Ai Dreams Forum

Artificial Intelligence => General AI Discussion => Topic started by: LOCKSUIT on March 01, 2017, 03:07:04 am

Title: Is short-term long-term algorithms what I think they are?
Post by: LOCKSUIT on March 01, 2017, 03:07:04 am

Is short-term long-term algorithms what I think they are?

Are they long-term memories, that, when selected, can also hold a short-term pointer, which are forgotten but not actually forgotten?

Ex. remember 6 h 8 3 h a p p u car fox drive god n a 6 6 99 cents yes cool me you

Forgottennnnnnnnnnnnn HA   HA

But in ur head, u will forever know ALL OF THOSE WORDS, LETTERS, AND NUMBERS.

The point (lol) is that you can remember car fox mothership. Then forget in a day say.

The real funy thing though, is, my "VideoSense" IS this.

Title: Re: Is short-term long-term algorithms what I think they are?
Post by: kei10 on March 01, 2017, 04:04:06 am

Long term memories are ones that are stored but difficult to retrieve, as it is stored through deep links and connections that require searching and similar functions.

Thus short-term memory is used to hold retrieved random memories, as well as probably allowing faster memory retrieval, and so on.

Much like the HDD/SSD and DRAM in perspective.

Title: Re: Is short-term long-term algorithms what I think they are?
Post by: 8pla.net on March 01, 2017, 04:53:09 am

Let's start programming that algorithm using a C Language pointer, then...

Code

#include <stdio.h>
#include <string.h>

void main()
{
 char longterm[80];
 char *pointer = "short-term memory";

 printf("hold a %s\n",pointer);
 fgets(longterm,80,stdin);
 longterm[strcspn(longterm, "\n")] = 0;

 printf("%s of %s.\n",pointer, longterm);
}

OUTPUT:

Quote

hold a short-term memory
blue skies
short-term memory of blue skies.

EXPLANATION:
The first line is the prompt for input.
The second line is the input.
The third line is the output.

Title: Re: Is short-term long-term algorithms what I think they are?
Post by: LOCKSUIT on March 01, 2017, 07:03:04 am

.....What are you trying to do there 8pla?


@kei
I can recognize long-term memories fast and easy if my mom says "remember aunt so-and-so" or "video game's certain level" or or or anything!!

The CNNs that my AI's long-term memory and short-term memory would use are both as easy/fast to search and retrieve memories & connections.

My AI uses 1 strengthening process.

The onlyyyyyyyyyyyy reason it'd use short-term memory is the reason I explained in my opening post. Tell me car. I'll forget the password. But I'll forever know car. There's a secondary memory.

Title: Re: Is short-term long-term algorithms what I think they are?
Post by: Korrelan on March 01, 2017, 11:22:02 am

I think that in the human brain there is no difference between long and short term memories.

To remember a list of words or numbers you have to be able to recognise and understand the items; you can’t remember what you don’t recognise/ understand.

So called long term memories are stored in the synapse patterns that connect the groups of neurons.  Remembered items are composed of sparsely distributed neuron groups that represent a particular facet of the item.

The so called short term memory is just the current thought pattern; or pattern of activation within the synapses/ neurons at any given moment.

Memories/ Information are never moved around inside the brain; it’s never moved from short to long term storage.  ‘Memories’ are never searched either (searching is a serial schema); because of the parallel architecture the brain has no need to search. The ‘thought pattern’ flows from one pattern of depolarized neurons to the next pattern through the axons/ synapses/ connectome; it’s the structure that stores the memories. 

When a particular pattern is recognised the resulting output pattern is added to the overall/ global pattern; this changes/ morphs/ phases the global pattern which is in turn recognised by other areas… repeat.

If you were to learn a new property of water for example; your brain doesn't have to update all its knowledge regarding water and it’s uses/ properties… it simply has to include the new property into the ‘thought pattern’ representing water when ever you think of it.

