I had very high hopes for this book. After reading it cover to cover, the single useful piece of information I got was this: Neurons in a bat (the kind that flies and uses echo location for locating its prey and obstacles) are able to spike once when receiving 2 sequential input spikes spaced some fixed amount of time apart -- measured in milliseconds. If the time between the two spikes isn't just right this particular neuron won't fire. And the bat can change its behaviour so fast in response, there isn't enough time for any more spikes to occur. The implication is that even a single spike can convey information.
Everything else in the book is garbage, and you absolutely should not waste your time on it. If you're interested in understanding how the brain works, this book will send you off on the wrong path.
Imagine one person writing a note, in ink, on a piece of paper, and handing it to someone else. The receiver then takes the note and then exaustively starts analyzing the fibers of the paper and the chemical makeup of the ink used. He spends an incredible amount of time on it, he determines the fibers are usually of a certain length and thickness, the ink is composed of a mixture of a number of organic compounds, if you add alcohol the ink will dissolve. He determines the ink is impervious to water but the paper will come apart...a myriad of details. The analysis is flawless, the information presented accurate in every detail.
In the receiver's analysis of the note, he mentions that perhaps a productive avenue to making sense of the note might be to look at the shapes and organizations of the figures, perhaps they convey some sort of information. But he doesn't actually do that analysis. In fact, even though the note is written in english, he never bothered to look at it and try to read it.
Well, the book Spikes: Exploring the Neural Code is like that.
The book is a tour de force of a variety of concepts. We are presented with Information Theory, Taylor Series, Fourier Analysis...many more as well. These are all applied to the question of how to analyze the processing that is going on in a single neuron. There is an assumption that there is a code associated with the firing patterns of neurons, and the book then attempts to divine what that code might be.
The book treats neurons like black boxes. Spikes come in, spikes go out, all the neuron has to do is decide whether to output a spike or not. The book presents measurement data on actual neurons, and shows the results of their various methods of analysis.
There is an implicit assumption that each neuron, alone, is passing information from input to output. The output spike pattern is assumed to be some representation of the input spike pattern. Then the book exhaustively tries to figure out how the output and input might be related. Information Theory comes into play at one point when there is an effort made to determine the bandwidth (bits per second) of information passing through the neuron. The authors remark that the nerve fibers seem to carry close to the maximum amount of data possible given the environment -- and are suitably impressed.
I never followed the math presented. All the time while I was reading, I was hoping to discover some new insight. It was obvious from the moment I realized their implicit assumptions were horribly flawed that there was no point in actually working through the math and trying to learn from their analysis.
Their most basic, fundamental assumption is that the information carried along nerve fibers is conveyed in the frequency of the spiking. That is, fewer spikes per unit time vs more spikes per unit time. Or in other words, the timing between successive spikes is what conveys information. Given this assumption, they are mystified when confronted with the evidence that a bat (like certain insects or other animals) is able to modify its behaviour based on the action of a single spike in isolation. Their reasoning is that the animal in question responds so fast that there isn't time for a second spike to occur -- therefore the first spike, alone, conveyed all the critical information necessary to influence behaviour.
The authors created this mystery by adopting their (flawed) assumption that the rate of spiking is what is carrying the information. Their entire analysis concerns the firing rate of single neurons acting alone. No pages are devoted to an analysis of how groups of neurons, acting in parallel, might be performing some function of greater complexity than single neurons acting alone.
The neurons the book speaks of are unchanging...there is another implicit assumption that the act of feeding test patterns into neurons and looking at their output firing patters in no way affects the internal state of the neuron itself. Their analysis doesn't take changing internal state into account. Instead the neuron is assumed to be equivalent to a function: output = function(input). Then they try and come up with mathematical approximations to what the function might be. Or perhaps they're trying to educate the reader on the appropriate math to use in order to do that.
Real neurons change their behaviour based on how the firing patterns presented on their inputs. Real neurons don't simply have one input. Some have as many as 100,000 inputs. Real neurons change over time. Do something on input, and something comes out. Wait a while. Do it again, and you get something different.
Real neurons exist in an environment of other neurons. They're all bathed in blood. Other neurons in the vicinity can cause a large number of diverse neurotransmitters to be diffused into the general area -- targeting pretty much every neuron. Spikes... doesn't go into any of this detail. Neurons form new connections. They can put out new tendrils, reach out and touch other neurons. Spikes... doesn't explore this. During the early life of a human, a great deal of neurons simply die, presumably to make room for other more important neurons. There are a whole lot of details about neurons that are important, that are critical to understanding how the mind and brain work.
However, Spikes... seems to concern itself with one hypothetical, constant, unchanging neuron that has only one input and one output.
Now, I only have my opinion to go on, but my take is that one should not even try to figure out a code associated with the spikes. There is no code. It's a spike. That's what it does. The spike itself is the information. It's like the baton in a relay race. As long as the baton gets where it's going, the system works.
As an example, we can look at the motor neurons. Each motor neuron is connected to a patch of muscle tissue. When a motor neuron fires, the muscle contracts a bit. If the motor neuron stops firing for a period, the muscle will relax and stretch out. More firing -- more contraction, up to some maximum I suppose where it can't contract any more. The spike is saying, "Please contract a bit." End of story. You pack more per unit time, the muscle will contract more and faster.
How would information theory come into the picture here? Where does fourier analysis fit in? Who cares what the shape of the graph of time vs firing frequency looks like? These things just don't matter. When you get spikes occuring, stuff happens.
In conclusion, even though I don't have all the answers, I know enough to know Spikes... isn't the answer. If you want to figure out how the brain works, don't expect any insight to come from this book.
Access count (e^i)