scholarly journals Hippocampal Anatomy Supports the Use of Context in Object Recognition: A Computational Model

2013 ◽  
Vol 2013 ◽  
pp. 1-19 ◽  
Author(s):  
Patrick Greene ◽  
Mike Howard ◽  
Rajan Bhattacharyya ◽  
Jean-Marc Fellous
Keyword(s):  
Entorhinal Cortex ◽  
Recognition Performance ◽  
Contextual Information ◽  
Network Models ◽  
Neural Network Models ◽  
Medial Entorhinal Cortex ◽  
Anatomical Features ◽  
Human Hippocampus ◽  
Types Of Information ◽  
Object Features

The human hippocampus receives distinct signals via the lateral entorhinal cortex, typically associated with object features, and the medial entorhinal cortex, associated with spatial or contextual information. The existence of these distinct types of information calls for some means by which they can be managed in an appropriate way, by integrating them or keeping them separate as required to improve recognition. We hypothesize that several anatomical features of the hippocampus, including differentiation in connectivity between the superior/inferior blades of DG and the distal/proximal regions of CA3 and CA1, work together to play this information managing role. We construct a set of neural network models with these features and compare their recognition performance when given noisy or partial versions of contexts and their associated objects. We found that the anterior and posterior regions of the hippocampus naturally require different ratios of object and context input for optimal performance, due to the greater number of objects versus contexts. Additionally, we found that having separate processing regions in DG significantly aided recognition in situations where object inputs were degraded. However, split processing in both DG and CA3 resulted in performance tradeoffs, though the actual hippocampus may have ways of mitigating such losses.

scholarly journals ARM-Cortex M3-Based Two-Wheel Robot for Assessing Grid Cell Model of Medial Entorhinal Cortex: Progress towards Building Robots with Biologically Inspired Navigation-Cognitive Maps

2017 ◽  
Vol 2017 ◽  
pp. 1-9
Author(s):  
J. Cuneo ◽  
L. Barboni ◽  
N. Blanco ◽  
M. del Castillo ◽  
J. Quagliotti
Keyword(s):  
Embedded System ◽  
Entorhinal Cortex ◽  
Cell Model ◽  
Systems Design ◽  
Grid Cell ◽  
Network Models ◽  
Time Algorithm ◽  
Neural Network Models ◽  
Biologically Inspired ◽  
Medial Entorhinal Cortex

This article presents the implementation and use of a two-wheel autonomous robot and its effectiveness as a tool for studying the recently discovered use of grid cells as part of mammalian’s brains space-mapping circuitry (specifically the medial entorhinal cortex). A proposed discrete-time algorithm that emulates the medial entorhinal cortex is programed into the robot. The robot freely explores a limited laboratory area in the manner of a rat or mouse and reports information to a PC, thus enabling research without the use of live individuals. Position coordinate neural maps are achieved as mathematically predicted although for a reduced number of implemented neurons (i.e., 200 neurons). However, this type of computational embedded system (robot’s microcontroller) is found to be insufficient for simulating huge numbers of neurons in real time (as in the medial entorhinal cortex). It is considered that the results of this work provide an insight into achieving an enhanced embedded systems design for emulating and understanding mathematical neural network models to be used as biologically inspired navigation system for robots.

scholarly journals A geometric attractor mechanism for self-organization of entorhinal grid modules

eLife ◽  
2019 ◽  
Vol 8 ◽  
Author(s):  
Louis Kang ◽  
Vijay Balasubramanian
Keyword(s):  
Entorhinal Cortex ◽  
Network Models ◽  
Recurrent Inhibition ◽  
Modular Structure ◽  
Self Organization ◽  
Periodic Lattice ◽  
Grid Cells ◽  
Medial Entorhinal Cortex ◽  
Attractor Network ◽  
Discrete Values

Grid cells in the medial entorhinal cortex (MEC) respond when an animal occupies a periodic lattice of ‘grid fields’ in the environment. The grids are organized in modules with spatial periods, or scales, clustered around discrete values separated on average by ratios in the range 1.4–1.7. We propose a mechanism that produces this modular structure through dynamical self-organization in the MEC. In attractor network models of grid formation, the grid scale of a single module is set by the distance of recurrent inhibition between neurons. We show that the MEC forms a hierarchy of discrete modules if a smooth increase in inhibition distance along its dorso-ventral axis is accompanied by excitatory interactions along this axis. Moreover, constant scale ratios between successive modules arise through geometric relationships between triangular grids and have values that fall within the observed range. We discuss how interactions required by our model might be tested experimentally.