Look at the letter ‘C’; that has just included a pattern into your global ‘thought pattern’ that represents (too you) that letter… now look at ‘A’; now that pattern is included. The two separate patterns have created a merged pattern that represents ‘CA’ and already your brain/ cortex is firing other patterns that relate to ‘CA’. Now look at ‘T’… bingo. The combined pattern of the three letters was instantly recognised by areas of your cortex that then ‘fired’ the pattern for ‘CAT’ back into you global thought pattern intern firing patterns that represent the general concept of ‘CAT’. At the same time there where patterns running that covered this topic, your body posture, how comfortable you feel, etc. Reading this paragraph and your thoughts/ opinions on it have altered your ‘global thought pattern’; you don’t need to remember the letters ‘CAT’ or even the basic method of explanation I’ve used; they are already well engrained… it’s the new/ different bits of the pattern that get etched into your synapses.

If you look at this sum 5+5= you don’t have to mentally count or add the numbers on your fingers; the visual pattern of seeing the sum fires a globally understood pattern that represents 10.

My AGI is based on the mammalian connectome and exhibits both/ all these memory traits; this is why I believe I am correct in my assumptions.

 :)

Title: Re: Is short-term long-term algorithms what I think they are?
Post by: 8pla.net on March 01, 2017, 02:58:57 pm

.....What are you trying to do there 8pla?

I attempting to point out with an algorithm, using a pointer in C Language, what long-term and short-term have in common...  Both are terms, words or phrases, and/or both are limited to a period of the 4th dimension (time).

Title: Re: Is short-term long-term algorithms what I think they are?
Post by: Art on March 01, 2017, 03:37:17 pm

@ Korrelan - No difference in structure perhaps but definitely a difference in how readily that stored information / data is accessible when needed. This is especially evident in some people that tend to misplace things or fail to recall a name, date or place that they know they know.

Short term memory is what you had for breakfast 20 minutes ago yet there are people who couldn't provide an accurate description. That same person might be able to tell you what they had for dinner on their 21st birthday (Long Term Mem).

While some of these examples might not necessarily be the norm, they are quite real and plentiful. Doctors address these as Short Term Memory and Long Term Memory as different area of the brain. Not being a neurosurgeon, I couldn't say with certainty.

I have seen this in people over the years so when you stated that "...there is no difference between long and short term memories.", I have to take exception to that, with all due respect.

Title: Re: Is short-term long-term algorithms what I think they are?
Post by: 8pla.net on March 01, 2017, 05:10:48 pm

Very interesting theory, korrelan... Especially as it may relate to a practical AGI design, I think.

My theory is that, if a short-term memory may be longer than a long-term memory, and a long-term memory may be shorter than a short-term memory, then we may calculate memories as mode-term (reoccurring memories), median-term (central memories),  mean-term (average memories).

Oh, I just passed 700 posts at AiDreams.

Title: Re: Is short-term long-term algorithms what I think they are?
Post by: Korrelan on March 01, 2017, 07:18:50 pm

@Art

Quote

No difference in structure perhaps but definitely a difference in how readily that stored information / data is accessible when needed. This is especially evident in some people that tend to misplace things or fail to recall a name, date or place that they know they know.

Perhaps a better way for me to describe it is weak to strong memories on a linear scale; using the same architecture/ schema.  So short is a weak memory and strong is a long term memory.  We don’t have mid term or super long term memories… where would you draw the line? I think the whole long/ short thing is very misleading.

When we form a memory we are just linking together existing knowledge/ understanding/ experiences with weak synaptic connections/ associations.

Quote

Short term memory is what you had for breakfast 20 minutes ago yet there are people who couldn't provide an accurate description.

To remember what I had for breakfast I have to use strong/ long term memories. To understand what the items were, what they were called, even the concept of ‘breakfast’ requires a lot of strong/ long term understanding/ knowledge/ memories.  The weak/ short term bit of the memory is what links all the various items together along with a location/ timestamp/ index/ etc that I would recall as ‘earlier today’.  For the unfortunate people who have difficulty retrieving today’s (short term) memories I would wager they have a problem with a brain region responsible for temporal tagging/ indexing the memory as ‘today/ recent’

Quote

Doctors address these as Short Term Memory and Long Term Memory as different area of the brain.

