Robust-first Computing Wiki - people:chris_symonds

abstract_outcome

anonymous@undisclosed.example.com (Anonymous) — Tue, 14 Oct 2014 05:48:11 +0000

@@ -3,6 +3,6 @@
  ===Revised Abstract===
- Computational models of biological evolution have shown that both parochial and altruistic behaviors increased the fitness of the host by working in concert with each other rather than co-evolving on their own beneficial merits. The benefits of these mutually reinforcing behaviors include better resources management within an in-group through the elimination of out-groups. We present new models in a robust computational system that allows for the evolution of parochial altruism. Populations of in-groups have the potential to evolve some combination of these behaviors to test whether the biological benefits gained from parochial altruism can find a correlate effect on fitness in a robust-first environment. However, the wide variation in resource availability within this computational system can give rise to "cooperative stalemates" among certain behavior categories. The new models display a dynamic interplay among these evolved behaviors as the populations compete for computational resources.
+ Computational models of biological evolution have shown that both parochial and altruistic behaviors increase the fitness of the host by working in concert rather than on their own beneficial merits. The benefits of these mutually reinforcing behaviors include better resources management within an in-group through the elimination of out-groups. We present new models that allow for the evolution of parochial altruism in a robust computational system that show whether the biological benefits gained from these behaviors can find a correlate effect on fitness in a robust-first environment. Wide variation in resource availability within this computational system can give rise to "cooperative stalemates" among certain behavior categories. The new models display a dynamic interplay among these evolved behaviors as the populations compete for computational resources.
  
  
  ===Introduction?===
  The paradigm of robust-first computing entails an infinitely scale-able environment in which elemental programs interact with each other to create emergent computational behaviors. The usefulness of such behaviors and the fitness of such elements can hold analogous properties to the system of biological evolution in the natural world. This paper will outline research that seeks to explore the evolution of parochial altruism in biological systems by modeling this behavior in a robust-first computational environment. We seek to determine whether the natural evolution of parochial altruism in the biological world, and the increased fitness that it brings, will find a correlate advantage to the elemental programs of robust-first computation, thereby increasing the overall fitness of those programs. We will create an elemental program that can take on one of four behavior categories: Parochial Altruists, Non-Parochial Altruists, Parochial Non-Altruists, and Non-Parochial Non-Altruists. We will allow these elements to interact with each other, both within an in-group of elements and other out-group elements, in a manner consistent with their behavior, and allow sexual reproduction that produces offspring programs with a genetic mix of the behavior patterns of both parents, subject to mutation, thereby modeling an evolutionary system. We will likewise be scaling a value of resource availability in the system from 10% - 50% of the occupied space to determine whether the presence or absence of abundant resources has a significant impact on the successful emergence of any one behavior category.

element_sytizen

anonymous@undisclosed.example.com (Anonymous) — Mon, 22 Sep 2014 02:27:17 +0000

@@ -1,5 +1,5 @@
- Element_Sytizen Specification
+ ====Element_Sytizen Specification====
  