scholarly journals Thinking like a naturalist: enhancing computer vision of citizen science images by harnessing contextual data

2019 ◽  
Author(s):  
J. Christopher D. Terry ◽  
Helen E. Roy ◽  
Tom A. August
Keyword(s):  
Neural Network ◽  
Machine Learning ◽  
Computer Vision ◽  
Contextual Information ◽  
Network Models ◽  
List Type ◽  
British Isles ◽  
Automated Identification ◽  
Neural Network Models ◽  
Contextual Data

AbstractThe accurate identification of species in images submitted by citizen scientists is currently a bottleneck for many data uses. Machine learning tools offer the potential to provide rapid, objective and scalable species identification for the benefit of many aspects of ecological science. Currently, most approaches only make use of image pixel data for classification. However, an experienced naturalist would also use a wide variety of contextual information such as the location and date of recording.Here, we examine the automated identification of ladybird (Coccinellidae) records from the British Isles submitted to the UK Ladybird Survey, a volunteer-led mass participation recording scheme. Each image is associated with metadata; a date, location and recorder ID, which can be cross-referenced with other data sources to determine local weather at the time of recording, habitat types and the experience of the observer. We built multi-input neural network models that synthesise metadata and images to identify records to species level.We show that machine learning models can effectively harness contextual information to improve the interpretation of images. Against an image-only baseline of 48.2%, we observe a 9.1 percentage-point improvement in top-1 accuracy with a multi-input model compared to only a 3.6% increase when using an ensemble of image and metadata models. This suggests that contextual data is being used to interpret an image, beyond just providing a prior expectation. We show that our neural network models appear to be utilising similar pieces of evidence as human naturalists to make identifications.Metadata is a key tool for human naturalists. We show it can also be harnessed by computer vision systems. Contextualisation offers considerable extra information, particularly for challenging species, even within small and relatively homogeneous areas such as the British Isles. Although complex relationships between disparate sources of information can be profitably interpreted by simple neural network architectures, there is likely considerable room for further progress. Contextualising images has the potential to lead to a step change in the accuracy of automated identification tools, with considerable benefits for large scale verification of submitted records.

scholarly journals Chemtables: A Dataset for Semantic Classification on Tables in Chemical Patents

2021 ◽  
Author(s):  
Zenan Zhai ◽  
Christian Druckenbrodt ◽  
Camilo Thorne ◽  
Saber Akhondi ◽  
Dat Quoc Ngueyn ◽  
...  
Keyword(s):  
Language Processing ◽  
Pharmaceutical Research ◽  
Network Models ◽  
Relevant Information ◽  
Classification Task ◽  
Neural Network Models ◽  
Data Set ◽  
Content Type ◽  
Semantic Classification ◽  
Types Of Information

Abstract Chemical patents are a commonly used channel for disclosing novel compounds and reactions, and hence represent important resources for chemical and pharmaceutical research. Key chemical data in patents is often presented in tables. Both the number and the size of tables can be very large in patent documents. In addition, various types of information can be presented in tables in patents, including spectroscopic and physical data, or pharmacological use and effects of chemicals. Since images of Markush structures and merged cells are commonly used in these tables, their structure also shows substantial variation. This heterogeneity in content and structure of tables in chemical patents makes relevant information difficult to find. We therefore propose a new text mining task of automatically categorising tables in chemical patents based on their contents. Categorisation of tables based on the nature of their content can help to identify tables containing key information, improving the accessibility of information in patents that is highly relevant for new inventions. For developing and evaluating methods for the table classification task, we developed a new dataset, called ChemTables, which consists of 788 chemical patent tables with labels of their content type. We introduce this data set in detail. We further establish strong baselines for the table classification task in chemical patents by applying state-of-the-art neural network models developed for natural language processing, including TabNet, ResNet and Table-BERT on ChemTables. The best performing model, Table-BERT, achieves a performance of 88.66 micro F1 score on the table classification task. Availability: The ChemTables dataset is publicly available at http://dx.doi.org/10.17632/g7tjh7tbrj.1, subject to the CC BY NC 3.0 license. Code/models evaluated in this work are in a Github repository https://github.com/zenanz/ChemTables.

