Hugo de Garis

A world survey of artificial brain projects, Part II: Biologically inspired cognitive architectures

A number of leading cognitive architectures that are inspired by the human brain, at various levels of granularity, are reviewed and compared, with special attention paid to the way their internal structures and dynamics map onto neural processes. Four categories of Biologically Inspired Cognitive Architectures (BICAs) are considered, with multiple examples of each category briefly reviewed, and selected examples discussed in more depth: primarily symbolic architectures (e.g. ACT-R), emergentist architectures (e.g. DeSTIN), developmental robotics architectures (e.g. IM-CLEVER), and our central focus, hybrid architectures (e.g. LIDA, CLARION, 4D/RCS, DUAL, MicroPsi, and OpenCog). Given the state of the art in BICA, it is not yet possible to tell whether emulating the brain on the architectural level is going to be enough to allow rough emulation of brain function; and given the state of the art in neuroscience, it is not yet possible to connect BICAs with large-scale brain simulations in a thoroughgoing way. However, it is nonetheless possible to draw reasonably close function connections between various components of various BICAs and various brain regions and dynamics, and as both BICAs and brain simulations mature, these connections should become richer and may extend further into the domain of internal dynamics as well as overall behavior.

Artificial Brains: An Evolved Neural Net Module Approach

This book explains how the author is building Chinas first artificial brain, using an evolved neural net module approach. These modules are evolved in special hardware very fast, each with its own little job. They are downloaded one by one into the memory of a supercomputer, and connected up according to the designs of human BAs (Brain Architects) to build artificial brains, which then control the hundreds of robots behaviors. These artificial brains contain thousands of pattern recognition circuits. This approach is expected to produce artificial brains with several 10,000s of evolved neural net modules. The robots will also be given language abilities for conversing with humans. Artificial brains could possibly be the missing piece of the puzzle that will make home robots more genuinely intelligent and useful. It is hence likely that by 2030, artificial brains will be one of the biggest industries in the world.

CuParcone A High-Performance Evolvable Neural Network Model

An algorithm for evolving recurrent neural network via the genetic algorithm was implemented on the CUDA, resulting in a system called CuParcone (CUDA based Partially Connected Neural Evolutionary). Run on a Nvidia Tesla “GPU supercomputer, ” CuParcone achieves a performance increase of 323 times in face gender recognition compared to the comparable Parcone algorithm on a state-of-the-art, commodity single-processor server. The accuracy on this task does not decrease in moving from Parcone to CuParcone, and is comparable to the published results of other algorithms.

Approach to controlling robot by artificial brain based on parallel evolutionary neural network

A novel approach to building an intelligent robot, which can be expected to show many intelligent behaviors as people do, is proposed. This approach exploits a computer device called Tesla S1070, a personal supercomputer, to control a humanoid robot named NAO. The Tesla S1070 acts as a human head, in which thousands of artificial neural network, the basis of an artificial brain (AB), evolve in a parallel way and act as a human brain. The CUDA (Computing Unified Device Architecture) technology is employed to solve the key problem, i.e., how to improve the evolution speed of thousands of neural network. The general architecture of this approach, the reason why the Tesla S1070 is selected, and how the data structure of neural net modules should be modified to meet the demands of CUDA are presented in detail. This approach provides a method both for controlling a robot by the AB, and for demonstrating the power of the AB through a robot.

A simple robot paths planning based on Quadtree

Based on traditional methods, according with practical application conditions, and the special structure of Quadtree, this paper gives a practical method of paths planning for the further expansion of Quadtree in the field of paths planning, except for segmentation obstacles. The experimental and simulation results demonstrate that the algorithm described in the text could be able to find all the relevant paths correctly, and make the better choice rapidly.

Editorial: Guest Editorial: Special issue on artificial brains

Now, at the start of the 21st century, there are strong reasons to believe the time has finally come for the creation of powerful artificial general intelligences, including ‘‘artificial brains’’ that incorporate key aspects of human brain structure and dynamics within digital computer programs. In the last decades, computer hardware has advanced tremendously, due to Moore’s Law and associated phenomena, which bring the computing power of available hardware closer and closer to that of the human brain. Our knowledge of neuroscience and cognitive science has also increased dramatically, due largely to advances in brain measurement instrumentation (which have been driven by advances in computer technology, applied physics and other areas). Additionally, the arsenal of AI and neurosimulation algorithms at our disposal has improved steadily in both strength and diversity. The purpose of this special issue is to review some of the recent work that has been done, leveraging these technological advances toward the design and creation of artificial brains. The first two papers, authored by the special issue editors and their colleagues, give a bird’s eye view of the artificial brains field, dividing it into two subdomains. First, large-scale brain simulation: the quest to simulate the internal dynamics and structure of the brain as well as the overall architecture and behavior. And second, biologically inspired cognitive architecture (BICA): the creation of software systems with high-level architectures and processes analogous to those in the brain, but in some cases utilizing data structures and algorithms with significant conceptual differences from their neural analogues. These papers review a sampling of the leading projects in each of these two subdomains, setting the stage for the following papers, which present more detailed technical contributions to the artificial brains field. The initial review papers reveal a striking diversity of aims as well as technical approaches in the artificial brains area, and this diversity is also reflected in the following more technical papers. Binding together the diverse concepts and systems, however, is a common spirit and methodology: to figure out, by looking at specific intelligent capabilities, sensible levels at which to model the human brain’s functionality and then create its digital analogues. This is a subtle matter and there need not be a single right answer, for any particular capability or in general. It may well be possible to emulate the brain’s capability at, say, audition or visual recognition or language generation – or holistic general intelligence – in multiple ways, some involving simulation at a detailed biological level (e.g. spiking neural networks) and others involving more qualitative high-level emulation. Following the introductory papers we have John Taylor’s contribution, which presents a specific neural net based artificial brain architecture, and then discusses the potential implications

Approach to Partially Connected Neural Evolutionary Model with Its Application to Image Recognition

To explore the method for the building of artificial brain, by combing neural network and genetic algorithm, Parcone model (PARtially COnnected Neural Evolutionary model) was proposed and represented, especially for its partially connected structure and evolution algorithm. Comparing with fully connected model, Parcone model can substantially decrease computing amount, while remain strong classification capability. A series of image recognition experiments (including arrow detection, face detection, and facial sex detection) showed the recognition power and effectiveness of the Parcone model. Since the Parcon model had great potential power, it might be expected to improve it to become the basis for the construction of Chinas first artificial brain.

GPU based Partially Connected Neural Evolutionary network and its application on gender recognition with face images

An algorithm for evolving neural network via the genetic algorithm based on GPU parallel architecture was implemented on the CUDA, resulting in a system called CuParcone (CUDA based Partially Connected Neural Evolutionary) and was used on gender face recognition. By using the powerful ability of GPU parallel computing, CuParcone achieves a performance increase about 323 times than Parcone algorithm, which runs on a single-processor. With this new model, a gender recognition experiment was made on 530 face images (265 females and 265 males from Color FERET database), including not only frontal faces but also the faces rotated from −40°∼40° in the direction of horizontal, and achieved the accuracy rate of 90.84%.