Lecture
Это окончание невероятной информации про мемристор.
...
mold — its ability to learn, forget what it has learned, and restore that memory when reminded.
The behavior of the mold Physarum polycephalum — this unicellular organism is capable of devising nontrivial ways to navigate mazes, solve geometric puzzles, and even serve as part of the simplest cyborg robot. For one of the studies on this mold, its authors even received an Ig Nobel Prize in cognitive science.
Slime mold is able to remember what it was taught, forget the lessons, and restore the memory of them when reminded. Japanese scientists not only observed this in an experiment but also built a model of it. Tetsu Saigusa and Toshiyuki Nakagaki of Japan's Hokkaido University reported in the pages of Physical Review Letters how they discovered the slime's capacity for learning. If, three times in a row, the temperature and humidity are lowered for 10 minutes each hour, then an hour later the cell will freeze for 10 minutes in anticipation of unfavorable conditions — even if in reality the humidity and temperature remain optimal. Over time, the mold forgets what it was taught. But if it is reminded of the unpleasant conditions by once again lowering the humidity and temperature, the memory returns: an hour later the mold again freezes in anticipation of the worst, like a professional lawyer who always prepares for the worst case.
Physarum Polycephalum experimental maze solving (from Nakagaki et al. [17])
The mold's possession of memory was not entirely unexpected. Earlier maze experiments had shown that at each new fork the mold remembers which way its tendrils turned at previous ones. What surprised the scientists was specifically the ability to learn, forget, and restore memory upon reminder. The Japanese researchers proposed a hypothesis to explain what was happening. According to it, inside the giant cell of Physarum polycephalum there is a huge number of diverse biochemical "internal clocks" with a wide range of different periods. Some of these clocks speed up the mold's movement, while others slow it down. But because they all run out of sync with one another, the cell as a whole crawls at a more or less constant speed.
But when external conditions change, the processes that were speeding up movement switch off: in cold, dry conditions they consume too many resources. The only clocks that remain active in the organism are those that slow the pace down. This state persists for some time, and the slime's movement stays slow regardless of whether fate deals it further blows. Then another jolt occurs, and after some time the mold forgets the imposed rhythm. This hypothesis has one significant drawback: it does not explain how the mold restores a forgotten rhythm when it is reminded by a change in conditions.
The fundamentally new simulation model of the mold is an electrical circuit that you could even solder together yourself, given one exotic component — a memristor. Your only problem would be finding that memristor itself. Otherwise, this circuit is a series oscillating LC circuit with losses. Why a circuit specifically? Because the circuit needed a source of oscillations. There's no getting away from the resistor, since in the real world every element has some resistance. The memristor was connected in parallel with the capacitor, since any other configuration either gave no result or made the model inadequate to the real behavior of the mold. The circuit parameters were optimized and tuned to match the behavior of real mold.
Fig. Electrical circuit model of the intelligence of the slime mold Physarum polycephalum (A) and its behavior under irregular (B) and regular (C) changes in external conditions (Pershin et al., 2008, arXiv.org)
The memristor has two basic states — one of high conductivity and one of low. It can be switched from one state to the other by an applied voltage, which is equivalent to a change in the external conditions for the mold's development — temperature and humidity. A positive voltage represents favorable conditions, a negative one unfavorable conditions. The system's response — the speed at which the slime crawls in the model — is the voltage across the memristor. If a single negative pulse is applied to the input of the circuit — placing the mold in unfavorable conditions — the circuit responds with rapidly decaying oscillations with a period determined by the parameters of the LC circuit. In this state the memristor's conductivity is at its highest. But when the input receives a series of pulses whose period is roughly equal to the period of the circuit's free oscillations, then at some point the voltage across the memristor reaches a threshold value, and it quickly switches into a high-resistance state. This is how the mold "remembers" a series of unfavorable impressions. Oscillations in a circuit that has stored this charge decay much more slowly.
All it takes is to "remind" the circuit with one more negative pulse, and it immediately responds with a series of slowly decaying oscillations — it "recalls" what it was taught. Moreover, the memory turns out to be very long-lasting — it is restored even after a hundred periods of the LC circuit's oscillations. The mold's memory is shorter.
Of course, this is only a model. How the memory of a real slime mold is actually organized remains unknown. It is not clear what sets the rhythm within this cell, how it remembers signals, and what makes it forget them. However, general knowledge of the mold's physiology suggests that the role of the memristor is played by a system of channels that transport cellular fluid within an elastic shell. The movement of the myxamoeba is essentially a continuous back-and-forth flow of intracellular fluid with a slight forward bias. Continuous vibrations of actin-myosin protein fibers create a pressure gradient that drives the fluid. These fibers themselves are attached to the cell's shell, which in turn is subject to friction against the surface over which the myxamoeba crawls. As a result, the frictional force partially balances the fibers' reaction to the fluid's motion, and the cell as a whole moves forward. The fluid in the cell exists as ectoplasm and endoplasm. The former is more viscous than the latter, and the endoplasm makes its way through channels in the ectoplasm, much like mercury through glass. But when the pressure on the ectoplasm at some point exceeds a threshold value, its viscosity drops sharply — one could say that another channel opens up in the ectoplasm. Clearly, the more channels that are open, the faster the myxamoeba can move. And the number of open channels is ultimately determined by the history of movement — in exactly the same way that the memristor's resistance is determined by the history of the voltage across its terminals. Incidentally, if one grasps the physics of how the memristor works, one can notice a deep resemblance between the two models.
Часть 1 The Memristor: the Fourth Passive Element of Electrical Engineering
Часть 2 - The Memristor: the Fourth Passive Element of Electrical Engineering
Comments