scholarly journals A geometric attractor mechanism for self-organization of entorhinal grid modules

2018 ◽  
Author(s):  
Louis Kang ◽  
Vijay Balasubramanian
Keyword(s):  
Entorhinal Cortex ◽  
Network Models ◽  
Recurrent Inhibition ◽  
Modular Structure ◽  
Self Organization ◽  
Periodic Lattice ◽  
Grid Cells ◽  
Medial Entorhinal Cortex ◽  
Attractor Network ◽  
Discrete Values

Grid cells in the medial entorhinal cortex (MEC) respond when an animal occupies a periodic lattice of “grid fields” in the environment. The grids are organized in modules with spatial periods, or scales, clustered around discrete values separated by ratios in the range 1.2–2.0. We propose a mechanism that produces this modular structure through dynamical self-organization in the MEC. In attractor network models of grid formation, the grid scale of a single module is set by the distance of recurrent inhibition between neurons. We show that the MEC forms a hierarchy of discrete modules if a smooth increase in inhibition distance along its dorso-ventral axis is accompanied by excitatory interactions along this axis. Moreover, constant scale ratios between successive modules arise through geometric relationships between triangular grids and have values that fall within the observed range. We discuss how interactions required by our model might be tested experimentally.

Dual Gated Graph Attention Networks with Dynamic Iterative Training for Cross-Lingual Entity Alignment

2022 ◽  
Vol 40 (3) ◽  
pp. 1-30
Author(s):  
Zhiwen Xie ◽  
Runjie Zhu ◽  
Kunsong Zhao ◽  
Jin Liu ◽  
Guangyou Zhou ◽  
...  
Keyword(s):  
Neural Network ◽  
Structural Information ◽  
Contextual Information ◽  
Network Models ◽  
Neural Network Models ◽  
Attention Networks ◽  
Individual Knowledge ◽  
Benchmark Datasets ◽  
Cross Lingual ◽  
Network Approaches

Cross-lingual entity alignment has attracted considerable attention in recent years. Past studies using conventional approaches to match entities share the common problem of missing important structural information beyond entities in the modeling process. This allows graph neural network models to step in. Most existing graph neural network approaches model individual knowledge graphs (KGs) separately with a small amount of pre-aligned entities served as anchors to connect different KG embedding spaces. However, this characteristic can cause several major problems, including performance restraint due to the insufficiency of available seed alignments and ignorance of pre-aligned links that are useful in contextual information in-between nodes. In this article, we propose DuGa-DIT, a dual gated graph attention network with dynamic iterative training, to address these problems in a unified model. The DuGa-DIT model captures neighborhood and cross-KG alignment features by using intra-KG attention and cross-KG attention layers. With the dynamic iterative process, we can dynamically update the cross-KG attention score matrices, which enables our model to capture more cross-KG information. We conduct extensive experiments on two benchmark datasets and a case study in cross-lingual personalized search. Our experimental results demonstrate that DuGa-DIT outperforms state-of-the-art methods.

scholarly journals ChemTables: A Dataset for Semantic Classification of Tables in Chemical Patents

2020 ◽  
Author(s):  
Zenan Zhai ◽  
Christian Druckenbrodt ◽  
Camilo Thorne ◽  
Saber A Akhondi ◽  
Dat Quoc Nguyen ◽  
...  
Keyword(s):  
Language Processing ◽  
Pharmaceutical Research ◽  
Network Models ◽  
Relevant Information ◽  
Classification Task ◽  
Neural Network Models ◽  
Data Set ◽  
Content Type ◽  
Semantic Classification ◽  
Types Of Information

Abstract Chemical patents are a commonly used channel for disclosing novel compounds and reactions, and hence represent important resources for chemical and pharmaceutical research. Key chemical data in patents is often presented in tables. Both the number and the size of tables can be very large in patent documents. In addition, various types of information can be presented in tables in patents, including spectroscopic and physical data, or pharmacological use and effects of chemicals. Since images of Markush structures and merged cells are commonly used in these tables, their structure also shows substantial variation. This heterogeneity in content and structure of tables in chemical patents makes relevant information difficult to find. We therefore propose a new text mining task of automatically categorising tables in chemical patents based on their contents. Categorisation of tables based on the nature of their content can help to identify tables containing key information, improving the accessibility of information in patents that is highly relevant for new inventions. For developing and evaluating methods for the table classification task, we developed a new dataset, called ChemTables, which consists of 7,886 chemical patent tables with labels of their content type. We introduce this data set in detail. We further establish strong baselines for the table classification task in chemical patents by applying state-of-the-art neural network models developed for natural language processing, including TabNet, ResNet and Table-BERT on ChemTables. The best performing model, Table-BERT, achieves a performance of 88.66 micro F1 score on the table classification task. Availability: A 10% sample of the ChemTables dataset has been made publicly available, subject to a data usage agreement.