I’m not sure about this; I have no reference to any research that specifically points to long/ short term memories being stored in different brain locations.  I could how ever explain why a scientist/ neuroscientist might think this was the case. Different brain regions supply different facets of a memory.  The Hippocampus (could be revised) for example provides an index/ temporal date stamp to a memory as its being stored (not read); if they where viewing frmi results on memory consolidation they would be measuring the difference between existing knowledge and new learning. The new learning would require a recent time/ stamp that would activate the Hippocampus (episodic), blood would flow to this region highlighting it.

Quote

That same person might be able to tell you what they had for dinner on their 21st birthday (Long Term Mem).

Once a memory has been consolidated it can be recalled by any small aspect/ facet of the original/ whole memory.  A smell or even an image (pareidolia) can trigger a memory/ recognition event.

If any of the original facets that the memory was comprised of are compromised in some way it can make retrieval difficult from that aspect, time, location, etc… we all use the tactic of thinking of related/ similar memories when trying to recall a weak memory, you’re just trying to produce a global pattern conducive to triggering the required memory; trying to fill in the missing blanks in the pattern that will trigger retrieval.

The limbic (emotional) network adds a very strong facet to a memories overall pattern; this is how repressed emotional memories are triggered.  The memory was painful so we made the effort to forget/ shun it from our normal thoughts… the links that would trigger the memory from day to day thoughts become weaker and so it doesn’t pop into our heads. A strong emotional response however from an unrelated experience can be enough to trigger a total recall.

This is a diagram showing a rough map of where the various categories within the circle where detected upon the human cortex surface.

(https://aidreams.co.uk/forum/proxy.php?request=http%3A%2F%2Fi.imgur.com%2FvJLHRKs.jpg&hash=498a2fbaf3bfba0835a8f2f725179ed847948200)

This is a short video showing the same/ similar category organisation within my AGI’s cortex.  As usual the forty test patterns (phonemes, images, etc) are shown on the right; the confidence in recognition (height of the bar) is shown on the bottom left. Notice the regular modulated input pattern below the pattern input on the right. The cortex section has very high confidence in its recognition of the patterns until I click ‘A’ in the lower right to turn this regular injected pattern off. Then the cortex sections confidence drops/ stops… I have removed a facet of the overall pattern that the system was using to recognise the patterns. This is a kin to disconnecting the hippocampus or limbic system… it makes a big difference.

https://www.youtube.com/watch?v=AGFk_QddPqo (https://www.youtube.com/watch?v=AGFk_QddPqo)

Quote

I have to take exception to that

And so you should, that’s the point of the forum, to discuss theories and ideas.  I’m constantly revising/ improving my code/ theories and welcome any and all input/ ideas/ theories.

@8pla.net

Glad you like it and congrat’s on passing 700 lol.

 :)

Title: Re: Is short-term long-term algorithms what I think they are?
Post by: yotamarker on March 01, 2017, 08:13:43 pm

what do U mean by memory pointer ?

Title: Re: Is short-term long-term algorithms what I think they are?
Post by: LOCKSUIT on March 02, 2017, 02:01:07 am

Oh my the writing after one day! I am so busy! I'm inclined to read it, but man! Man!



Yous are confusing something.



You store draco-doodle-woman for the first time.

If you don't keep sensing it then it will weaken fast, and faster if weaker. This gives you short/long term. No difference between short/long term - correct. It's the same memory.

Once it's strong as hell, you forever will know the secret meaning behind draco-doodle-woman, but, there is a SECONDARY memory to play with, meaning you can be told draco-doodle-woman in 2 months, along with 9 car f f tea drive mr fox wand 8 8 e q d q, and forget the even just the one word/letter! again - If told h, you will forget if asked in 2 months! Yet forever know the letter h!



Memory pointer as in directing to a stored memory without wasting space, while also creating a secondary memory.

Title: Re: Is short-term long-term algorithms what I think they are?
Post by: Korrelan on March 02, 2017, 09:49:13 am

Quote

Yous are confusing something.

I was born confused… I am confused… and I will always be confused lol.

