Tag Info

Hot answers tagged

13

In my experience, the term "semantic knowledge" (or semantic memory or conceptual knowledge) is generally used to refer to knowledge of objects, word meanings, facts and people, without connection to any particular time or place. The neural basis of this kind of knowledge is more or less agreed to depend on a distributed network of cortical brain regions ...


13

The standard complexity metric in theoretical computer science and machine learning, in particular in statistical learning theory, is the Vapnik–Chervonenkis (VC) dimension. It is of interest because it gives us a very good tool to measure the learning ability of a neural network (or any other statistical learner, in general). A good introduction to the use ...


13

Caenorhabditis elegans is probably not an ancestor to Humans. As found in Sponge proteins are more similar to those of Homo sapiens than to Caenorhabditis elegans, certain sponges were found to have more similar protein structures to humans than C. elegans suggesting the sponges are the ancestor. For your second point, it depends what you mean. The actual, ...


12

There is a huge body of literature on axon growth cone guidance which will give you some insights into how the biology works. Unfortunately, incorporating it all into a model is probably going to make it unwieldy unless your express purpose is to model the physiology, which doesn't seem like the case. Here are some references: Hong K, Nishiyama M. ...


11

There are some important relationships between the c. elegans nervous system and the human nervous system that should be pointed out here: Neurons in both animals communicate with each other via synapses that use special molecules called neurotransmitters to convey activity. All major neurotransmitters used in humans are also used in c. elegans ...


10

I'd like to add to Chuck's excellent answer; the computational approach is very well-represented in neuroscience, and actually involves a large number of very heterogeneous methods. Thus, a very different set of neuroscientists and examples have sprung to mind for me. To my mind, the best single example of the utility of a computational approach to ...


9

It's a local rule. All that it means is that the connection between two neurons gets stronger if you use that specific connection more. The specific connection (the synapse) must be used though; it doesn't apply to two random neurons that aren't connected that happen to fire at the same time. Hebbian learning is generic term for outcome; there are ...


9

There are many neuroscientists who use the techniques of advanced mathematics and statistics to analyze actual neural data for patterns. George Gerstein, who is now retired, has been a pioneer in applying "particle" methods in analyzing neuronal interactions. The originator of the Gravity transform, he used this tool to untangle some of the stochastic ...


8

I don't know of any NN algorithms that match your definition entirely, and I have looked for them (previously and recently). Here are some papers that I think are close or in the direction that you are exploring. Using theoretical models to analyze neural development (review) An Instruction Language for Self-Construction in the Context of Neural Networks ...


8

A (probably incomplete) list of ongoing whole brain simulation projects can be found at http://www.artificialbrains.com/. However, based on the information reported on this site, it is sometimes hard do distinguish what already has been achieved and what is still in the planning stage. Nevertheless it gives a good overview to start with.


7

Many parts of the fetus brain begin showing neural activity before the senses that feed them are sufficiently developed to provide actual sensory information. In other words, it is unlikely that spiking activity in the brain is initiated by the senses. Some of the cells that become sensory organs, however, often fire in very specific patterns similar to the ...


7

Artificial neural networks (ANNs) are mathematical constructs, originally designed to approximate biological neurons. Each "neuron" is a relatively simple element --- for example, summing its inputs and applying a threshold to the result, to determine the output of that "neuron". Several decades of research went into discovering how to build network ...


6

For the dentate gyrus, which is probably more closely analogous to a feedforward hidden layer in a memory network, here are some answers: Axon and dendrite connectivity is essentially local and can probably be assumed to be initially random within that local region. That is, a neuron integrating into the DG at the midpoint (along the long hippocampal ...


6

The use of neural-networks in the cognitive sciences has been around since Turing. However, many of the networks common in connectionism suffer from a lack of biological plausibility. Of these abstract ones, even the ones that try to capture some properties of biological neural networks only do some metaphorically. See for instance the limitations of cascade ...


6

I think that your intuition about the lower "energy ratio" of spurious states explaining their greater susceptibility to unlearning might be correct. In a Hopfield Network spurious states are activity patterns that have not been explicitly embedded in the synaptic matrix, but are nonetheless stable. They are in other words "unwanted" attractor states that, ...


