One problem, or feature depending on your viewpoint of Content addressable memory, is that since each memory cell, individually tests for content individually, the more memory cells involved, the more tentative solutions you get. David Marr, who suggested that the cerebral neo-cortex acts like a self-categorizing content addressable memory back in 1970, noted that redundancy, was a problem (Feature).

To understand how the brain, deals with redundancy, we need to understand the idea of a Neural Group, and transfer it to our circuit. Essentially we could say that our content addressable memory cell, is somewhat equivalent to a neuron. Neural Groups are clusters of neurons that act together to vote on whether or not one or two of them will fire.

Obviously a BIT does not have that much content, so we will want to connect a number of them together to form a representational grouping that contains more information. For the purpose of illustration, I have implemented a 4 bit array, which is suitable for storage of BCD representations.

Obviously we want to detect, the matching of a BCD representation with the data within the array, in order to recognize a specific numerical digit. We can do this, a number of ways, for instance we could use a 4 input AND gate on the match lines and thus always activate when there is an exact match, or alternately we might want to vote on the output, using a Majority Gate, that will trigger if the majority of the match lines are triggered. Alternately we might want an even weaker (Fuzzier) approach, where redundancy is tolerated in order to increase the information content reaching later stages. Satisfycing rather than detecting conditions of matching between data and match lines.

One reason we might not want to implement a majority gate, is simply, by definition you have to have an odd number of bits in order to have a majority.

Since most computer codes are represented in strings of a natural power of 2 bits, a majority gate will never be useful if we want to take advantage of existing computer technology.

The slightly less selective Two-or-more gate will have to do. This gate will always be triggered if two or more of the bits, in the 4 bit array are triggered. I call it the Satisfycing Gate, to indicate that it is less selective (Fuzzier) than either straight logic, or the majority gate.

The problem with lower selectivity, is that your output is degenerate in the sense that more than one code will result in activation of the same output. If any two bits in the BCD bitstring match, the output for the group will turn on. Obviously you have not detected the BCD value at this level, since you have a muich wider field of output at which the Neural Group will fire the satisfycing gate. This means that recognition of a BCD representation, is not done at the Neural Group Level, just as it is not done at the bit level.

Why would the brain use degenerate codings for memory? The answer lies in the fact that individual cells in the brain do not survive the life of the brain. This means that the brain needs to fail safe, in other words, when a single cell in the brain is lost, the information remains intact, even though that neural group fires according to a slightly different data grouping.

By making the cells redundant, and detecting patterns in firing of the cells, at a higher level, the brain allows individual cells to die without seriously affecting its ability to detect the same content. Unfortunately it also means that the brain has to go through more stages in its processing than an equivalent logical detection circuit would require. To lose the efficiency of a detection circuit, but gain the survivability of the information over the life of the individual, is a serious trade-off.

As a response to the recent ict/programme/FET call, I asked the FET group if this project would be suitable and was told that it was.

Essentially FET is looking for emerging technologies like implicit memory research, that they can support Ideally, to get funding for a proposal like this, we would need to form a Consortium with at least three EU Organizations in it that are registered on ECAS. I have recently added GreySmith Institute and added myself as Lear for the Institute.

The OPEN funding thread seems to be the most useful for a project like this, and it requires the presentation of a 5 page proposal describing the project in general, which will be followed by a larger proposal at a specified date.

The first proposal is Anonymous, and essentially is a peer review of the project concept

The second proposal goes into detail. The First Proposal has to be accepted before you will be allowed to enter the second.

Proposals are in Black and White, and links are not encouraged.

Ok I was looking a little closer at the paperwork, and found that the ICT/Programme/FET proposal typically offers between 1 and 2M Euros for a 2-3 year project. Given a small consortium of 4 centers, that would still be about 250,000.00 EU per research center. Giving you a budget of about 83,000.00 EU/year of which up to 60% might be indirect expenses depending on your accounting model. Meaning you might be able to absorb 49,000.00 EU /year into operational expenses that do not directly affect your productivity.