     1. Sliders
        1.1. Team
           1.1.1. Up to 7 team colors to select: 
@@ -22,69 +22,73 @@
              1.3.1.1. 0 indicates that the value will be randomly selected from a uniformly distributed set of 
                       values from 1-16.
              1.3.1.2. [1,8] will be considered “non-parochial”
              1.3.1.3. [9,16] will be considered “parochial”.
- 2.	Properties
- 2.1.	u8 team; //team value
- 2.2.	u8 genes; //genetic makeup; bits 0-3 represent altruism, 4-7 represent parochialism
- 2.3.	s32 energy; //the resources of the element
- 3.	Behavior
- 3.1.	Initial Behavior
- 3.1.1.	Pay a metabolic cost by way of subtracting an amount from energy if energy <= 0, element deletes itself.
- 3.1.2.	Loop through the event window and track everything seen with its location.
- 3.1.3.	If there is res in the immediate neighborhood (distance of 1 from active element) consume 1 res and add an amount of energy to the existing pool.
- 3.2.	Genetic-driven behavior
- 3.2.1.	PA: Attack any single out-group elements in the neighborhood.
- 3.2.1.1.	Outcome is based on probability
- 3.2.1.1.1.	P[kPA] = 0.5
- 3.2.1.1.2.	P[kP-] = 0.6
- 3.2.1.1.3.	P[k-A] = 0.75
- 3.2.1.1.4.	P[k--] = 0.85
- 3.2.1.2.	Success
- 3.2.1.2.1.	Target element is deleted
- 3.2.1.2.2.	Active receives ½ of target energy value
- 3.2.1.3.	Failure
- 3.2.1.3.1.	Probability that active will be deleted
- 3.2.1.3.1.1.	P[dPA] = 0.5
- 3.2.1.3.1.2.	P[dP-] = 0.4
- 3.2.1.3.1.3.	P[d-A] = 0.25
- 3.2.1.3.1.4.	P[d--] = 0.15
- 3.2.2.	PNA: Attempt to steal from the least-threatening out-group agent in the neighborhood.
- 3.2.2.1.	Outcome is based on probability
- 3.2.2.1.1.	P[sPA] = 0.5
- 3.2.2.1.2.	P[sP-] = 0.6
- 3.2.2.1.3.	P[s-A] = 0.75
- 3.2.2.1.4.	P[s--] = 0.85
- 3.2.2.2.	Success
- 3.2.2.2.1.	Target element loses ¼ energy value
- 3.2.2.2.2.	Active receives target energy value lost
- 3.2.2.3.	Failure
- 3.2.2.3.1.	Small probability that active will be deleted
- 3.2.2.3.1.1.	P[dPA] = 0.25
- 3.2.2.3.1.2.	P[dP-] = 0.2
- 3.2.2.3.1.3.	P[d-A] = 0.15
- 3.2.2.3.1.4.	P[d--] = 0.1
- 3.2.3.	NPA: Evaluate the needs of each agent in the neighborhood and share based on cost/benefit
- 3.2.3.1.	Range is immediate neighborhood
- 3.2.3.2.	If target.energy < active.energy then share (Active.energy-Target.energy)/2
- 3.2.4.	NPNA: No actions
- 3.3.	Reproduction
- 3.3.1.	If multiple potential mates exist within the event window, choose uniformly at random
- 3.3.2.	Mate must be from same team
- 3.3.3.	Offspring will appear in any available location within the event window. If no such free location is available, reproduction is canceled.
- 3.3.4.	Active.energy and mate.energy must be > 20. Each will lose energy/3
- 3.3.5.	Offspring energy = active.energy/3 + mate.energy/3
- 3.3.6.	Offspring genetics inheritance
- 3.3.6.1.	Altruism value selected at random from one parent with normal distribution
- 3.3.6.1.1.	Subject to 0.5 chance of mutation within +-2 points
- 3.3.6.2.	Parochialism value selected at random from one parent with normal distribution
- 3.3.6.2.1.	Subject to 0.5 chance of mutation within +-2 points
- 3.3.7.	Team color inherited from parents
- 3.4.	Movement
- 3.4.1.	The event window is checked for the nearest RES location.
- 3.4.2.	In the absence of available RES, each type will check for the nearest genetically favorable conditions in the event window, if such conditions exist.
- 3.4.2.1.	PA looks for outsider team members
- 3.4.2.2.	PNA looks for outsider team members
- 3.4.2.3.	NPA looks for other team members but will avoid P from outside groups
- 3.4.2.4.	NPNA will diffuse if no RES detected
- 3.4.3.	A single-step move is made, biased towards the preferable location
+    2. Properties
+       2.1. u8 team; //team value
+       2.2. u8 genes; //genetic makeup; bits 0-3 represent altruism, 4-7 represent parochialism
+       2.3. s32 energy; //the resources of the element
+    3. Behavior