6

Humans actually exhibit both slow and fast learning and they have somewhat different properties. One distinction is between "declarative" memory (for example, facts like "tigers have stripes" or "Paris is the capital of France") and "procedural" learning (such as perceptuo-motor skills like riding a bike or playing a musical instrument). Declarative memory ...


5

After doing some additional research, I think the answer is yes. It just means using a fixed timestep for the continuous-time activation equation (as described here). Since this is a differential equation, implementing it in software requires implementing a numerical integration method. I recommend the Exponential Euler Method as a starting point, because ...


5

Wen & Chklovskii (2005) looked at exactly this question through a simulation study. They assumed that the segregation of white and gray matter was the the result of evolutionary pressure to maximize some aspect of connectivity. They tested the idea that simultaneously maximizing interconnectivity (neurons should be able to connect to all other neurons ...


4

Traveling waves are a developmental mechanism to "prime" neural circuits. In a mature adult, certain mechanisms allow the strengthening and weakening of synapses (LTP/LTP, dendritic spine growth, etc...). These same mechanisms are used during development to achieve initial connectivity. For these to work however, spiking activity must be present. Experience ...


4

I think you are beginning to push at the limits of human brain versus computer metaphor.(mildly related link here) I'll list the objections as: While neurons firing can be translated/compared to 0 or 1 states, am not convinced it is a valid equivalence. (i.e: simulating 100 billion neurons on a computer would still not be able to match a human brain). My ...


4

From my understanding, no, this is not the case: action potentials, in the axon terminal(s) and other non-myelinated areas, travel like a wave, thus activating all terminals with the same starting energy. In the myelinated areas of the axon, however, the charge instead performs Saltatory conduction, where the charge jumps from one node of Ranvier to the ...


4

There is a passage in On intelligence about the differences between parallel processing in human versus computers : From the dawn of the industrial revolution, people have viewed the brain as some sort of machine. They knew there weren't gears and cogs in the head, but it was the best metaphor they had. Somehow information entered the brain and the ...


4

I think this recent paper fits your requirements. It considers biological plausibility by showing that the number of neurons required in the proposed method is within a reasonable size for the human brain, and dismisses a series of unreasonable models. Specifically, they create a neural network that contains 2.5 million neurons to contain a network of ...


4

Computational neuroscience and neural networks are both studied on this MSc at the University of Sussex. When I took the course in 2004/5, the Neural Networks module was compulsory, and the Computational Neuroscience was optional in the 2nd semester, so that would suggest the course designers (world leaders in biologically inspired computing) thought that ...


3

The question of how "rapid" learning could be possible relates to Hume's problem of induction -- how can we learn so much from so little. Historically, in both philosophy and psychology, the solution has fallen into one of two camps: either some form of the knowledge was already there to begin with (a 'nativist' view), or we use statistical inference to ...


3

Maybe I just don't get it, but I see your question as confusing because: 1/ Your brain is capable of running multiple parallel processes. Actualy each one of tasks you've mentioned consists of number of processes that are done at the same time. Lots of your neurons and neuronal networks are being used at the very same moment. 2/ If you can do something ...


3

Well firstly, what are those statements really saying? How do you measure "processing" or "power" as it relates to the brain? For an electrical engineer working in communicatons, it's easy: you just take the square of the amplitude of the signal (signal power) or count the number of instructions per second (processing power). But that's the discipline ...


3

For the question of whether humans could fight neuro"mind-reading", I can only say that there is very little evidence that information we know but are not thinking about could be extracted using current methods. This means that the old scifi mind-reading-fighting standby of thinking of something else might work--although then we have the "don't think of a ...


2

I'm not sure I fully understand your design; perhaps you can clarify what you want your network to learn, why TD-learning "isn't cutting it", and what you mean by 'reinforcement' and 'prediction' learning. In particular, TD-learning is a reinforcement learning model, and it does reward based on predicted (and not just observed) outcomes. However, you seem ...



Only top voted, non community-wiki answers of a minimum length are eligible