Event Abstract

Back to Event

BiCAmon: Activity monitoring tool on 3D connectome structures for various cognitive architectures

Hirokazu Kiyomaru1*, Masahiko Osawa2 and Hiroshi Yamakawa3, 4
  • 1 Doshisha University, Japan
  • 2 Keio University, Japan
  • 3 Dwango ltd, Japan
  • 4 Whole Brain Architecture Initiative, Japan

1. Introduction The development of deep learning technology in recent years has accelerated the evolution of various cognitive architectures composed of artificial neural networks (ANNs) [1]. By using massive quantities of data, these technologies can learn the ability to solve various tasks. The ultimate goal of AI research is to build artificial general intelligence (AGI) that gains knowledge of various tasks and can flexibly combine this knowledge to solve more complex problems. However, currently, each AI only possesses specialized problem-solving abilities for specific tasks, and it cannot reach the level of AGI. Existent general intelligence is only the brains of organisms now, so it is effective to approach AGI by considering how a brain works. Also, neuroscience is steadily progressing, along with a clarified understanding of the entire brain's mesoscopic network construction (connectome). Therefore, if cognitive architecture can be designed on the model of the brain's mesoscopic network, it would be expected to approach AGI. These cognitive architectures inspired by the brain are often referred to as biologically-inspired cognitive architecture (BiCA) (c.f. http://bicasociety.org/). However, it is difficult to combine the entire cognitive architecture with the whole brain all at once. For example, if IT technicians attempted this work, it would require a deep and broad understanding of the field of neuroscience. Realistically, this is quite difficult. On the other hand, computational neuroscientists often construct more local neural network models with detailed functions; however, construction of a precise model for the entire brain is too grand. 2. BiCAmon (biologically-inspired cognitive architecture monitor) To attack this immense problem, we developed the Biologically-Inspired Cognitive Architecture Monitor (BiCAmon), a 3D visualization tool that monitors the internal activity status of various cognitive architectures [2] as an activity on the brain network. In the following, we will explain the functions of BiCAmon. Basic Functions: Structure Display Function: BiCAmon displays 3D information about nodes in every area of the brain. The information includes data of their locations, volume, and structures of connections between nodes. Exploration Function: The viewpoint can be moved to different locations and directions with intuitive operation. Monitoring Function: Dynamically displays the activity status of cognitive architecture’s functional units that have been equipped with particular nodes. Keyword Search Function: Nodes can be searched by entering search terms in the toolbox and clicking on the black button. Searching is done by matching the beginning of the search term; multiple searches are handled by the OR search. Also, detailed information of the search results are displayed at the lower left of the screen, and the node included in the results can be narrowed down by clicking on its X buttons in the detail information box. Half-Mode Switch Function: It is possible to hide or display a brain hemisphere by clicking the green button. This time, we used the mouse connectome [3] at a mesoscopic level. We utilized the Allen SDK [4] for the information of coordinates of every area in the brain. The cognitive architecture in the diagrams was constructed using Nengo [5], but we also confirmed the connections in BriCA [6] and Brain Simulator [7]. We considered the ease of using BiCAmon when we developed cognitive architecture in various calculation platforms. Thus, communication between cognitive architecture and BiCAmon uses the socket communication, and the program is related to displaying operates on Web browsers. The composition of BiCAmon is shown in Figure 1. Cognitive architecture is the monitoring target. It is assumed that this will be written according to a platform that a user chooses. The “BiCAmon server” operates in environments that can execute Python. On WebGL support browsers, the “BiCAmon client” script is executed and the drawing performed. 3. AGI development with a BiCAmon Using BiCAmon allows cognitive architecture’s activity status visualization and facilitates understanding of that internal state. However, more important is ultimately making BiCA capable of supporting a mesoscopic network construction in the brain through development processes such as those described below. 1． First, create a primitive cognitive architecture that roughly corresponds with the brain (sensory/motor/decision-making), specify functional activity that supports it, and display it on BiCAmon. 2． Then, enable dispersing processes like the following. (1) IT technicians can develop and improve more detailed cognitive architectures by focusing on specific local networks. This reduces the requirement of neuroscientific expertise for a broad area. It is also possible that people will come up with new ideas while considering the association with the brain. (2) Neuroscientists will be able to hold functional hypotheses about nervous activity by caring about cognitive architectures. The significant hypothesis will be validated through experiments. Whichever of these two processes is employed, BiCAmon is a useful tool for comparing between operations of cognitive architecture and activities of the nervous system. 4. Related Works To date, the following kinds of tools have been developed. BrainNet Viewer [8]　can help researchers visualize structural and functional connectivity patterns from different levels of work on MATLAB. However, it cannot dynamically display cognitive architecture activity status. Brain X3 [9] is a human brain activity simulator with a display tool that shows brain activity in 3D. However, its goal is not to display the activity state of the relevant cognitive architecture in an accessible manner, and at present it has not been opened to the public (April, 18, 2016). 5. Conclusion The idea that neuroscientific expertise can be of use in AI research is not a novel one. For example, there has been research where the activity state of the representative cognitive architecture “ACT-R” [10] supported FMRI experiment results. Yet now, a functional brain map is finally being constructed for the connectome at a mesoscopic level. Such research is also expected to progress. Against such a background, a “BiCAmon” cognitive architecture monitoring tool is introduced. Using tools like BiCAmon to access the connection information of the connectome surely allows many researchers to accelerate the development of cognitive architecture. Thus, such tools encourage cross-disciplinary collaborations between the fields of neuroscience and AI. Furthermore, the source is already publically available through Github and can be used for free [11].

