THE HUMAN BRAIN. observations and foundations

Transcription

1 THE HUMAN BRAIN observations and foundations

2 brains versus computers a typical brain contains something like 100 billion miniscule cells called neurons estimates go from about 50 billion to as many as 500 billion each neuron is made up of a cell body with a number of connections coming off it: numerous dendrites (for information toward the cell body) and a single axon (carrying information away). the body s size is about 10 µm. neurons make up only 10 percent of all the cells in the brain; the rest are glial cells, or neuroglia, that support and protect the neurons and feed them with energy that allows them to grow and function inside a processor the transistor, a tiny switching device (of about 1 nm), is the counterpart to a brain cell the latest single-chip processors contain over 2 billion transistors; even a basic microprocessor has about 50 million transistors, all packed onto an integrated circuit of just 25mm 2

3 wires versus neurons chips are wired in relatively simple chains that process sequences of instructions each transistor is connected to maybe few other transistors in basic arrangements to form gates, modules, supermodules, etc neurons in a brain are densely interconnected in complex, parallel ways each one is connected up to 10,000 others computers are designed for processing and storing vast amounts of meaningless data according to precise instructions (programs) brains, on the other hand, learn slowly, by a more roundabout method, often taking months or years to make complete sense of something but, unlike processors, brains seem to spontaneously put information together in astounding new

4 artificial neural networks a neural network is to simulate interconnected brain cells (copy in a simplified but reasonably faithful way) inside a computer without a program it to learn explicitly: it seems to learn all by itself so you can get it to learn things, recognize patterns, and make decisions in a humanlike way. so, neural networks are made by programming ordinary computers, working in a very traditional fashion with connected logic gates, in a serial way, to behave as though they are built from interconnected brain cells working in parallel computer simulations are just implementations ofcollections of algebraic variables and mathematical equations linking them together artificial neural networks (or ANNs) are also referred to by names like connectionist machines (the field is also called connectionism), parallel distributed processors (PDP), thinking machines, and so on

5 layered neural networks a typical neural network has anything from a few dozen to maybe even millions of artificial neurons arranged in layers, each of which connects to the layers on either side input units, are designed to receive various forms of signals from the outside world output units sit on the opposite side of the network and signal how it responds to the information it has learned in between the input units and output units are one or more layers of hidden units most neural networks are fully connected: each hidden unit is connected to every unit in the layers either side the connections between one unit and another are represented by a number called a weight, which can be either positive (if one unit excites another) or negative (if one unit suppresses or inhibits another) the higher the absolute value of the weight, the more influence one unit has on another

6 flows in neural networls feedforward flows: when operating normally (after being trained), patterns of information are fed into the network via the input units, which trigger the layers of hidden units, and these in turn arrive at the output units each unit receives inputs from the units to its left, and the inputs are multiplied by the weights of the connections they travel along every unit adds up all the input intenisties and if the sum exceeds a certain threshold, the unit "fires" and provides input to units on its right to learn, there has to be feedback, typically by a feedback process called backpropagation (sometimes abbreviated as "backprop"). this involves comparing the output a network produces with the output it was meant to produce, and using the difference between them to modify the weights of the connections in time, backpropagation causes the network to learn, reducing the difference between actual and intended output to the point where the two exactly coincide

7 backpropagation once the network has been trained with enough learning examples, it responds to entirely new set of inputs for example, suppose you've been teaching a network by showing it lots of pictures of chairs and tables, represented in some appropriate way it can understand, and telling it whether each one is a chair or a table. that doesn't mean to say a neural network can just "look" at pieces of furniture and instantly respond to them in meaningful ways consider the example we've just given: the network is not actually looking at pieces of furniture. The inputs to a network are essentially binary numbers: each input unit is either switched on or switched off. So if you had five input units, you could feed in information about five different characteristics of different chairs using binary (yes/no) answers. The questions might be 1. Does it have a back? 2. Does it have a top? 3. Does it have soft upholstery? 4. Can you sit on it comfortably for long periods of time? 5. Can you put lots of things on top of it? A typical chair would then present as Yes, No, Yes, Yes, No or in binary, while a typical table might be No, Yes, No, No, Yes or So, during the learning phase, the network is simply looking at lots of numbers like and and learning that some mean chair (which might be an output of 1) while others mean table (an output of 0).

8 competition artificial intelligence neural networks where is time? feedback? hierarchy? auto-associative memory retrieving data upon presenting part from that data consistent neurological anatomy six -layer neocortex columnar organization dedicated areas (by genetics) accepts some observation Mountcastle's common algorithm Felleman and Van Essen's "hierarchical" structure

9 the human brain: foundations Vernon Mountcastle: one region of the cortex looks slightly different from another, because of what it is connected to; there is a common algorithm that is performed by all cortical regions. Daniel Felleman, David Van Essen: they have created a detailed map of the monkey cortex global brain architecture (hierarchical structure)

10 Vernon B. Mountcastle 1950 s: columnar functional organization neurons passing each other at more than a half mm from each other do not have overlapping sensory receptive fields cortical columns through 6 layers: connections up and down the column are dramatically denser than the ones that spread from side to side genes specify how the regions of the cortex are connected, which is very specific for function and species, but the cortical tissue is doing the same thing everywhere. 6-layer neocortex appears to be a distinguishing feature of mammals: it has been found in the brains of all mammals but not in any other animals. in avians there are clear examples of cognitive processes that are thought to be neocortical in nature, despite the lack of the distinctive six-layer structure.

11 brain map of a monkey Felleman and Van Essen

12 Modified by T. Serre from Ungerleider and Haxby, and then copied by me.

13 Modified by T. Serre from Ungerleider and Haxby, and then copied by me.

14 Modified by T. Serre from Ungerleider and Haxby, and then copied by me.

15 identifying different visual areas connectivity analysis relies on finding a characteristic pattern of inputs and outputs for each cortical area architectonics relies on finding a distinctive structure topographic organization relies on an orderly mapping of the visual field in each area, as revealed physiologically or anatomically

16 the adjacency matrix of (A) cat and (B) macaque. for the cat cortical network, the colors represent the following communities: (i) black: cognitive, (ii) blue: visual, (iii) green: auditory, and (iv) red: sensory system. for the macaque, (i) black: memory, (ii) blue: visual, (iii) green: motor, and (iv) red: detection of movement. connections between communities are shown in gray.

17 connectivity matrix each row indicates whether region on the left sends to the region at the top each column indicates whether region on the top receives from one on the left large + indicates a pathway reported in archival literature small + indicates a pathway reported in abstracts or reports dots explicitly passed and found absent blanks indicate pathways not carefully tested question marks indicate conflicting reports NR and NR? absent in that direction even though the reciprocal is reported shaded boxes on the diagonal indicate connections inside the region

18 connectivity among visual areas

19 hierarchy for visual areas Felleman and Van Essen, 1991

20 hierarchy for somatosensory and motor areas