Evolving a neural network using dyadic connections

Advances in Reinforcement Learning, Intech, 2011

“There exist many robots who faithfully execute given programs describing the way of image recognition, action planning, control and so forth. Can we call them intelligent robots?” In this chapter, the author who has had the above skepticism describes the possibility of the emergence of intelligence or higher functions by the combination of Reinforcement Learning (RL) and a Neural Network (NN), reviewing his works up to now.

BI 2018, 2018

This paper deals with Brain-Inspired robot controllers, based on a special kind of artificial neural structures that burn "dark" energy to promote the self-motivated initiation of behaviors. We exploit this ambient to train a virtual multi-joint robot, with many moving parts, muscles and sensors distributed through the robot body, interacting with elements that satisfy Newtonian laws. The robot faces a logical-mechanical challenge where a heavy, slippery ball, pressed against a wall has to be pushed up by means of coordinate muscles activation , where energy, timing and balancing conditions add noticeable technical complications. As in living brains our robots contains self-motivating neu-ral agents that consumes energy and function by themselves even without external stimulus. Networks that handle sensory and timing information are combined with agents to construct our controller. We prove that by using appropriate learning algorithms, the self-motivating capacity of agents provides the robot with powerful self-programming aptitudes, capable of solving the ball lifting problem in a quick, efficient way.

In this paper we introduce a novel mechanism for controlling autonomous mobile robots that is based on using Spiking Neural Networks (SNNs). SNNs are inspired by biological neurons that communicate using pulses or spikes. As SNNs have shown to be excellent control systems for biological organisms, they have the potential to produce good control systems for autonomous robots. In this paper we present the use and benefits of SNNs for mobile robot control. We also present an adaptive Genetic Algorithm (GA) to evolve the weights of the SNNs online using real robots. The adaptive GA using adaptive crossover and mutation converge in a small number of generations to solutions that allow the robots to complete the desired tasks. We have performed many experiments using real mobile robots to test the evolved SNNs in which the SNNs provided a good response.

Lecture Notes in Computer Science, 2001

An evolutionary algorithm for the creation of recurrent network structures is presented. The aim is to develop neural networks con- trolling the behaviour of miniature robots. Two neuro-modules are created separately using this evolutionary approach. The first neuro- module gives the agents the ability to move within an environment without colliding with obstacles. The second neuro-module provides the agents with

2011

Artifacts having to perform in the real world should be able to cope with uncertain situations and react quickly to changes in the environment. Living systems, starting from a pre-structured set of functions, develop competence to better adapt to the environment all life long, from childhood to maturity. This phenomenon is called growing up. A developmental intelligence for growing up robots must be able to generate autonomously representations for unknown knowledge and skills. Models of development can play an important role in specifying the minimal preferences, faculties, and processes needed for this skill to emerge. A perspective on the analysis of some adaptive learning processes originating from literature and experiments in natural world result in some suggestion intelligent system architecture modeling. Chapter 2 Representation of Neuro-Information and Knowledge

International Journal of Neural Systems, 2001

Setting up a neural network with a learning algorithm that determines how it can best operate is an efficient way to formulate control systems for many engineering applications, and is often much more feasible than direct programming. This paper examines three important aspects of this approach: the details of the cost function that is used with the gradient descent learning algorithm, how the resulting system depends on the initial pre-learning connection weights, and how the resulting system depends on the pattern of learning rates chosen for the different components of the system. We explore these issues by explicit simulations of a toy model that is a simplified abstraction of part of the human oculomotor control system. This allows us to compare our system with that produced by human evolution and development. We can then go on to consider how we might improve on the human system and apply what we have learnt to control systems that have no human analogue.

2023

During the first part of life, the brain develops while it learns through a process called synaptogenesis. The neurons, growing and interacting with each other, create synapses. However, eventually the brain prunes those synapses. While previous work focused on learning and pruning independently, in this work we propose a biologically plausible model that, thanks to a combination of Hebbian learning and pruning, aims to simulate the synaptogenesis process. In this way, while learning how to solve the task, the agent translates its experience into a particular network structure. Namely, the network structure builds itself during the execution of the task. We call this approach Self-building Neural Network (SBNN). We compare our proposed SBNN with traditional neural networks (NNs) over three classical control tasks from OpenAI. The results show that our model performs generally better than traditional NNs. Moreover, we observe that the performance decay while increasing the pruning rate is smaller in our model than with NNs. Finally, we perform a validation test, testing the models over tasks unseen during the learning phase. In this case, the results show that SBNNs can adapt to new tasks better than the traditional NNs, especially when over 80% of the weights are pruned. CCS CONCEPTS • Computing methodologies → Neural networks; Genetic algorithms; Artificial life; • Theory of computation → Models of learning.

Lecture Notes in Computer Science, 2010

The paper presents an architecture of a biologically inspired agent. Its physical body is described first. The agent's movement is directly controlled by Spiking Neural Network. To achieve this goal, the network is trained by a genetic algorithm. The agents move in a 3D physical environment. Their main goal is to effectively translocate themselves using a virtual body structure and muscles. This approach is inspired by a biological assumptions, where the neural network receives signals from sensors and directly controls the muscles. The application of Spiking Neural Network needs a suitable signal encoding method, which is also described. The system is flexible and it allows to create agents with various body structures and different neural controllers. Experiments presented in the paper refer to a simple snake-like creature. The effectiveness of controllers based on a standard threshold network and the spiking one are compared.

NASA/DoD Conference on Evolvable Hardware, 2003. Proceedings.

A major problem in artificial brain building is the automatic construction and training of multi-module systems of neural networks. For example, consider a biological human brain, which has millions of neural nets. If an artificial brain is to have similar complexity, it is unrealistic to require that the training data set for each neural net must be specified explicitly by a human, or that interconnections between evolved nets be performed manually. In this paper we present an original technique to solve this problem. A single large-scale task (too complex to be performed by a single neural net) is automatically split into simpler sub-tasks. A multi-module system of neural nets is then trained so that one of these sub-tasks is performed by each net. We present the results of an experiment using this novel techinque for pattern recognition.