+       3.1. Initial Behavior
+          3.1.1. Pay a metabolic cost by way of subtracting an amount from energy if energy <= 0, element 
+                 deletes itself.
+          3.1.2.	Loop through the event window and track everything seen with its location.
+          3.1.3.	If there is res in the immediate neighborhood (distance of 1 from active element) consume 1 
+                 res and add an amount of energy to the existing pool.
+       3.2. Genetic-driven behavior
+          3.2.1. PA: Attack any single out-group elements in the neighborhood.
+             3.2.1.1. Outcome is based on probability
+                3.2.1.1.1. P[kPA] = 0.5
+                3.2.1.1.2. P[kP-] = 0.6
+                3.2.1.1.3. P[k-A] = 0.75
+                3.2.1.1.4. P[k--] = 0.85
+             3.2.1.2. Success
+                3.2.1.2.1. Target element is deleted
+                3.2.1.2.2. Active receives ½ of target energy value
+             3.2.1.3. Failure
+                3.2.1.3.1. Probability that active will be deleted
+                   3.2.1.3.1.1. P[dPA] = 0.5
+                   3.2.1.3.1.2. P[dP-] = 0.4
+                   3.2.1.3.1.3. P[d-A] = 0.25
+                   3.2.1.3.1.4. P[d--] = 0.15
+          3.2.2.	PNA: Attempt to steal from the least-threatening out-group agent in the neighborhood.
+             3.2.2.1. Outcome is based on probability
+                3.2.2.1.1. P[sPA] = 0.5
+                3.2.2.1.2. P[sP-] = 0.6
+                3.2.2.1.3. P[s-A] = 0.75
+                3.2.2.1.4. P[s--] = 0.85
+             3.2.2.2. Success
+                3.2.2.2.1. Target element loses ¼ energy value
+                3.2.2.2.2. Active receives target energy value lost
+             3.2.2.3. Failure
+                3.2.2.3.1. Small probability that active will be deleted
+                   3.2.2.3.1.1. P[dPA] = 0.25
+                   3.2.2.3.1.2. P[dP-] = 0.2
+                   3.2.2.3.1.3. P[d-A] = 0.15
+                   3.2.2.3.1.4. P[d--] = 0.1
+          3.2.3. NPA: Evaluate the needs of each agent in the neighborhood and share based on cost/benefit
+             3.2.3.1. Range is immediate neighborhood
+             3.2.3.2. If target.energy < active.energy then share (Active.energy-Target.energy)/2
+          3.2.4. NPNA: No actions
+       3.3. Reproduction
+          3.3.1. If multiple potential mates exist within the event window, choose uniformly at random
+          3.3.2. Mate must be from same team
+          3.3.3. Offspring will appear in any available location within the event window. If no such free 
+                 location is available, reproduction is canceled.
+          3.3.4. Active.energy and mate.energy must be > 20. Each will lose energy/3
+          3.3.5. Offspring energy = active.energy/3 + mate.energy/3
+          3.3.6. Offspring genetics inheritance
+             3.3.6.1. Altruism value selected at random from one parent with normal distribution
+                3.3.6.1.1. Subject to 0.5 chance of mutation within +-2 points
+             3.3.6.2. Parochialism value selected at random from one parent with normal distribution
+                3.3.6.2.1. Subject to 0.5 chance of mutation within +-2 points
+          3.3.7. Team color inherited from parents
+       3.4. Movement
+          3.4.1. The event window is checked for the nearest RES location.
+          3.4.2. In the absence of available RES, each type will check for the nearest genetically favorable 
+                 conditions in the event window, if such conditions exist.
+             3.4.2.1. PA looks for outsider team members
+             3.4.2.2. PNA looks for outsider team members
+             3.4.2.3. NPA looks for other team members but will avoid P from outside groups
+             3.4.2.4. NPNA will diffuse if no RES detected
+          3.4.3. A single-step move is made, biased towards the preferable location

