Modeling Function of Nectar Foraging of Honeybees using Operant Conditioning

Transcription

1 International Journal o Computer Applications ( ) Volume 4 No.7, October 24 Modeling Function o Nectar Foraging o Honeybees using Operant Conditioning Subha Fernando Faculty o Inormation Technology University o Moratuwa Sri Lanka ABSTRACT Behaviors o the colonies o small unsophisticated agents have been analyzed in the literature with the purpose o developing eicient algorithms to solve complex, dynamic and burden problems in other societies. Among them, only a ew research have been conducted in the area o swarm cognition which tries to understand the cognitive behaviors exhibited by human brain by using the cognitive behaviors demonstrated by a colony as a sel-organized entity. In this aspect, the role o a neuron and a role o a insect have been equally considered as an unsophisticated agent which adjusts its actions according to the luctuations o local environment without knowing any global inormation. The cognitive behavior, such as eective labor division o honeybees at ood oraging process, was analyzed in this paper and has been exploited under operant conditioning. The paper has proposed a simple but eective computational model which demonstrates that, the positive reinorcement and the negative reinorcement in operant conditioning are the real actors that aect to the emergent o cognitive behaviors at swarm level when swarm is observed as a sel-organized entity. General Terms Artiicial Intelligence, Swarm Cognition Keywords Swarm Cognition, Operant condition, Honey Bee colonies.. INTRODUCTION A swarm is a large number o unsophisticated agents with limited capabilities, interacting cooperatively and locally among themselves, and environment, with no central control. These local interactions o the agents in a swarm allow the emergent o globally interesting behaviors that is necessary or their survival. The interaction o these social insects can be direct or indirect. Visual or audio contact, such as waggle dance and tremble dance, o honeybees are examples o direct communication while stigmergy or pheromones based communication between social insects are some examples or indirect communication[]. The collective intelligence behavior o these natural organizations such as ood oraging, division o labor, nest building, etc. are emerged in these colonies when the swarm is sel-organized. Sel organization is a set o dynamical mechanisms whereby global representation appears rom the interactions o its lower level components. The ingredients o sel-organism are multiple interactions which results in positive eedback and negative eedback that allows ampliication o random luctuations and control the evolution. This analysis o the natural sel-organized systems has been urther diversiied into the development o eicient algorithms when it comes to the discussion o how natureinspired sel-organized societies can be used to solve and optimize the complex problems in other societies. This swarm-based algorithms are capable o providing low cost, eicient, and robust solutions to solve complex problems in other inormation societies. Ant Colony Optimization (ACO)[2], Artiicial Bee Colony (ABC)[3], and Particle Swarm Optimization(PSO) are such signiicant swarm algorithms which have been introduced by observing the ants ood oraging, bee ood oraging and nest selection, and birds locking respectively. The principle o these algorithms are successully applied in image and data analysis, machine learning, operational research, and in inance and business applications[4]. Moreover many signiicant attempts[5,6] have been made in developing models to demonstrate the dynamicity o these societies, and to investigate how random luctuations o the key parameters o sel-organization aect to the decision making process o the swarm. In the ield o Swarm Cognition, the cooperative interactions o swarm that is necessary or the emergent o collective intelligence such as census decision making, inding the shortest path to the ood sources, etc are exploited in the perspective o understating higher cognitive behaviors in human beings. Swarm cognition[7] works on the basic postulation that a neuron as a part o the brain can be expressed in similar to a social insect as a part o colony. A neuron in isolation has very limited capabilities and depends only on local interactions, however, brain demonstrates highly complex cognitive process similar to what swarm displays as a colony. Sel-organism is the common mechanism that enables these simple units to display higher cognitive behaviors through the cooperative and local interactions between them. Thereore, behavior o a colony and cognitive process o a brain can be explained using simple rules o selorganization. In sel-organizing communities each individual either a insect or a neuron acts according to the inormation it receives rom its local community, without having any global representation, but by ollowing simple individual rules. 2. SELF-ORGANIZATION IN HUMAN BRAIN In order to understand how the key components o selorganization aect to the decision making process o human brain, we explored the swarm o honeybees ood oraging process as a decision making process, which demonstrates smart division o observers in the colony into oragers and receivers to have eicient nectar take in and storing mechanism. This decision making process can be characterized by the inormation which is processing through 45