scholarly journals Explaining heterogeneity in medial entorhinal cortex with task-driven neural networks

2021 ◽  
Author(s):  
Aran Nayebi ◽  
Alexander Attinger ◽  
Malcolm G. Campbell ◽  
Kiah Hardcastle ◽  
Isabel I.C. Low ◽  
...  
Keyword(s):  
Entorhinal Cortex ◽  
Network Architecture ◽  
Path Integration ◽  
Neuronal Response ◽  
Grid Cell ◽  
Cell Types ◽  
Neural Network Models ◽  
Medial Entorhinal Cortex ◽  
Wide Range ◽  
Response Profiles

Medial entorhinal cortex (MEC) supports a wide range of navigational and memory related behaviors. Well-known experimental results have revealed specialized cell types in MEC --- e.g. grid, border, and head-direction cells --- whose highly stereotypical response profiles are suggestive of the role they might play in supporting MEC functionality. However, the majority of MEC neurons do not exhibit stereotypical firing patterns. How should the response profiles of these more "heterogeneous" cells be described, and how do they contribute to behavior? In this work, we took a computational approach to addressing these questions. We first performed a statistical analysis that shows that heterogeneous MEC cells are just as reliable in their response patterns as the more stereotypical cell types, suggesting that they have a coherent functional role. Next, we evaluated a spectrum of candidate models in terms of their ability to describe the response profiles of both stereotypical and heterogeneous MEC cells. We found that recently developed task-optimized neural network models are substantially better than traditional grid cell-centric models at matching most MEC neuronal response profiles --- including those of grid cells themselves --- despite not being explicitly trained for this purpose. Specific choices of network architecture (such as gated nonlinearities and an explicit intermediate place cell representation) have an important effect on the ability of the model to generalize to novel scenarios, with the best of these models closely approaching the noise ceiling of the data itself. We then performed "in-silica" experiments on this model to address questions involving the relative functional relevance of various cell types, finding that heterogeneous cells are likely to be just as involved in downstream functional outcomes (such as path integration) as grid and border cells. Finally, inspired by recent data showing that, going beyond their spatial response selectivity, MEC cells are also responsive to non-spatial rewards, we introduce a new MEC model that performs reward-modulated path integration. We find that this unified model matches neural recordings across all variable-reward conditions. Taken together, our results point toward a conceptually principled goal-driven modeling approach for moving future experimental and computational efforts beyond overly-simplistic single-cell stereotypes.

scholarly journals DeepPPPred: An Ensemble of BERT, CNN, and RNN for Classifying Co-mentions of Proteins and Phenotypes

2020 ◽  
Author(s):  
Morteza Pourreza Shahri ◽  
Katrina Lyon ◽  
Julia Schearer ◽  
Indika Kahanda
Keyword(s):  
Network Models ◽  
Ensemble Classifier ◽  
Biomedical Literature ◽  
Supervised Machine Learning ◽  
Neural Network Models ◽  
Unstructured Text ◽  
Sentence Level ◽  
Machine Learning Approach ◽  
Human Proteins ◽  
Types Of Information

AbstractThe biomedical literature provides an extensive source of information in the form of unstructured text. One of the most important types of information hidden in biomedical literature is the relationships between human proteins and their phenotypes, which, due to the exponential growth of publications, can remain hidden. This provides a range of opportunities for the development of computational methods to extract the biomedical relationships from the unstructured text. In our previous work, we developed a supervised machine learning approach, called PPPred, for classifying the validity of a given sentence-level human protein-phenotype co-mention. In this work, we propose DeepPPPred, an ensemble classifier composed of PPPred and three deep neural network models: RNN, CNN, and BERT. Using an expanded gold-standard co-mention dataset, we demonstrate that the proposed ensemble method significantly outperforms its constituent components and provides a new state-of-the-art performance on classifying the co-mentions of human proteins and phenotype terms.

Sign in / Sign up

Export Citation Format

Share Document