graphs

anonymous@undisclosed.example.com (Anonymous) — Mon, 17 Nov 2014 05:37:05 +0000

@@ -1 +1,12 @@
+ =====Graphs and Data=====
+ Below are some of the results of the Paralta simulation
  
+ A typical simulation run with the genetic behavior threshold set to 250 = ~50% energy
+ 
+ {{:people:chris_symonds:250_01_team_one_graph.png?direct&300|}}
+ {{:people:chris_symonds:250_01_team_two_graph.png?direct&300|}}
+ 
+ 
+ Overall results from all simulation runs
+ 
+ {{:people:chris_symonds:graph_2.png?direct&300|}}

model

anonymous@undisclosed.example.com (Anonymous) — Sun, 16 Nov 2014 23:07:00 +0000

@@ -30,9 +30,9 @@
  
  ===Simulation Events===
  At the beginning of an event, a Sytizen will first pay a metabolism cost of 1 energy. If this depletes the energy to 0, the Sytizen is removed and the event ended. The neighborhood of the Sytizen is then scanned, and the location and type of any Element or empty space is recorded within a Manhattan distance of four. The Sytizen will then consume one neighboring RES within two Manhattan distance from them. A consumed RES bestows 10 energy on the Sytizen. Then, any genetic-specific behavior is performed as described above. 
  
- If the Sytizen is still alive and their total energy is greater than 30, they will attempt to procreate. Any in-group members within a Manhattan distance of four are sorted by energy. If the highest energy amount is greater than 40 RES, that Sytizen is selected to breed. If an empty location is available, an offspring will appear with a starting energy of 20, and the energy of each parent reduced by 10. The offspring will inherit one of the parent's Altruism values and Parochial values with 0.5 probability each. Thus, a PA and an NPNA could produce any of the 4 categories of phenotype. Once the genes are established, the parochial and altruism values are each subject to mutation with probability 0.02. If a mutation occurs, the new value is increased or decreased by 1 with 0.5 and 1-0.5 probability respectively.
+ If the Sytizen is still alive and their total energy is greater than 30, they will attempt to procreate. Any in-group members within a Manhattan distance of four are sorted by energy. If the highest energy amount is greater than 40 RES, that Sytizen is selected to breed. If an empty location is available, an offspring will appear with a starting energy of 20, and the energy of each parent reduced by 10. The offspring will inherit one of the parent's Altruism values and Parochial values with 0.5 probability each. Thus, a PA and an NPNA could produce any of the 4 categories of phenotype. Once the genes are established, the parochial and altruism values are each subject to mutation with probability 0.02. If a mutation occurs, the new value is increased or decreased by 1 with 0.5 probability.
  
  Finally, the Sytizen engages in movement. All Sytizens are movement-biased towards any RES visible to them in a given event. The normalized locations of any RES within manhattan distance four are summed, and a vector is determined from the average of those locations. They take a single step, and the event is concluded.
  
  ===Methodology===
@@ -40,5 +40,5 @@
  
  [[people:chris_symonds:model:Draft_1|Model Description Draft 1]]
  
  =====Figure 1=====
- {{:people:chris_symonds:sytizens_v02.png?direct&300|}}
+ {{:people:chris_symonds:fig_01_v01.png?direct&300|}}

news_archive

anonymous@undisclosed.example.com (Anonymous) — Fri, 05 Dec 2014 22:11:01 +0000

@@ -1,5 +1,16 @@
  ===== News Archive =====
+ 
+ ===05-Dec-2014 03:09:49PM-0700===
+ 
+ [[:people:chris_symonds:project_submission|Final Project Page]] has been posted.
+ 
+ ===15-Nov-2014 01:09:07PM-0700===
+ Ok, so here's where I'm at:
+ 
+ After a conversation with Prof. Ackley, it was concluded that my model needed a lot of simplification. So back to the drawing board I went, and implemented the changes he suggested. That done, and several wasted simulation runs later during which I realized I had some implementation errors in my model, I've gone back to running simulations. Now, instead of varying the amount of resources on the board, I am varying the threshold by which the Sytizens will engage in their genetic behavior. Parochials now have a single behavior. Altruists have another. Thus, PAs will engage in both. PNAs and NPAs will engage in one or the other, and NPNAs will engage in neither. Both genetic behaviors are tied to an energy threshold, below which they will not fire. For example, if set to 20, no Sytizen with less than 20 energy will share with or attack another Sytizen. This threshold is now what I am adjusting over runs. And holy crap, __I think I have a U-shaped curve__.
+ 
+ The bad news is, this has placed me very much behind. Now, my model description must be redone. My figure 1 redone. And I have many more lengthy runs to do before I can produce graphs 1 and 2. The bright side is that I am pretty much working solely on this over the weekend, but I wonder if that will be enough time...
  