2 International Journal o Computer Applications ( ) Volume 4 No.7, October 24 multiple local communication and the adaptive behavior o bees which adjust their responses according to their neighborhood environments[8]. This adaptation o honeybees is similar to the adaptation made by neurons when learning occurs at human brain. There a neuron adjust its internal metabolic processes according to its local environmental luctuations by updating its synaptic strength by creating new synaptic connections or by withdrawing existing synaptic connections[9]. Two key types o learning mechanisms have been discussed in the literature which deine the change o behavior according to the responses o the surrounding environments. They are: classical conditioning and operant conditioning. 2. Classical Conditioning Classical conditioning [] explains the learning as: it occurs through the associations between neutral stimulus and environmental stimulus. While in operant conditioning[] learning occurs as changes in behaviors o an individual that are results o the individual's responses to the events that occur in the environment. For example, assume a dog receives tasty oods, soon ater it hears a bell. The dog is very happy to receive the ood and he starts dancing by wagging his tail. Repeating this process many times, the dog starts dancing when he hears the bell. Pavlov[] explained this change o behavior o the dog under classical conditioning. He described the sound o the bell as the neutral stimulus, presentation o oods to the dog as environmental stimulus. The dance o the dog is called naturally occurring response. So by associating this neutral stimulus with the environmental stimulus, the sound o the bell could alone can make the dog dance. Thus, by building this association, the neutral stimulus becomes conditioned stimulus while the dance o the dog is the conditioned response. The conditioned response is the learned response to the neutral stimulus. By pairing and un-pairing this association between the neutral stimulus and environmental stimulus, the dog can be made to learn or unlearn the conditional behavior. 2.2 Operant Conditioning On the other hand, in operant conditioning, learning is described as a change o behavior that is resulted by causes o actions and its consequences [] [3]. According to Skinner, three types o operants can be identiied based on the type o responses o the environment that change the probability o the behavior being repeated: neutral operant (i the responses do not change the probability o the behavior being repeated), reinorces (i the responses increase the probability o the behavior being repeated and these reinorces can either be positive or negative), and punishers (i the responses decrease the probability o the behavior being repeated. The negative reinorce increase the probability when behavior being repeated is withdrawn). The key dierences between the two learning approaches are: in operant conditioning the learner is actively and voluntarily participated while in classical conditioning the learner is passively involuntary participated. In act many studies in the literature have thoroughly discussed the behavior o honeybees when they operate under classical conditioning[4] and rewarding mechanism[5]. In our research, we ocused on an extensive study about the adapting behavior o honeybees under operant conditioning than classical conditioning because the role played by honeybees in a colony is active and voluntary. This paper postulates that cognitive behaviors that are emergent rom a swarm as a sel-organized system can be explained using operant conditioning. And thereby the swarm behaviors can be used to explain the emergent o cognitive behavior at human brain. By doing so, we argue that positive reinorcement and negative reinorcement in operant conditioning are the key actors that control the selorganization o any dynamic system and the emergent o cognitive behavior. The paper is organized as ollows: section 3 describes the nectar oraging process o honeybee colonies, section 4 mathematically models the process described in section 3. And section 5 presents the results o the simple computational model that align to the model presented in section 4. Section 5 discusses and concludes our indings. 