This is quite hard to explain; its doing me good to write this down lol.

Knowledge and memories are all part of the same structure, they are the same thing. Our consciousness is the pattern of activation running within this structure. 

I’ve shown this vid before but it’s a good example of how complex the ‘global though pattern’ is. Each pixel represents a neuron in the AGI’s frontal cortex.  Linked to the back of each pixel are connections (white matter tracts) to other areas of the overall cortex. The machine is sleeping and what you are looking at are patterns formed from memories/ experiences blending together; each memory is triggering other memories; the pattern constantly flows from one state of ‘thought’ to the next. At 6 seconds into the vid a blue (high activity) region appears (lower middle)… this was me talking. The machine was still listening even though it was sleeping… that activity influenced the whole pattern as it tried to make sense of the sensory input; a ‘thought’ pattern is very fluid and complex.

https://www.youtube.com/watch?v=C6tRtkyOAGI (https://www.youtube.com/watch?v=C6tRtkyOAGI)

When we use a piece of knowledge or a skill we also have access to when and where we learned it; sometimes it will effect our use of the knowledge (bedside manner for a doctor); how the use of the memory/ knowledge effects us is governed by the global thought pattern and what the main situation/ topic/ goal/ task is.  It’s our focus/ attention (also part of the global pattern) that dictates the effect of the memory/ knowledge on our current consciousness and what sections of the memory/ knowledge are relevant to the current task.

Any new memory is just the brain recording our current state of consciousness; to understand this moment in time we are using hundreds of ‘long term’ memories and knowledge.   The brain doesn’t record the incoming experience… it records its current understanding of it though its previous experiences/ knowledge.

We build our knowledge representations in a hierarchical manner; new knowledge is based/ built on our understanding of old knowledge.  We access our memories in a holographic manner; any part of a memory will trigger the rest. We have to have an understanding of the facets of the current experience/ knowledge to be learned to be able to form new learning/ memories. 

A young child has more synapse than you or I but could never grasp the concepts of this conversation because they have no prior experience/ knowledge to create their memories on/ from. 

Quote

but, there is a SECONDARY memory to play with

I don’t believe this to be true, we don’t need a second memory system.  When we are experiencing something we are activating thousands of long term memories/ knowledge; we have to be… just to understand the present.  Our understanding of the current experience is created from our ‘global thought pattern’ which is comprised of the parts of memories/ knowledge relevant to this moment of time… it’s the state of this pattern that new memories are formed from.

The brain tends to record only novel experiences; a new memory is formed from novel differences the brain has not encountered before.  This happens at the synaptic level and so is very difficult to relate to the consciousness level.

Anyway… so a short term memory is initially held in the ‘global thought pattern’ and laid down through weak synaptic links between the existing hierarchical structures that our knowledge base is comprised from. 

 :)

Edit: Arggg… I've just read through my posts… I really need to tie this description down… it shouldn't take me three posts to explain my concept of how human memory works lol.

Title: Re: Is short-term long-term algorithms what I think they are?
Post by: LOCKSUIT on March 02, 2017, 09:52:47 pm

1 sensory/motor memory.

Senses get fat (strong) or shrink quickly.

Our strongest fill up with energy fast and fire rapidly.

When you walk around you have Video-Sense. In ex. 5 seconds all the selected memories's secondary memory reservoirs now will begin to decrease their energy and never raise, as a *slope, but where they lay rather. The reservoirs are needed. And they won't become long-term either. I know car. Tell me car is the password. I'll forget it.

Post 800

Title: Re: Is short-term long-term algorithms what I think they are?
Post by: 8pla.net on March 03, 2017, 07:30:56 pm

Relating a reservoir, energy and a slope, Locksuit in part mentioned, "... reservoirs now will begin to decrease their energy and never raise, as a *slope" 

A surface of a reservoir is at a higher level than a surface further(for it to to be as a slope). By this theory, reservoirs are a source of energy.

Interestingly, with all things considered...  Notwithstanding a dearth of a scientific explanation, the result is an excellent theory for discussion,

Happy 800th.

