Revisiting the XOR problem: a neurorobotic implementation
| Autor: | Frédéric Thériault, André Cyr, Sylvain Chartier |
|---|---|
| Rok vydání: | 2019 |
| Předmět: |
Spiking neural network
0209 industrial biotechnology Computer science business.industry Cognition 02 engineering and technology Task (project management) 020901 industrial engineering & automation Action (philosophy) Artificial Intelligence 0202 electrical engineering electronic engineering information engineering Robot Embedding 020201 artificial intelligence & image processing Operant conditioning Artificial intelligence business Software |
| Zdroj: | Neural Computing and Applications. 32:9965-9973 |
| ISSN: | 1433-3058 0941-0643 |
| Popis: | The exclusive-or (XOR) classification task still represents a challenge in the study of cognition since the precise neural circuit sustaining the general ability to learn nonlinear problems remains to be discovered in natural organisms. As such, this paper focuses on a neurorobotic application embedding a specific spiking neural network built to solve these types of tasks. This experiment proposes a 2-bit task (XOR) with visual compound binary images acting as inputs and a left/right action for the output. The robot learns to solve it in both virtual and real environments from an operant conditioning procedure. Furthermore, the robot also adapts its behavior from learning all other simpler associative rules, even when switching them at runtime. Finally, this study explores the impact on the neural architecture, when passing from a 2-bit to a 3-bit task. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|
Full text from SpringerLink