  ===08-Nov-2014 10:05:54AM-0700===
  I completely forgot to make an update last week, so I guess I'll lump it in with this weeks, and attempt to keep it concise.

preliminary_results

anonymous@undisclosed.example.com (Anonymous) — Mon, 20 Oct 2014 05:21:12 +0000

@@ -18,5 +18,11 @@
  ====Implementation====
  There have been a great deal of compromises had over the course of the implementation. While I have certainly adhered in the general to the overal trajectory of my project, the area that took the most compromise was in the details. In retrospect, this might be to the better, as the less complicated my simulation is, the less convoluted the results. The details I'm specifically referring to is in the genetic behavior of the Sytizens. Intitially, the probability parameters for attackng, stealing, giving were far more nuanced. Now, for example, it is a straight 1 in 2 chance of killing an opponent, regardless of their condition. Likewise, Altruists no longer convert other team members, thereby likely destroying one of the few benefits they enjoy. Consequently, you see very few altruists in my simulation. Finally, parochial altruists no longer form roving gangs to terrorize other out-groups. However, they seem to be doing just fine on their own. I am glad to see that sexual reproduction appears to be sustainable; I was worried that would not work out very well.
  
  ====Testing and Debugging====
- This is a short section. Debugging was very hard, however became much easier once I learned how to write out to LOG.debug. This helped me to help myself, however I did still need to reach out from time to time whenever I needed to ask something of the MFM at large, outside the behavior of my Sytizens. I'm still not convinced that the whole thing is even working correctly, but I get data out of it, so that's a start.
+ Debugging was initially very hard, however became much easier once I learned how to write out to LOG.debug. This helped me to help myself, however I did still need to reach out from time to time whenever I needed to ask something of the MFM at large, outside the behavior of my Sytizens. I'm still not convinced that the whole thing is even working correctly, but I get data out of it, so that's a start.
+ 
+ ====Observation and Analysis====
+ To me, the biggest surprise was the dominance of the parochial non-altruist and the locked-in scenarios that would arise from this. These are the Sytizens that will steal from out-group members at a potential risk to themselves. I had to make the action of stealing particularly risky, perhaps more risky than the act is worth, just to keep them from dominating the map. In all, this makes me feel like the simulation itself is rather arbitrary in it's parameters, and the rather non-scientific way I implement these genetic behaviors speaks more to what I see than biological evolutionary tendencies. The fact that parochial altruists are now the dominant force, though consistent with previous findings, strikes me as less of a surprise. A good defense still seems to be a good offense, and the first population that generates a dominant group of PAs tends to win the day. It's not clear to me that varying the resource availability will affect this at all. So while I continue to run simulations and gather results, I feel mired somewhere between feeling like I may have implementation bugs, or worse, model bugs. 
+ 
+ ====Going Forward====
+ I will continue to adjust the simulation parameters to allow myself more flexibility in resource control. Once this is finalized, I will turn my attention back to the genetic behaviors and see how they are now different, and make a determination then as to whether I need to revisit implementation of that behavior. Oh, and at some point stop and collect useful data.

project_submission

anonymous@undisclosed.example.com (Anonymous) — Mon, 08 Dec 2014 20:41:01 +0000

@@ -1,15 +1,15 @@
  {{http://cs.unm.edu/~csymonds/CS591/Paralta_csymonds_final.pdf|Final Paper}}
  
  Final Presentation:
     *[[http://cs.unm.edu/~csymonds/CS591/Presentation.pptx|.pptx]]
-    *[[http://cs.unm.edu/~csymonds/CS591/20141122031816.mp4|.mp4]]
-    *[[http://cs.unm.edu/~csymonds/CS591/Presentation.odp|.odp]]
     *[[http://cs.unm.edu/~csymonds/CS591/Presentation.pdf|.pdf]]
+ 
+ [[http://youtu.be/vabfdEfDM1g|Project Video (YouTube, ~5 mins)]]
  