Title: Re: Is short-term long-term algorithms what I think they are?
Post by: LOCKSUIT on March 03, 2017, 08:54:30 pm

VideoSense.

There's also FaintSense and FullSense. But I won't get into those.

Image a bunch of stored senses. Everywhere. In 3D. You have a clump. Like a ball in your hands. It's a little moist but not too bad.

Some of these senses are fat & strong.

But now imagine that 100 of the senses are shining bright like a star. Each a little less than the other. Like a linear slope of buildings. When you peak in you won't see a lineup of buildings though. They'll be all over. This is VideoSense. And when senses fire, it doesn't kill one of the VS reservoirs, it leaves it to decrease.

They draw searches. Enter as internal input. They'll echo in your head after hearing someone speaking.

It makes your Recognizers VideoSense-Recognizers. Big advantage. You can see your hand/clock-hand move. This requires many input images to match to such a memory (not 1 blurry image).

Title: Re: Is short-term long-term algorithms what I think they are?
Post by: LOCKSUIT on March 03, 2017, 10:00:49 pm

(Please read last page's reply)



Also we have a nice safeguard for figuring out

C0000

                                                                                    A0000

T000000000

Even if we tend to try to mentally figure out what the morse code someone tells us, the letters are still not the word "bulldozer", rather we can ONLY say the letters we GOT more faster together i.e. c a t. We only pop "cat" into our head when the letters fetch it.

I'm working on this. Give me a moment. I tend to make discoveries fluently like air i breath.

Title: Re: Is short-term long-term algorithms what I think they are?
Post by: 8pla.net on March 03, 2017, 10:52:45 pm

Locksuit said, "I tend to make discoveries fluently like air i breath." after discovering that   "You can see your hand/clock-hand move." is missing a dimension if it is, "Everywhere. In 3D." since,  time (clock) is 4D (the fourth dimension).

Title: Re: Is short-term long-term algorithms what I think they are?
Post by: LOCKSUIT on March 03, 2017, 11:08:26 pm

1. We are robots not 4D consciousnesses.

2. Importantly - the 3D clump of senses are each differently brighter than the other as the video sense, and as each frame is matched for vid-matching.

Also when I say p o p c o r n (ask a friend on your house) you begin to see it being built! Then it matches. Theoretically the sound could match the clip-recognizer but c a t sounds different a lot than cat. So there's my algorithm again! the place, test (recognize) then place here, repeat, until the letters are neatly placed on a paper and you see the word being built as you're told.

Title: Re: Is short-term long-term algorithms what I think they are?
Post by: Korrelan on March 03, 2017, 11:41:20 pm

Quote

We are robots not 4D consciousnesses

If your brain didn't reference/ incorporate the 4th dimension (time) you wouldn't even be able to read this sentence.

Reality/ life/ knowledge/ etc has a sequence/ order... sequence=time.... end of.

In my opinion incorporating time into a AGI design is extremely important.

 :)

Title: Re: Is short-term long-term algorithms what I think they are?
Post by: LOCKSUIT on March 04, 2017, 01:49:00 am

To recognize motion or all this Language you have suspect memories be hit multiple times. Ex. c a t. can hit the cat memory 3 times. Each ballerina spin frame matches that tv shows babe about 9 times that you catch from the wikipedia GIF.

To reference if someone asks what just did you can internally select those displaced slope of stars (videosense) OR match n link from "today".

VideoSense senses recently selected hold charge and echo.

My AI has a big sequence History-Bar in the center.

Title: Re: Is short-term long-term algorithms what I think they are?
Post by: 8pla.net on March 04, 2017, 05:14:49 pm

Quote from: korrelan on March 03, 2017, 11:41:20 pm

In my opinion incorporating time into a AGI design is extremely important.

Paradoxically,  narrow A.I. designs also depend on past events.

Title: Re: Is short-term long-term algorithms what I think they are?
Post by: 8pla.net on March 04, 2017, 05:25:17 pm

Quote from: LOCKSUIT on March 04, 2017, 01:49:00 am

To recognize motion or all this Language...

A model for representation of NLG (Natural Language Generation).