NNRG Research Areas

Research Areas

Also show archived

Applications	Much of our work on applications involves using neuroevolution in real-world domains, but also includes reinforcement learning in robotics, packet routing, and satellite communication, unsupervised learning for pattern recognition and visualization, and supervised learning for intrusion detection and process control.
Cognitive Science	Our research in Cognitive Science aims to better understand human development and learning through computational modeling. We have active projects in Natural Language Processing, Schema Learning, Episodic Memory, Perceptual Grouping, Face Detection, and Hi-Level Concept Acquisition. Our models are applied to applications in neuroscience, biology, linguistics, psychology and education. Details of our projects in each of these areas can be found in the links below.
Computational Neuroscience	A computational model is a complete description of how a neural system functions, and in that sense the ultimate specification of neuroscience theory. The models are constrained by and validated with existing experimental data, and then used to generate predictions for further biological experiments. Our work in this area focuses on understanding the visual cortex, episodic and associative memory, aphasic and dyslexic impairments of the lexical system, and language impairments in schizophrenia.
Concept and Schema Learning	An agent can truly understand the meaning of its knowledge structures, and utilize them most effectively, only if that knowledge is grounded on sensorimotor interactions with the world. We aim at building systems that learn such grounded representations, from basic causal concepts to high-level schemas of visual scenes.
Episodic Memory	Episodic memory is the record of events that the individual experiences throughout his/her/its life. Recent neurobiological results suggest that the events are stored temporarily in the hippocampus and related areas immediately as they happen, and over days and weeks, transferred to other locations (most likely in the neocortex) for permanent storage. Our research in episodic memory started with the long-term memory component for the DISCERN story processing system. More recently, we have focused on the episodic memory system of the brain, trying to understand the structure, capacity and memory effects through computational modeling and mathematical analysis.
Evolutionary Computation	Our research in this area focuses primarily on evolving neural networks, or Neuroevolution, but also includes work in probabilistic model-based algorithms and particle swarming.
Game Playing
Natural Language Processing	Our research in Natural Language Processing aims at bridging the gap between subsymbolic representations and complex high-level behavior. The models are based on subsymbolic mechanisms but aim at explaining how people learn word meanings, organize their lexicon, understand sentences and stories, and answer questions about them. An important aspect of this research is also understanding linguistic disorders, and most recently, the evolution of language.
Neuroevolution	In difficult real-world learning tasks such as controlling robots, playing games, or pursuing or evading an enemy, there are no direct targets that would specify the correct action for every situation. In such problems optimal behavior must be learned by exploring different actions and assigning credit for good decisions based on sparse reinforcement feedback. Our research in this area focuses on methods for evolving artificial neural networks with genetic algorithms, an Evolutionary Reinforcement Learning method called neuroevolution. Compared to standard Reinforcement Learning, neuroevolution is often more robust against noisy and incomplete input, and it allows representing continuous states and actions very naturally. Our methods include utilizing subpopulations, population statistics, knowledge embedded within the population(s) or training regimen, and evolving network structure. Much of this research involves comparisons of neuroevolution to traditional methods in several benchmark tasks such as pole balancing and mobile robot control. In addition to the standard benchmark tasks our research in this area focuses on difficult and unusual tasks such as robot control, coordination in multi-agent systems, game playing, resource optimization, music generation, theorem proving, and modeling language evolution. This research is supported in part by the National Science Foundation under grant IIS-0083776 (and previously under IRI-9504317) and the TexasHigher Education Coordinating Board under grant ARP-003658-476-2001.
Other Areas	This category includes papers published by neural network group members that are not in our traditional research areas.
Reinforcement Learning	Reinforcement Learning tasks are learning problems where the desired behavior is not known; only sparse feedback on how well the agent is doing is provided. Reinforcement Learning techniques include the value-function and policy iteration methods on the. Our research using this approach includes real-world applications of packet routing and satellite communication, as described below. See our Neuroevolution area for a different approach to learning with sparse feedback.
Robotics	Our work in robotics focuses on learning hierarchies of sensorimotor schemas through unsupervised and reinforcement learning, and on developing intelligent controllers through neuroevolution.
Self-Organization	Our work in this area includes extending the Self-Organizing Map architecture (SOM; Kohonen, 1982; 1997; von der Malsburg, 1975) with lateral connections, hierarchies, sequential inputs, and growing network structures. This work has been done mainly with cognitive science and computational neuroscience applications in mind, as described in the visual cortex, concept and schema learning, episodic memory, and natural language processing pages. We have also applied such maps to character recognition, visualizing high-dimensional data, modeling multi-sensory integration, and speech recognition, as described below.
Visual Cortex	visual cortex