We are interested in developing computational models of embodied intelligence by developing biologically plausible learning algorithms for signal processing in self-organized learning networks of processing components (artificial neurons).
The aim is to develop self-organizing architectures for future computers, electronic gadgets, robots and intelligent machines. We do so using motivated learning approach, based on internal mechanism of goal creation, to avoid externally applied environmental pain signals.
Embodied intelligence emerges gradually in real-time interaction with environment within the constraints of its processing power, sensory inputs, motor outputs and environmental changes. In our study we aim at developing brain-like processing and learning that may lead to creation of intelligent machines. We have proposed to combine sensory and motor activities with goal creation and goal driven behavior. We develop models for sparse pattern representation, attention driven cognition, spatio-temporal memories, and goal driven action planning and execution. We aim at developing neuronal foundation for knowledge, thinking, deliberation, reasoning, and emotion.
Embodied Intelligence research is inspired by models of brain and its operation. The research involves a broad range of areas concerning learning and neural systems that include building sensory representations for vision and image processing, active vision with sensory motor coordination; audition, speech and language understanding, text based learning; invariant pattern recognition; cognitive information processing; self-organization and emergence on neural structures; associative learning, building short and long-term memories; unsupervised, supervised and reinforcement learning, motivation, goal creation, and goal driven behavior; attention and attention based learning; adaptive sensory-motor planning, verification and skill building, autonomous control in robotics and robotic applications; self navigation, spatial orientation and perception; self-awareness and consciousness; and the mathematical and computational intelligence methods to support structured modeling of embodied intelligence and its applications.
This research is biologically inspired and is supported by discoveries in neuroscience, neurobiology, and psychological study of human brain, its models, models of behavior, including disease based disorders (dementia, Parkinson’s disease, attention deficit disorder, schizophrenia, and depression). It considers existing models of human brain and the functional organization of brain regions such as the visual, auditory, temporal, parietal, motor, and prefrontal cortex, hippocampus, hypothalamus, cerebellum, superior colliculus, basal ganglia, thalamus, retina, and spinal cord. We take our inspiration from neurophysiology and cognitive neuroscience for model building and experimental verifications of our models.
Our approach is to build and study models of intelligence in large sparsely connected self-organizing structures of simple processing elements capable of associative memory, spatio-temporal learning, goal creation, sensory and motor representation building in interaction with the environment. We study properties of such networks, representation building and categorization, memory self-organization, short and long-term memory, cognitive information processing, as well as sensory-motor anticipation and planning.