, in which they clearly learn about the layout of an environment in the absence of goals or motivation, helping them to navigate once a goal is provided. This is reflected in the way in which the graph is automatically constructed during exploration. Similarly, the ability to perform detours and shortcuts exemplifies the type of navigation impaired by hippocampal lesions and can be naturally incorporated in the model. Detours occur because placing an obstacle on or near a place field tends to inhibit the firing of the place cell, thus taking that cell out of the path-planning process. Shortcuts can be found because, while place cells with fields on ei ther side of a barrier will tend to have weak synaptic connect ion because of the time taken to travel f rom one field to the other, the rat will tend to explore the area if the barrier is removed and will pass directly from one to the other, causing the connect ion to strengthen. Another strength of the model is that the directional modulat ion of place cell firing that occurs when a rat is constrained to run along restricted paths does not destroy the model. However, directional firing would tend to exaggerate the problem of the animal's exploratory behavior, biasing the representat ion of distance in the graph, for example, so as to underest imate distances along familiar routes compared with unfamiliar ones (see below). The graph model appears to work well; indeed, it is reminiscent of a successful path-planning scheme based on resistive grids (e.g., see Connelly et al., 1990) that has been extensively studied in robotics

The hippocampus as a cognitive graph (abridged version).

The hippocampus as a cognitive graph (abridged version).