3. THE FUNCTION OF NECTAR FORAGING OF HONEYBEES The complexity and dynamicity o the nectar oraging o honeybees was analyzed in the means o identiying key parameters that are necessary or eicient nectar gathering process. It is well known that honeybees communicate through various communication channels such as using pheromones, or tactile dancing with or without some vibrating sounds. Among these techniques, waggle dance is the key communication channel that has been used by honeybees to inorm new rewarding lowering sites to the colony and to recruit new oragers to the newly ound lower patches[6]. The process o waggle dance with its dependent parameters in general ecology o nectar oraging o honeybees can be briely summarized as ollows: A honeybee colony is generally composed o a queen, workers (all workers are emale) and male or drones. The worker bees are almost completely responsible or caring hives, such as cleaning the hive, caring the larvae and youngs, eeding the queen and the drones, making honey, and gathering and storing nectar, water, etc. Thereore around 9% o the bees in the colony are workers, rom % o them are scouts who ind lower patches by searching[7]. A scout keep searching until its energy level is depleted or it inds lower-patches. I it ound a lower patch, it comes to the comb and unloads the nectar sources to receivers at the comb. The receiver takes it to the storage area o the hive. The returned scout can either be a scout again, or a orager, or an unemployed bee. I the returned scout eels that the lower patch rom where it brought the nectar is in high quality and quantity, it perorms waggle dance to recruit more oragers. I it is in considerable quality it simply returns and brings the nectar rom the lower patch. Otherwise, a scout can orget the visited lower patch and settle as an unemployed bee until it is recruited, or it can become a scout again searching or new lower patches. Sometimes orager bees perorm tremble dances to get unemployed bees to engage in nectar-receiving task [8, 9]. Key message that a orager wants to convey through this tremble dance is that it has ound more-rewarding nectar source and no enough receivers to unload them eiciently. Meanwhile by perorming this dance the orager tries to inorm other mate-oragers not to recruit additional oragers to their nectar sources. Thereore, the tremble dances o honeybees help to the colony to keep the balance between nectar storing and nectar take-in. Once oragers unloaded their nectar they may start to perorm waggle dance to recruit unemployed oragers to visit their explored lower patches. I 46

3 International Journal o Computer Applications ( ) Volume 4 No.7, October 24 a orager bee had to wait too long to unload the nectar (when it was unable to ind a receiver), then the orager bee does not perorm the waggle dance to recruit additional oragers bees, because it does not have enough receivers to handle the unloading process. The receiver bee may get delayed or searching urther away through the hive when the hive has not enough vacant storage cells, and it is almost ull. Waggle dance[6,6,2-22] is a communication behavior which conveys the inormation about location, and quality (high concentration, distance, easy to collect, etc) o the ood sources that have been ound. A dancing bee runs orward and perorms the waggle dance as shown in the igure, while she is on the run, she vibrates her abdomen laterally and then comes back to her starting point. According to the sources in the literature, a distance to the ood source is proportioned to the length o this waggle run and the angle o the run to the sun represents the direction o the ood sources. The higher the quality o nectar source, the higher the number o waggle runs per dancing bout and that increases the number o recruiters. Furthermore, the number o recruiters increases in proportion to the probability o returning orger would dance and the number o waggle runs made by her per visit. Through this mechanism honeybee colonies gather nectar more eiciently by sending their oragers to the better lower patches by abandoning less reward lower-patches, recruiting to more-rewarding patches, and searching or new patches. Fig : Waggle dance o a honeybee According to Seeley[2], honeybees measure the proitability o located nectar source by sensing the energetic eiciency o their oraging. Thereore, the number o waggle runs per dance is not directly a linear unction o closeness o the located nectar source to the hive but the energy expenditure per oraging trip. Moreover, through its waggle dance, a dancing bee reports on the current level o energy proitability o her orage site. Thereore, the number o waggle runs per bout is high when the nectar source is abundant, and it is low when it is scarce. The dancing bee does not perorm her dance in one place but distributes it over the dance loor. Thereore when number o orage sites are being reported on the dancing loor, oragers can easily take a random sample o the dance inormation. This allows the colony to allocate oragers to more-rewarding ood sources. 