If you used this as a basis for asking support from the local government in some sort of matching grant, you could quite likely afford to do an open source project, and pay a few post-docs to help you.

I see three different threads, at least, although a fourth one is possible but might be questioned by the ICT project because it is more Histo-Psychology than ICT.

Scala Neuron for Simulation of Neural Networks with Heterogeneous groups and cluster/grid based processing engines

Data Driven implicit memory database with rulebase like parallel search and fuzzy datacloud output

Digital Simulation circuits based on standard logic gate technology

Rodent based studies of implicit memory

I found this in the FET programme documentation:

4.2 Challenge 2: Cognitive Systems, Interaction, Robotics

Engineering systems with the capability to sense and understand an unstructured environment

is a challenge which goes beyond today's systems engineering paradigm. Present day systems

engineering relies on specifying every eventuality a system will have to cope with in the

execution of its task(s), and programming the appropriate response in each case. With the

abundance of ever cheaper, smaller sensors, actuators and wireless tranceivers that link

systems to the real world and with other systems, this approach faces serious limitations:

- The real world is generally too nuanced, too complicated and too unpredictable to be

summarised within a limited set of specifications; there will inevitably be novel situations

and the system will always have gaps, conflicts or ambiguities in its own knowledge and

capabilities.

- Even in situations where unpredictable events are less likely, the problem of extracting

meaning and purpose from bursts of sensor data or strings of computer code arises,

because we don't have a formalisation of information processing that embodies semantics.

This makes a very good case for implicit memory as being valuable in order to add semantics to stimulus detected with inexpensive sensors. Where current systems require that the software and hardware be specified for each other, a system with a sophisticated implicit memory should be able to extract the data from a number of sensors, without requiring specific design time data. Essentially it should cozy up to the sensors, and learn how to evaluate their information content despite the nature of the sensor itself.

The same tolerance for genetic indeterminacy that is required for biological systems to make sense out of their universe, is needed in engineered systems that will have maintenance indeterminancy. Unlike the eye, for one instance where the genetic parameters have not changed in millions of years, but in which some range of function can be found depending on the specific DNA and development history of the organism, Electronic components have a specific lifecycle that is usually not measured in hundreds of years. They also have a specific marketing cycle that is similarly constrained. Thus it is entirely plausible that a specified optical sensor will have passed its marketing life-cycle before it has passed its operational life cycle.

If this is the case we can pretty well guarantee that even if new equipment has similar optical sensors, that they will be incompatible with the old circuit. In circuits where the software and the electronic components are matched by specifications, this means that maintenance will require the replacement of not only the hardware that has failed but the software that drives it.

However replacement of sophisticated software requires the ability to determine what the original software did, which due to licensing and copyright limitations, will not be readily apparent where the company writing the software is no longer available to offer software listings. Thus with specified systems, degradation is guaranteed, if only because software maintenance becomes impossible after some point of time. Hence Obsolescence and E-waste. Aware of this, manufacturers, tend to design their systems to arrive at a catastrophic failure rate, after approximately 2 times the duration of the warrantee.

One aspect of implicit memory is that it degrades gracefully, and adapts to inconsistencies in function between one part and another, thus adapting to the eventual failure of components. As well, because it adapts to the components in the first place, it is open to maintenance by replacement with a slightly different standard of component. There is a learning curve while a new component is fitted, but once that learning curve is done, the system has the capability of working with the new components different capabilities, instead of working only to a pre-designed specification.

Implicit memory therefore is quite important to this Challenge, which might be one reason why the FET people are interested in possibly funding it.

Coffee is a known stimulator of the brain.

What this means is that it either replaces, or increases the volume of a Neuro-transmitter, thus increasing the strength of its signal.

Long known affects, are addiction, adaptation so that people cannot function before their first coffee of the morning, tremor, hypertension and an increase in the number of headaches. However we digress from the topic of this forum.