+ All Categories
Home > Documents > Evolutionary Approach to Investigations of Cognitive Systems Vladimir Red ’ ko a), Anton Koval ’...

Evolutionary Approach to Investigations of Cognitive Systems Vladimir Red ’ ko a), Anton Koval ’...

Date post: 13-Jan-2016
Category:
Upload: anne-heath
View: 213 times
Download: 0 times
Share this document with a friend
Popular Tags:
13
Evolutionary Approach to Investigations of Cognitive Systems Vladimir Red’ko a) , Anton Koval’ b) a) Scientific Research Institute for System Analysis, Russian Academy of Science, Moscow b) National Nuclear Research University “MEPhI”, Moscow
Transcript
Page 1: Evolutionary Approach to Investigations of Cognitive Systems Vladimir Red ’ ko a), Anton Koval ’ b) a) Scientific Research Institute for System Analysis,

Evolutionary Approach to Investigations of Cognitive Systems

Vladimir Red’koa), Anton Koval’ b)

a) Scientific Research Institute for System Analysis, Russian Academy of Science, Moscow b) National Nuclear Research University “MEPhI”, Moscow

Page 2: Evolutionary Approach to Investigations of Cognitive Systems Vladimir Red ’ ko a), Anton Koval ’ b) a) Scientific Research Institute for System Analysis,

Epistemological problem

Epistemological problem: why human logical thinking is applicable to cognition of nature?

To emphasize the problem, let us consider physics. The power of physics is due to effective use of mathematics. However, a mathematician makes logical inferences, proves theorems, basing on his mind, independently from physical world. Why are his results applicable to real nature, to real physical world?

To investigate problem, it is reasonable to analyze cognitive evolution (evolution of animal cognitive abilities), evolutionary origin of human logical thinking.

So, it is reasonable to model cognitive evolution

Page 3: Evolutionary Approach to Investigations of Cognitive Systems Vladimir Red ’ ko a), Anton Koval ’ b) a) Scientific Research Institute for System Analysis,

Sketch program: steps of modeling cognitive evolution

(from simple animal cognitive abilities to mathematical deductions):

1) Modeling of adaptive behavior of autonomous agents that have natural needs: food, safety, reproduction

2) Investigation of the transition from the physical level of information processing in nervous system of animals to the level of the generalized “notions”

3) Investigations of processes of generating causal relations in animal memory

4) Investigations of “logical conclusions” in animal minds. Comparison of animal “logic” with human logic

Page 4: Evolutionary Approach to Investigations of Cognitive Systems Vladimir Red ’ ko a), Anton Koval ’ b) a) Scientific Research Institute for System Analysis,

Model of several needs and motivations

(step 1 of the sketch program)

Page 5: Evolutionary Approach to Investigations of Cognitive Systems Vladimir Red ’ ko a), Anton Koval ’ b) a) Scientific Research Institute for System Analysis,

Several needs and motivations

Population of autonomous agents is considered. Any agent has the following needs: food, safety, reproduction.

Needs are characterized by motivations MF , MS , MR , and factors FF , FS , FR . Any time moment only one motivation is leading.

Agent control system is set of rules

Sk Ak

Rule weights Wk are adjusted by means of both reinforcement learning and Darwinian evolution of agent population.

Situation Sk : 1) activity of the predator in vicinity of the agent, 2) previous action of the agent, 3) current leading motivation of the agent.

Actions: 1) searching for food, 2) eating of food, 3) preparing for reproduction, 4) reproduction, 5) defence from a predator, 6) resting.

Page 6: Evolutionary Approach to Investigations of Cognitive Systems Vladimir Red ’ ko a), Anton Koval ’ b) a) Scientific Research Institute for System Analysis,

Several needs and motivations Scheme of choosing of leading motivation

TF , TS , TR are thresholds

MN is additional motivation (it becomes leading very rare)

Changes of the factor (FF , FS or FR ) corresponding to the leading motivation are rewards at reinforcement learning

Page 7: Evolutionary Approach to Investigations of Cognitive Systems Vladimir Red ’ ko a), Anton Koval ’ b) a) Scientific Research Institute for System Analysis,

Several needs and motivationsResults of computer simulations

Dynamics of factors

Dynamics of motivations

Cycles of agent behavior and chains of actions are observed

Page 8: Evolutionary Approach to Investigations of Cognitive Systems Vladimir Red ’ ko a), Anton Koval ’ b) a) Scientific Research Institute for System Analysis,

Model of formation of generalized notions

(step 2 of the sketch program)

Page 9: Evolutionary Approach to Investigations of Cognitive Systems Vladimir Red ’ ko a), Anton Koval ’ b) a) Scientific Research Institute for System Analysis,

Agent is searching for food in cellular environment

Agent

There are 10x10 cells. Portions of food are randomly distributed in 50 cells.

Agent control system is set of rules:

Sk Ak ,

Sk and Ak are situation and action.

Situation Sk : presence or absence of food in agent field of vision.

Actions Ak : moving forward, turning left/right, eating, resting.

Rule weights Wk are adjusted by means of reinforcement learning

Circles indicate agent field of vision. Arrow shows forward direction of agent

Page 10: Evolutionary Approach to Investigations of Cognitive Systems Vladimir Red ’ ko a), Anton Koval ’ b) a) Scientific Research Institute for System Analysis,

Formation of internal notions

1) food is here “eating”;

2) food is forward “moving forward”, then “eating” ;

3,4) food is right/left turning right/left, then “moving forward”, then “eating”;

5) there is no food in field of vision “moving forward”…

5 heuristics generalize selected rules:

Internal notions of the agent are formed:

1) food is here, 2) food is forward, 3,4) food is right/left, 5) there is no food in field of vision

Page 11: Evolutionary Approach to Investigations of Cognitive Systems Vladimir Red ’ ko a), Anton Koval ’ b) a) Scientific Research Institute for System Analysis,

Adaptive behavior of modeled “organisms”

Witkowski M. An action-selection calculus // Adaptive Behavior, 2007. V. 15. No. 1. PP. 73-97.

Butz M.V., Sigaud O., Pezzulo G., Baldassarre G. (Eds.). Anticipatory Behavior in Adaptive Learning Systems: From Brains to Individual and Social Behavior. LNAI 4520, Berlin, Heidelberg: Springer Verlag, 2007.

Vernon D., Metta G., Sandini G. A survey of artificial cognitive systems: Implications for the autonomous development of mental capabilities in computational agents // IEEE Transactions on Evolutionary Computation, special issue on Autonomous Mental Development, 2007. V. 11. No. 2. PP. 151-180.

Intelligent autonomous agents

Page 12: Evolutionary Approach to Investigations of Cognitive Systems Vladimir Red ’ ko a), Anton Koval ’ b) a) Scientific Research Institute for System Analysis,

Sketch program: steps of modeling cognitive evolution

1) Modeling of adaptive behavior of autonomous agents that have natural needs: food, safety, reproduction

2) Investigation of the transition from the physical level of information processing in nervous system of animals to the level of the generalized “notions”

3) Investigations of processes of generating causal relations in animal memory

4) Investigations of “logical conclusions” in animal minds. Comparison of animal “logic” with human logic

Page 13: Evolutionary Approach to Investigations of Cognitive Systems Vladimir Red ’ ko a), Anton Koval ’ b) a) Scientific Research Institute for System Analysis,

Conclusion

Comparing steps of the sketch program with our models and other works, it is possible to conclude that we can see some small fragments of a picture of cognitive evolution now, but we do not see the whole picture yet

Nevertheless, investigations of cognitive evolution are interesting and important


Recommended