4. MODELING FOOD FORAGING FUNCTION OF HONEYBEES The emergent o cognitive behavior at swarm level, a decision making process which divides the labors at the colony eiciently to streamline the ood oraging process was analyzed in our research. This analysis was carried out to identiy the key actors or reinorcements that enable the swarm as a sel-organized system to demonstrate this highly cognitive task. Here, we have restricted our analysis by considering only the impacts o direct communication o honeybees on this oraging task. Further, we have assumed these bees can remember one lower patch at a time and can harvest only one lower per trip rom the comb as in [6]. Let mathematically model the world o swarm as square-grid o length m (m > and odd integer in meters) in which lower patches are placed on the environment according to a 2 Gaussian distribution N (, ) where is m/ 2 when the comb is placed at the middle o the square. Let the maximum distance a scout can ly without returning the comb is l >. Let quality o the nectar concentration at lower patches ( n U ) are uniormly distributed in three states: low ( unit), moderate (2 units) and high (3 units). Moreover, the available quantity o the nectar in a lower patch is measured in terms o the number o bloomed lowers it has. Let the number o bloomed lowers are randomly distributed using a discrete uniorm distribution rom to N ( n U ~ U(, N) ). Then, the overall quality o nectar a lower patch has, is deined by n n n. Honeybees, in a colony o size U N colony_size, are initially distributed to scouts, oragers, receivers, and observers with probabilities Pr s, Pr, Pr r, and Pro respectively. We adapted the equation rom[22] which measures the quality o the ood sources in terms d and d where d is the max distance rom the comb to a lower patch and d max is the maximum direct distance rom the comb to the edge o the square through the lower patch and, n - the overall quality o the nectar at a lower patch. We can deine the quality o a lower patch as in eq. (). As given in the equation, the higher quality lower patches which locate near to the comb gets higher Q. A scout honeybee starts Levy light to ind these lower patches. dmax Q d d max n () The number o waggle runs (w) perormed by a honeybee is based on Q and it is in the orm o w Q. A returned orager waits time t been brought-in to a receiver in a colony; t wait_ to _ unload wait_ to_ unload to unload nectar that have exp( t / number_ o _ receivers) where number_o_receivers is the number o receivers in the colony at time step t. The expected time duration a honeybee has to wait to unload nectar t wait is measured in terms o the quality o the lower patches that have been reported to the colony so ar. i.e. t.5 ( Q k). I the oraging honeybee has to wait wait t wait t wait _ to_ unload, the orager bee tends to perorm tremble dance. The duration o the tremble dance is t. The parameter k is wait t wait _ to _ unload determined by a honeybee by randomly sampling its neighborhood, i.e. the roughly taking the quality o the lower n patches that have been reported to the comb; so k Qi n. i 47

4 International Journal o Computer Applications ( ) Volume 4 No.7, October 24 I t wait t wait _ to_ unload, is a parameter, the orager stops oraging the current site. A given orager bee can turn either to a scout, or an observer. A orager perorms a waggle dance mainly based on Q, the duration o the waggle bout is restricted by i any tremble dances perormed at the colony. I orager bee encountered trem number o tremble dances that have being currently reported in the comb, then w Q trem/ n / c ; where n is the size o the random sample and c is a constant. A observer bee can turn to a orager, or a receiver or remain as a observer based on the dominant requirement that is exhibited by the random sample that it has taken rom the colony. A receiver bee can change its role to a observer i it has to wait too long to receive nectar, t wait_ to _ receive ; where is a parameter. Under this mathematical model, the energetic eiciency o the nectar source o lower patches oraged by honeybees inluence the entire nectar oraging cycle o the colony. Furthermore, this energetic eiciency, is the key actor that determines the duration o waggle runs perormed and in act it aects to the proper division o labors in the colony. Thereore, we can postulate that, energetic eiciency o the nectar source is the positive reinorcement that a honeybee receives to perorm waggle dance and to recruit more oragers, and the arises o cognitive unction which makes the decision on dividing the labors at the colony appropriately. On contrast, the time duration, a orager bee has to wait to unload its nectar source to a receiver is the negative reinorcement that a honeybee receives rom the environment to discourage the oraging o low quality lower patches compared to what have been reported to the colony. So we can postulate that these two orms o reinorcements rom the environment allow the entire colony to keep its balance between nectar take-in and storage process by eiciently dividing its labors to appropriate task according to the dynamic environmental luctuations (according to the distribution o the quality o the nectar sources that have been reported) and according to the inormation communicated by its surrounding environment to the honeybees. Thereore, a cognitive process, such as decision making which is exhibited by a sel-organized system, is mainly arisen as learning adjustments which are made by each individual as responses to environmental positive and negative reinorcements. 