HBP web editor Greg Meylan talked to theoretical neuroscientist Dr Misha Tsodyks about his first principles approach to understanding memory recall.
The day after I spoke to Misha Tsodyks I was standing at a pedestrian crossing in the bright Geneva sunlight waiting for the lights to change. My mind was idling, memory circuits ready to fire at the smallest provocation.
A car with a roof-box turned in front of me. As my eyes followed its movement, the shape of the roof-box ran down my neural pathways and hit upon a memory: of my friend’s father, an idiosyncratic hydrologist with a fine knowledge of fluid dynamics. He always attached his roof box the other way round to everyone else — with the flat end facing forward and the slope at the back.
More like the shape of a falling raindrop, I thought. Then I remembered that the classic raindrop shape is not a true depiction (see below). I’d read somewhere what shape raindrops actually take on their descent to earth, but I couldn’t recall exactly what it was.
From car roof-box to raindrop. This was my memory firing, flitting from one thing to the next, sparked by the random passing of a car. Which was pretty much how Tsodyks had described his theories of memory recall to me the day before.
With his collaborators Mikhail Katkov and Sandro Romani, Tsodyks recently published a paper in Neuron that sought to demonstrate a first principles approach to understanding memory retrieval.
The paper provides a framework for how memories are triggered sequentially. When one set of neurons in a memory network is activated, it triggers the retrieval of the next item. The stronger the connections between two given memory networks, the more likely one will trigger the other.
The theory is based on two principles:
1. An encoding principle, which states that an item is encoded in the brain by a specific group of neurons in a dedicated memory network. When it is recalled, this specific group of neurons is activated.
2. The associativity principle, which states that in the absence of sensory cues, a retrieved item plays the role of an internal cue that triggers the retrieval of the next item.
The Neuron paper describes the process in what Tsodyks says is fairly simple mathematics. The value of this description is that it can produce predictable, testable results, something which is fairly rare in this area of neuroscience, he says.
“The surprising thing is that by applying it in an analytical framework, you get really clear predictions. If you are trying to remember something that has no structure, e.g. a list of words, or the movies you have watched, and if you apply this idea of sequential triggering, then you can predict how many things you are going to remember out of a list of a certain size.”
When tested against classic free recall experiments Tsodyks’ formula produced results close to those observed. Further experiments are planned, and Tsodyks is hopeful that the theory will hold. One of the surprising predictions of the model was that each word participants were exposed to in the free recall experiment had an intrinsic recall probability. Furthermore, the retrieval of easy words leads to a reduction of the overall number of retrieved words, due to the circuits following strong connections around in a circle.
Because memory is so important to how the brain functions, Tsodyks believes theories of its processes will help in other areas. Indeed, he sees no fundamental difference between memory, thought or talking.
“I think thinking and talking are memory recall. The way I think about it is that thinking and talking are actually the same process. Because the way I think it works is that every thought you have is the signal for the next, every idea is linked to another one.”
“In Neuroscience, there is nothing you can really predict. We do not know how things really work, and the brain is so complex. Both these things mean you cannot make quantitative predictions.”
The physical world is the same, he says; for example, you cannot predict temperature of the sun. But you can simplify it, and with basic rules you can predict things like how long it takes a ball to fall. His paper on first principles seeks to provide a similar framework.
As the Neuron paper says: “Despite thousands of years of astronomical observations, a clear picture unifying the motion of celestial objects appeared only after the formulation of a very few basic principles, known as Newton’s laws.”
The theory should be able accommodate more biologically realistic mechanisms and representations, says Tsodyks, with the possibility of still deeper explanatory powers.
In the meantime, I’ve looked up what a rain drop looks like. It looks a bit like a hamburger bun, flat on the bottom and curved on top. This description is now entangled in a memory circuit that should forever overlap with roof boxes, my friend’s father, and a theory on memory itself. If the theory holds.
Dr Misha Tsodyks works at the Weizmann Institute of Science, Israel. He is part of the HBP’s Theoretical Neuroscience team (HBP Subproject 4), whose theories help inform the work of other areas of the HBP such as brain simulation, neuromorphic computing and neurorobotics.