Open in TheBrain

HOLISM Bizarre Systems Monica April 18 2010

<p>Apr 19, 2010 10:15 PM

&#160;

Monica first played the clip &#160;PLAY &#160;by &#160;David Kaplan and Eric Zimmerman

&#160;

&#160;

list of &#160;scientists and writers contrib to the discussion of HOLISM &#160;(and holism vs reduc debate)

&#160;

including &#160;Teilhard de Chardin.

Schrodinger &#160;1946 ... &#160;What is life .... &#160;exclt slide with &#160;perhaps 30 scientisits.

&#160;

Reductionism is well named...

&#160;

REDUCE the problem by breaking it into components; &#160;main strategy of Sci Method.

REDUCE complexity by dividing it into parts.

reduce the # of variables by isolating the system from the environment in a lab.

Simple Explanation &#160;(to get a theory of everything)

&#160;

In contrast to HOLISM

When u reduce, &#160;u may discard the v thing you're looking for ... throw the baby out with the bathwater.

&#160;

When u split up the problem u throw away the info that only exists at the SYSTEM LEVEL

u throw away the EMERGENT&#160;

&#160;

System &#160;needs to be understood at the appropriate level and not below.

&#160;

4 &#160;features present in BIZARRE SYSTEMS

&#160;

CHAOTIC (massive feedback) &#160;IRREDUCIBLE &#160;(ridden with interactions)

Ambiguous &#160;(context dependent) &#160;EMERGENT &#160;(System level ... &#160;tranformative emergence)

&#160;

Appropriate for those &#160;problems are &#160;MFM &#160;MODEL FREE SYSTEMS

must be generated by experience &#160;(learned)

Work on &#160;bizarre systems

... no guaranteed optimal result; &#160;MFM &#160;are best effort only, fallible.

&#160;

The MFM &#160;ZOO

Discovery by Trial and Error &#160;(generate and test)

Gross Enumeration (rather than elegant proof) &#160;eg &#160;&#160;4 color theorem proof.

PATTERN RECOGnition

Learning; &#160;ADAPTATION &#160;eg &#160;Model Free adaptive controller &#160;(rather than PID controller)

EVOLUTION &#160;&#160;genetic algorithms

Language .. &#160;just ask someone, &#160;consultation or &#160;Google search.

&#160;

Example 1: &#160;Movie &#160;Recommendations

NetFlix &#160;Prize &#160;for &#160;&#62; &#160;10% &#160;improvement in accuracy.

she shows a simple &#160;MFM &#160;appraoch.

(what NOT to do... do NOT model movies or model &#160;customers or genres).

&#160;

She shows diagram &#160;(node and link graph) &#160;nodes are movies connected by

people: &#160;eg &#160;&#160;Bob &#160;Likes &#160;Matrix &#160;&#160;and Bob Likes Existenz...

so link from Bob to each of those.

&#160;

Now, &#160;just eliminate &#160;Bob and draw &#160;one arc &#160;(the Bob arc ... altho unlabelled)

connecting Matrix to Existenz... &#160;it means &#160;one customers &#160;rented both Matrix and Existenz.

If Sean also &#160;rented both, then there are TWO arcs connecting them.

&#160;

Now, &#160;let's say &#160;Bill &#160;has rented MATRIX and u want to give him recommendations....

just count the links from MATRIX... &#160;say &#160;there are &#160;17 links to Existenz and that is

the biggest number connecting Matrix to anything... then ur 1st recom is Existenz.

&#160;

(I prev. asked that same Q of Monica)

&#160;

Winning Team was an amalgamation of teams that used a combo of methods

that produced a &#160;10.!% &#160;improvement over baseline and won &#160;the Million dollar prize.

&#160;

weighted a variety of methods, &#160;eg. just stick with Oscar Winners.

&#160;

To Do &#160;MFM, u need Big Data

&#160;

Data Mining on massive DB.

&#160;

Note: &#160;Silicon Valley &#160;Bay Area &#160;ACM

&#160;

Continuous Industrial Processes... &#160;pipelines &#160;(hard to do)... weather,

robotics;

rational drug design is BS per Monica ... &#160;actually do &#160;generate and test.

&#160;

RFLP &#160;Reflep...

restrictive &#160;fragmentation &#160;length &#160;polymorphism &#160;to sequence &#160;DNA

or to idnetify genes &#160;eg &#160;in rare diiseases.

&#160;

In Science: &#160;generation of hypotheses is done holistically &#160;(I joke by LSD)

and testing only is done reductionistically.

&#160;

&#160;

&#160;

&#160;

&#160;</p>