5. RESULTS A simple computer model was developed to simulate the eect o the positive ( Q ) and negative ( t wait_ to _ unload) reinorcements on the labor division at swarm ood oraging process. The total quality o a lower, i.e. Q (reerred as in igures), the mean o the quality o lower patches that have been really reported to the colony, i.e. k (reerred as meanqltyflower in igures), the time duration a honeybee waits to unload the nectar, i.e. t wait_ to _ unload (reerred as in igures) and the expected time duration a honeybee must wait to unload the nectar, i.e. t wait (reerred as m in igures) were monitored in this process in addition to the recording o the number o receivers, the number o oragers and the number o observers in the colony. The computer model had a colony o 5 oragers, 3 observers, and 2 receivers. The overall quality o a lower patch, i.e. Q - (,) was randomly generated. Initially the value o k was set to.3. I Q > k, a observer is turned to a orager, otherwise a orager is turned to an observer. I t, we increase the number o receivers wait t wait _ to _ unload by one and decrease the number o observers by one. Otherwise we decrease the number o receivers by one and increase the number o observers by one. With this simple model we were able to demonstrate the eect o positive and negative reinorcements on each individual's action and thereby the emergent o cognitive behavior at swarm level. From igure 2 to igure 5, each igure consists o three subigures, namely, (a), and. Subigure (a) demonstrates the distribution o honeybees in the colony in terms o the number o receivers, the number o oragers and the number o observers when all the our parameters (, meanqltyflower,, and ) were calculated as deined in the mathematical model. The subigures and show the luctuations o, and meanqltyflower, and meanwtime and related to the scenario depicted in subigure (a) respectively. Figure 2 shows the situation when we let the system to run around steps without enorcing any other constraints. As shown in igure.2, the increases o the number o oragers have decreased the number o observers and has increased the number o receivers. As shown in igure.3. we artiicially set the value o meanqtyflower to. So that we artiicially inormed the colony that the colony has been reported by high quality lower patches which are in the range o. As shown in igure 3, since the quality o the lower patches that have been really reported by the honeybees are less than this artiicial value, o oragers have been increased compared to the expected waiting time, i.e. m. In order to observe behaviors o the colony or lower quality lower patches, we artiicially set meanqtyflower to. So that we artiicially inormed the colony that it has been reported by low quality lower patches and it is in the range o. As shown in igure 4, since the quality o the lower patches that have been really reported by the honeybees are higher than this artiicial value, honeybees were not requested to wait too long to unload the nectar and in act the time they really waited is less than the expected time duration, i.e. < meanwtime. This adjustment o the system has decreased both the number o observers and the number o oragers and consecutively it has increased the number o receivers. Finally, we evaluated the behavior o the system by artiicially setting mtime to.7 as shown in igure.5. As shown in igure 5. when is higher than the expected waiting time, i.e, the system has increased the number o observers, and decreased the number o receivers. When was less than the mwtime, the system has increased the number o receivers and decreased the number o observers. 48

5 Number o Honey Bees Number o Honey Bees International Journal o Computer Applications ( ) Volume 4 No.7, October 24 (a) meanqltyflower - Fig 2: Distribution o Honeybees in the Colony (a) meanqltyflower = mwanqltyflower Fig 3: Distribution o Honeybees in the Colony when meanqltyflower was artiicially set to 8 (a) meanqltyflower = meanqltyflower - Fig 4: Distribution o Honeybees in the Colony when meanqltyflower was artiicially set to 49

6 (a) =.7 International Journal o Computer Applications ( ) Volume 4 No.7, October meanqltyflower Fig 5: Distribution o Honeybees in the Colony when was artiicially set to.7 6. DISCUSSION As shown in igure.2 when the colony was reported by higher quality lower patches compared to what have being reported to the colony so ar, the system has increased the number o oragers, and the number o receivers while the number o observers have been decreased. This was an accepted behavior by the system which has encouraged the oragers who have reported higher quality lower patches by decreasing their real waiting time than the expected waiting time. The that has been reported to the colony has aected to the system as a positive reinorcement, as given igure.3. By increasing the number o observers in the colony, system has assumed that, it allows the oragers who have reported quality lower patches in range, to broadcast their oraging sites eectively to the colony. So the oragers who have reported poor quality lower patches have been discouraged by increasing their real waiting time, i.e., than the expected waiting time, i.e.. Thereore, in this scenario, has been applied on the system as a negative reinorcement to discourage oraging o poor quality lower patches. When quality o the lower patches that have been reported is set to, the system has encouraged oragers who have reported high quality lower patches (igure. 