Figure 1
Image

Acknowledgements

We are very grateful to the many members of the nonprofit Whole Brain Architecture Initiative for their critiques and advice during the development of this software.

References

1. Yann L, Bengio Y, Hinton J. Deep learning. Nature (2015) 521:7553, 436-444. doi:10.1038/nature14539
2. Goertzel B, et.al. A world survey of artificial brain projects, Part II: Biologically inspired cognitive architectures. Neurocomputing (2010) 74, 30–49. doi:10.1016/j.neucom.2010.08.012
3. Oh SW, et al. A mesoscale connectome of the mouse brain. Nature (2014) 508, 207–214. doi:10.1038/nature13186
4. Available from: http://alleninstitute.github.io/AllenSDK/
5. Eliasmith C. How to Build a Brain A Neural Architecture for Biological Cognition.(2013), Oxford Series on Cognitive Models and Architectures.
6. Takahashi K, Itaya K, et.al. A Generic Software Platform for Brain-inspired Cognitive Computing. Procedia Computer Science, (2015) 71, 31–37. doi:10.1016/j.procs.2015.12.185
7. Available from: http://www.goodai.com/#!brain-simulator/c81c
8. Available from: http://www.nitrc.org/projects/bnv/
9. Available from: http://www.brainx3.com/
10. Available from: http://act-r.psy.cmu.edu/
11. Available from: https://github.com/kiyomaro927/bicamon

Keywords: mouse models, visualization software, Nengo, Brica, connectome mapping

Conference: Neuroinformatics 2016, Reading, United Kingdom, 3 Sep - 4 Sep, 2016.

Presentation Type: Demo

Topic: Visualization

Citation: Kiyomaru H, Osawa M and Yamakawa H (2016). BiCAmon: Activity monitoring tool on 3D connectome structures for various cognitive architectures. Front. Neuroinform. Conference Abstract: Neuroinformatics 2016. doi: 10.3389/conf.fninf.2016.20.00025

Copyright: The abstracts in this collection have not been subject to any Frontiers peer review or checks, and are not endorsed by Frontiers. They are made available through the Frontiers publishing platform as a service to conference organizers and presenters.

The copyright in the individual abstracts is owned by the author of each abstract or his/her employer unless otherwise stated.

Each abstract, as well as the collection of abstracts, are published under a Creative Commons CC-BY 4.0 (attribution) licence (https://creativecommons.org/licenses/by/4.0/) and may thus be reproduced, translated, adapted and be the subject of derivative works provided the authors and Frontiers are attributed.

For Frontiers’ terms and conditions please see https://www.frontiersin.org/legal/terms-and-conditions.

Received: 27 Apr 2016; Published Online: 18 Jul 2016.

* Correspondence: Mr. Hirokazu Kiyomaru, Doshisha University, Kyoto, Japan, kiyomaru.tennis927@gmail.com