There seems to be increasing evidence that sleeping and dreaming are important for learning because they are facilitating "unlearning" of detailed information that is not relevant for the learning goals and that is better forgotten in favor of task relevant information. Stickgold et al. tested this hypothesis with subject who were asked to learn the computer game tetris:
"People in both groups reported that, as they fell asleep, they dreamed about images of blocks falling and rotating, as they do on the computer screen when the game is in progress. They did not actually dream about the game itself.
The amnesia patients did not remember playing the game and they did not ever improve, unlike the volunteers with normal memory. Three of them did report the strange dreams, however. "
"The researchers found that people who have just learned to play Tetris have vivid images of the game pieces floating before their eyes as they fall asleep, a phenomenon the researchers say is critical for building memories. Much more surprisingly, the team also found that the images appear to people with amnesia who have played the game--even though they have no recollection of having done so."
2. Learning Chaotic Attractors By Neural Networks, Neural Comp.
Abstract: An algorithm is introduced that trains a neural network to identify chaotic dynamics from a single measured time series. During training, the algorithm learns to short-term predict the time series. At the same time a criterion, developed by Diks, van Zwet, Takens, and de Goede (1996) is monitored that tests the hypothesis that the reconstructed attractors of model-generated and measured data are the same. Training is stopped when the prediction error is low and the model passes this test. Two other features of the algorithm are (1) the way the state of the system, consisting of delays from the time series, has its dimension reduced by weighted principal component analysis data reduction, and (2) the user-adjustable prediction horizon obtained by "error propagationpartially propagating prediction errors to the next time step. The algorithm is first applied to data from an experimental-driven chaotic pendulum, of which two of the three state variables are known. This is a comprehensive example that shows how well the Diks test can distinguish between slightly different attractors. Second, the algorithm is applied to the same problem, but now one of the two known state variables is ignored. Finally, we present a model for the laser data from the Santa Fe time-series competition (set A). It is the first model for these data that is not only useful for short-term predictions but also generates time series with similar chaotic characteristics as the measured data.
3. Swarms of Robots Solve Complex Problems, Business Week Online
Agent based modeling is an increasingly popular way of solving a number of complex problems in different fields. Swarms of simulated agents follow their local rules and interact with each others while searching for a solution to a computational problems. Dedicated software environments like the Swarm allow the user to use high level tools and libraries to set up the simulations. While all of these activities are happening in CyberSpace and the agents are made of software only, researchers at Sandia National Laboratories in New Mexico have implemented some of these artificial life ideas into hardware: the agents are actually little robots that autonomously navigate in difficult but real terrains and cooperate to solve real problems of military and civilian applications. Among them are clearing of mine fields and other operations in dangerous environments or for instance the search for victims of avalanche accidents (Business week asks: "But Will They Bring A Little Keg Of Brandy?").
The autonomy and communication among the robots would indicate a major step forward in robotics applications that in the past depended on remote operators that basically had to individually radio-control each robot. The next generation robot would be smart enough to make for example many navigational decisions only based on information about the goal, the current location (determined by global positioning systems), the observed environment and perhaps input from other robots in the swarm.