4) by decreasing their real waiting time than the expected waiting time. So it has increased the number o receivers and the number o oragers and decreased the number o observers. Thereore, as expected, the system has considered as a positive reinorcement. Finally when we set to.7 as in igure.5. It has negatively impacted on the system behavior. That is, when is higher than the expected, it has urther decreased the number o receivers, and increased the number o observers, by assuming that, the system is being reported by low quality lower patches. And accordingly, when is less than the expected waiting time, it has increased the number o receivers and the number o oragers, while it has decreased the number o observers, by assuming that the colony has been reported by higher quality lower patches. Through this simple but eective simulation, we could show that the operant conditioning, which explains the responses made by a simple individual to the environmental positive and negative reinorcements as learning adjustments, is the underline phenomena o sel-organization o dynamic system and the exhibition o cognitive behavior, such as division o colony labors eectively at the nectar oraging process. Thereore, similar to the emergent o cognitive behavior at swarm level, the cognitive behavior emergent at human brain can also be explained using operant conditioning by taking the brain as a sel-organized system. As the uture work, the deeper analysis into the emergent o a particular cognitive behavior at human brain is needed to be exploited under operant conditioning as a means o discovering the underlying positive reinorcement and negative reinorcement. Doing so, the adjustment that happen at individual neuronal level to these reinorcements can be taken as the process that resulted in emergent o cognitive behavior. 7. REFERENCES [] Ahmed H. and Glasgow J. 22 Swarm Intelligence: Concepts, Models and Applications, Technical Report , School o Computing, Queen's University, Canada. [2] Blum C. 25 Ant colony optimization: Introduction and recent trends, Physics o Lie Reviews 2 (25) [3] Karaboga D. and Akay B. 29 A Survey: Algorithms Simulating Bee Swarm Intelligence; Artiicial Intelligence Review; 3 (), pp

7 International Journal o Computer Applications ( ) Volume 4 No.7, October 24 [4] Krink T. Swarm Intelligence - Introduction, EVALie Group, Department o Computer Science, University o Aarhus. [5] Myerscough M.R. 23 Dancing or a decision: a matrix model or nest-site choice by honeybees, Proc. R. Soc. Lond. 27(55) [6] Bailis P., Nagpal R. and Werel J. 2 Positional Communication and Private Inormation in Honeybee Foraging Models, in Swarm Intelligence, Eds. Springer Berlin Heidelberg, pp [7] Trianni V. and Tusi E. 2 Swarm Cognition and Artiicial Lie, Advances in Artiicial Lie. Darwin Meets von Neumann, Lecture Notes in Computer Science:5778: [8] Trianni V., Tusi E., Passino, E.M., Marshell J.A.R. 2 Swarm Cognition: an interdisciplinary approach to the study o sel-organizing biological collectives, Swarm Intelligence:5:3-8. [9] Abbott L.F. and Nelson S.B 2 Synaptic Plasticity: taming the beast, Neuroscience, vol.3.pp [] Cherry K, How Classical Conditioning Works, About.com Psychology, 5-Aug-24. [Online]. [] Skinner, B.F. (954). The science o learning and the art o teaching. Harvard Educational Review, 24(2), [2] Skinner, B.F. (95). Are theories o learning necessary? Psychological Review, 57(4), 93-2 [3] Cherry K, What s Dierent Between the Classical and Operant Conditioning?, About.com Psychology, 5-Aug- 24.[Online] [4] Cammaerts M. 24 Classical conditioning, temporal learning and spatial learning in the ant Myrmica sabuleti, Biologia, Bratislava, 59/2: [5] Gil M. 2 Reward expectations in honeybees, Communicative & Integrative Biology 3:2, 95-. [6] Seeley T.D., Mikheyev A. S., and Pagano G. J. 2 Dancing bees tune both duration and rate o waggle-run production in relation to nectar-source proitability, J. Comp. Physiol. [A], vol. 86, no. 9, pp [7] Yahya H. The miracle o the Honeybee 27, chapter 2-3. [8] Thom C 23 The tremble dance o honey bees can be caused by hive-external oraging experience, J. Exp. Biol., vol. 26, no. Pt 3, pp [9] Seeley T.D. 992 The tremble dance o the honey bee: message and meanings, Behav. Ecol. Sociobiol., vol. 3, no. 6, pp [2] Passino K.M. and Seeley T.D. 26 Modeling and analysis o nest-site selection by honeybee swarms: the speed and accuracy trade-o, Behav. Ecol. Sociobiol., vol. 59, no. 3, pp [2] Seeley T. D. 994 Honey bee oragers as sensory units o their colonies, Behav. Ecol. Sociobiol., vol. 34, no., pp [22] Okada R., Ikeno H., Kimura T, Ohashi M, Aonuma H and Ito E. Error in the Honeybee Waggle Dance Improves Foraging Flexibility, SCIENTIFIC REPORTS 4 : 475 IJCA TM : 5