Basic Architecture
Basic Architecture
raw data [from the system/states “under observation”] is provided to an “interface”
the interface goes to a “sensor” whose job is to aggregate raw data and reduce it to something much smaller
this goes to the “internal state” of the observer
Features:
Features:
The observer has to have some persistence
Minimal observer
Minimal observer
[What is an observer: it is a “””memory location””” where things get persistently stored ; i.e. it’s a thing that may change but where you can always have a pointer to the current state]
Single hyperedge / single cell in CA / ...
The observer has to be somewhat big so that the planks of the ship of Theseus can be replaced gradually... [“observer inertia”]
Within the observer there needs to be more rapid interaction between its parts, than with the outside world... i.e. the observer must maintain internal coherence ; it can’t be buffeted too much by the outside world
Number of memory states is tiny compared to the number of external states of the world... The sensor reduces the states of the world to something small enough to be “memorable”
You can either filter at initial sensing, or you can “store compete logs” and filter later... [But you always have to filter if you want a conclusion]
Mechanism
Mechanism
The mechanism is pruning ; the effect is equivalence classes
E.g. the Ising case: there is no contraction, but there are distinct equivalence classes.
E.g. the Ising case: there is no contraction, but there are distinct equivalence classes.
Minimal CA Observer
Minimal CA Observer
On the left is a “gas” with a certain density/pressure
In the middle is a wall that lets through only certain “pruned behavior”
On the right is a “memory” that records
In the middle is a wall that lets through only certain “pruned behavior”
On the right is a “memory” that records
Simplest case: let through 1 in 10 particles
[ When do we erase the memory ]
[ When do we erase the memory ]
Two levels of pruning:
1. The original sensor prunes down the initial data
2. The memory only remembers so much [e.g. an absorbing RH wall that erases old memories]
1. The original sensor prunes down the initial data
2. The memory only remembers so much [e.g. an absorbing RH wall that erases old memories]
[ First the irreversible case, then the reversible case ]
Analog-to-Digital Converters
Analog-to-Digital Converters
Constructed from successive comparators
In[]:=
GraphTree[KaryTree[15]]
Out[]=
What we will have is a decision tree....
[ This is similar to weights in a balance measuring a mass ]
Average [+Threshold]
Average [+Threshold]
Average finders will have Gaussian noise....