# Tag Info

5

Replicating fMRI signal was first used in the Tower of Hanoi task as seen in section 5.8 of "How to Build a Brain": There is strong evidence that dendritic processing, driven by neurotransmitter usage, underwrites the BOLD signal (Logothetis & Wandell, 2004). It is that BOLD signal that is actually measured by MRI machines. Consequently, MRI ...

5

First, let's take a look at the basic principles of NEF. The first two principles (Representation and Computation) do seem analogous to trained ANN models. Additionally, with the hPES learning rule that I've described here, they seem to have the same learning (gradient-descent) capabilities. Where the NEF differentiates itself the most from ANNs is when it ...

5

From the comments: I'm going to hazard a guess that these are neurons that are tuned to a particular direction in space and that the x-axis is the angle in multiples of π radians, particularly since these are related to the work of Georgopoulos and colleagues. Since we know these are positionally tuned neurons, you can see some other examples in this ...

4

SPA is used (among other things) for combining (binding) and extracting (unbinding) knowledge representations for processing. This is a (purposely) lossy compression. In the "Learning Rule Generation for Induction" case, the clean-up memory is used to convert a general transformation that is being learned (lots of different transformations convolved together)...

3

My answer is that you have the beginnings of a grasp of neuronal tuning. But the point is not that neurons can represent functions. The point is generally that neurons contain information about certain experimental conditions. Rather, neurons can represent functions, but in most cases they tend to represent something closer to propositions. The seminal ...

3

According to the paper, the advantage of this new approach over conventional ANNs, Deep Belief Networks (DBN) and Self-Organising Networks (SON) are: Remains functional during online learning. Requires only two layers connected with simultaneous supervised and unsupervised learning Employs spiking neuron models to reproduce central features of ...

2

For a learning rule to be biologically plausible, it has to only depend on knowledge/information local to the neuron (no global information about the neuron population) and has to match experimental neuroscience data. Only Neuron-Local Information As discussed in "Simultaneous unsupervised and supervised learning of cognitive functions in biologically ...

2

The NEF has no formal stance on whether rate information or phase information is of importance. All it cares about is spikes. As proof, consider the derivation of the decoders, which is the core of the NEF. The Decoders, $d$ are a vector of synaptic weights applied to the activities of neurons (one decoding weight per neuron) to approximate a given ...

2

From Wikipedia: Content-addressable memory (CAM) is a special type of computer memory used in certain very-high-speed searching applications. It is also known as associative memory, associative storage, or associative array [...] It compares input search data (tag) against a table of stored data, and returns the address of matching data (or in the ...

2

To begin to answer this question, we must first unpack the concepts in their current context. The NEF makes no prediction about how error is propagated in the brain. It explains how to do computation using vectors in spiking neural networks. Also, it defines how error signals can be used to change how the signal is encoded (take in) and decoded (sent out) ...

2

An Associative Memory is a classifier and is equivalent to a Hamming Network. For documentation on the NEF Associative Memory, see this practical Nengo documentation and the paper "A biologically realistic cleanup memory: Autoassociation in spiking neurons". Basically, each ensemble of an associative memory computes the similarity measure of the input ...

2

As seen in "Neural Engineering" by Eliasmith et al. Chapter 4, complicated neuron models have greater computational abilities and match neural data more realistically. Computational Capabilities As seen in the following table (taken from Neural Engineering) Adaptive LIF neurons, by virtue of their temporally varying firing patterns (i.e. a constant input ...

1

Counting task numbers in Spaun becomes somewhat meaningless due to it's instruction following capabilities. However, the missing task was the stimulus response task, wherein given an image from ImageNet, classify it according to it's given identifier.

1

The model does not care what x is. You first create a population of 100 neurons to represent an input that will vary from -1 to 1. That input, which varies from -1 to 1, is x. For sensory neurons, the variable x would likely represent a transformation of some aspect of the sensory stimulus (e.g., the visual angle, sound level, or temperature), but it can ...

1

The Neurological Engineering Framework does not explicitly state a mechanism for memory. There is no "hard-drive" in the brain for easy retrieval and access. Rather, memory is captured in the connection weights between neural populations and the dynamics of the network. In the Hierarchical Reinforcement Learning example, linked to in the previous question, ...

1

At this point in time, the difference in neurotransmitter types affect the synaptic time constant (i.e. the filter on the incoming spike train) between neurons in Nengo. See Neural Engineering p.112 and these notes (see the section called "Biologically plausible filter") from a course covering the book. Neurotransmitters are also leveraged more ...

1

To review how neurons encode information, please check out these class notes review encoding. In those notes, you'll notice the intercept $J_{bias}$ and the maximum firing rate $\alpha$ are randomly selected when encoding functions in large populations of neurons. These variations can account for heterogeneity in attributes of neurons.

1

Within the context of the tutorial referenced, this graph is showing the first principle of the NEF, which is that neurons approximate functions by encoding them with their firing rates. Here the input being represented it the range 1 to -1. What the graph shows is the firing rates of all the neurons given the value being represented. So say you have the ...

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