  [[http://cs.unm.edu/~csymonds/CS591/data/|Project Data]]
  
  [[http://cs.unm.edu/~csymonds/CS591/MFMv2_csymonds.tar|Modified MFM (Includes Element data) .tar]]
  
  [[http://cs.unm.edu/~csymonds/CS591/source/|Source code for Element_Sytizen and scripts I used]]
  
  [[http://cs.unm.edu/~csymonds/CS591/2007ScienceParochialAltruismWar.pdf|Choi and Bowles Paper]]

project

anonymous@undisclosed.example.com (Anonymous) — Sat, 13 Sep 2014 19:46:42 +0000

@@ -53,7 +53,13 @@
  Reproduction will involve selecting either parent's Altruistic and/or Parochial values at random, subject to some probability of mutation. Thus, a child could receive one or both of either parents values, potentially modified. Starting resources will likewise be a shared portion of both parents at a fixed amount. They immediately enter the world as teenagers, striking out on their own with no interest in cultivating a relationship with either parent and will probably get into drugs or run off to Hollywood or some similar nonsense.
  
  ====Questions Examined====
  
- While the model I am attempting to create has roots in the work of Choi and Bowles, I am hoping to model a pre-paleolithic society; one more nomadic wherein the in-groups do not pool and share resources, nor do they engage in outright warfare with other groups. Rather, the interactions among agents are restricted to the individual territory of the agents themselves, reflecting a period before social coagulation took hold with early hominids, roughly 3-5 million years ago. ((Edward H. Judge & John W. Langdon. //Connections: A World History, Combined Volume, 2/E//. Pearson; 2 edition 2011)) This muddies the waters a little bit on what it exactly means to be an in-group, though Judge and Langdon point out that the process of forming groups was gradual would be on the order of millions of years before societies were more firmly established.
+ While the model I am attempting to create has roots in the work of Choi and Bowles, I am hoping to model a pre-paleolithic society; one more nomadic wherein the in-groups do not pool and share resources, nor do they engage in outright warfare with other groups. Rather, the interactions among agents are restricted to the individual territory of the agents themselves, reflecting a period before social coagulation took hold with early hominids, roughly 3-5 million years ago. ((Edward H. Judge & John W. Langdon. //Connections: A World History, Combined Volume, 2/E//. Pearson; 2 edition 2011)) This muddies the waters a little bit on what it exactly means to be an in-group, though Judge and Langdon point out that the process of forming groups was gradual and it would be on the order of millions of years before societies were more firmly established.
  
- I suspect that altruism and parochialism were both still in the very early stages of development in this period and it would be interesting to see how each might have come about, and what the conditions were that made their appearance more probable. 
+ I suspect that altruism and parochialism were both still in the very early stages of development in this period and it would be interesting to see how each might have come about, and what the conditions were that made their appearance more probable. Will the simulation results reflect what Choi and Bowles propose, that Altruism and Parochialism reinforce each other? 
+ 
+ ===So What? (Computer Science Edition)===
+ 
+ So what is the computational tie-in to all of this? Why do we, as computer scientists, care about how altruism and parochialism came about under pressure? And under traditional computational paradigms, these are legitimate questions. Processes don't attack other processes for the purposes of resource competition (they do for other reasons). They might be included in an group, and a group of processes belonging to a virus might very well be considered an out-group, but does the concept of sacrifice mean anything to a computational process? I would argue no, under the traditional computational paradigm of efficiency-first computing, the concept of altruism doesn't apply. 
+ 
+ Under robust-first computing however, under the resource-limited confines of the MFM, it might. If a program needs space to operate, it may very quickly need to start making sacrifices if that space is limited. If it is competing for space with other process groups, it may need to be aggressive. What constitutes the fitness of a program in a robust-first environment? Perhaps Parochial Altruism might be a successful behavior to engage in.