In the Palaces of Memory: How We Build the Worlds Inside Our Heads

Hopfield student, Terry Sejnowski created an invention that could learn to read. The network had a input layer that read the letters, a middle layer which generated the phonemes, and an output layer. Each neural was connect to eighteen thousand synapses. The weights on the synapses were adjusted according to the back-progragation error routine. NetTalk went through the babbling stage, after a half day pronounced a thousand words, and by the end of the week, it pronounced twenty thousand words. "Unlike an A.I. programmer, Sejnowski didn't start with symbols as the primitive units, the letters and phonemes that would be manipulated according to rules." Rules like long e, in she and he were generalized by the strength of units in the middle layer. Minsky: 1. The biggest problem with neural networks is scaling: "There is no reason to thing that a small network capable of learning a fairly easy task could be scaled up to solve the kinds of harder problems that brains do." 2. Many of the new NN take tens of thousands of trails to learn a simple skill, like recognizing a small number of objects. 3. If NN scale exponentially, so that multiplying the size of the body of material to be learned by a factor of n, meant raising the processing time to the nth power. Learning something difficult might take longer than the universe would exist. 4. Perceptron was not written to kill NN. The field had already died. Minsky and Papert were explaining why. Minsky listen to NetTalk and said he couldn't understand many of the words it said. 5. Minsky embraced both the NN and Symbolic agent programming model of the brain, "society of the mind". The agents interact in a complex cooperative hierarchial system where agents communicated with other agents simulating intelligent behavior. High level agents control low level agents. Conditional, if-then-do rules active agents to return a positive or negative result. The world of sensation was defined as: sensation -> reception -> recognition -> cognition. The brain received an input from one of the five senses. The sensory input activates various processes from connected agents. A polyneme signals different agencies (color, shape, or texture agencies) too turn on process in their agencies Agents interaction can be diagrammed into deterministic asynchronous finites states and used by robots. "How do you make a management structure in which some of them are good at learning how to manage how other learn?" That is what the brain is. The brain has 300 kinds of neural networks, and some specialize for the controlling the input and outputs of others. Other NN specialize in retaining memories. 6. Perceptrons is about how you'd use a lot of different types of neural nets to make something smart. <br /> <br />Nestor Inc, is a leader in commercial NN technology. Nestor, Inc. is a leading provider of advanced intelligent traffic management solutions. Nestor is creating intelligent scan radars and laser b

"Whenever you read a book or have a conversation, the experience causes physical changes in your brain. In a matter of seconds, new circuits are formed, memories that can change forever the way you think about the world. [...] I'll never forgive David Lynch for his movie Erasorhead." The first two pages of In the Palaces of Memory introduce remembrance as an act not only of acquisition but of self-exposure. Memories make it possible for us to function; they may also lodge themselves in us "like a shard of glass healed inside a wound," never to be expelled. Some memories are desired and some become a part of the structure of our minds against our will.Memory's palaces, though, may be as much the edifices the theorists construct as they are the ones inside our heads. This slim volume is not only an analysis of the way memory works but also an exposé of the way memory morphs depending on who's studying it. The underlying question, as in so much of Johnson's work, is really "how a theory matches up with some kind of real world," and what the world (in this case the brain) looks like from the point of view of the brain-children, scientific or philosophical, that purport to explain it. In this book the "unruly, creative art of theory-building" occupies center stage with memory.What is remarkable about Johnson's writing is the uninhibited intimacy he seems to have with his subjects and with us, his readers, so that we can feel ourselves to be as close to the Thing, whatever it is, as he is. Johnson has granted me the delightful illusion of being nose to nose with a neuron, with Gell-Mann, with Planck's constant -- almost as though the experience were unmediated by an author. The man's a master story teller. But what comes across is also -- and here's the clincher -- a profound sense of amusement. If I'm not mistaken George Johnson is given to quiet chuckles in the dark over theoreticians and theorems. He infuses his translations of science in the making with a persistent, ironic-affectionate grin.How can we resist.

"One of the last great mysteries is the one we carry inside our heads: how we remember, what we remember, why we remember. "In the Palaces of Memory" is a rich and lucid guide to this entangled and enchanting domain." -- James Gleick "Johnson has written a fascinating book, which perhaps throws as much light on how science is done and on the scientists who do it as any book since "The Double Helix" -- Stuart Sutherland, Nature "Johnson has achieved a rare blend of scientific and literary sophistication. Faithful to its complexities and controversies, the book is a fully dimensional portrait, a hologram of the field." -- Richard Mark Friedhoff, USA Today