Top PDF Evolution and learning in artificial ecosystems

Evolution and learning in artificial ecosystems

Evolution and learning in artificial ecosystems

1 Introduction Stuart Wilson defined animats as a type of artificial an- imals, whose sole goal, from the individual’s perspec- tive, is homeostasis [Wilson, 1986]. He also suggested the animat path to AI as a way of creating artificial intelligence by modeling animal behavior, which is a wider notion than natural intelligence [Wilson, 1991]. In this paper we propose to follow the animat path to AI by simultaneously modeling two fundamental processes underlying animal intelligence: evolution, which oper- ates at the population level between generations, and learning, which operates at the individual level. Both these processes are fundamental to the ability of animal populations to adapt and survive in new environments. Ecological models that omit either evolution or learning
Show more

14 Read more

Skill learning and the evolution of social learning mechanisms

Skill learning and the evolution of social learning mechanisms

Developmental mechanisms are also an important con- sideration in the evolution of SE and OL. We implement SE and OL as tendencies that are fixed over a lifetime. However, animals could plausibly ‘learn to socially learn’ [55], including through domain-general mechanisms [54]. If such learning would be governed by immediate rewards, however, our analysis points to difficulties in how it would arise. To the extent that our model is realistic, it implies that the immediate direct effect of SE is a reduction in average rewards since foragers tend to become biased to resources for which they have low skill (Fig. 4d), while the direct effect of OL is a time cost and no reward. In principle, foragers could therefore learn to associate the choices of others with low rewards, in which case learn- ing to socially learn might be more challenging than it first appears, at least in a ‘skill learning’ context. Where this occurs, we would predict SE and OL to be underpinned by attentional or motivation biases, which, together with the aforementioned cognitive pre-requisites of OL, may be regarded as social learning adaptations. The question of whether the evolution of social learning requires the evo- lution of social-learning-specific adaptations remains one the major unresolved issues in the field, but it is likely that the answer will depend on the social learning mechanism involved.
Show more

19 Read more

Accelerating Chemical Discovery with Machine Learning: Simulated Evolution of Spin Crossover Complexes with an Artificial Neural Network

Accelerating Chemical Discovery with Machine Learning: Simulated Evolution of Spin Crossover Complexes with an Artificial Neural Network

Department of Chemical Engineering, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, United States * S Supporting Information ABSTRACT: Machine learning (ML) has emerged as a powerful complement to simulation for materials discovery by reducing time for evaluation of energies and properties at accuracy competitive with first-principles methods. We use genetic algorithm (GA) optimization to discover unconventional spin-crossover complexes in combination with e fficient scoring from an artificial neural network (ANN) that predicts spin-state splitting of inorganic complexes. We explore a compound space of over 5600 candidate materials derived from eight metal/oxidation state combinations and a 32-ligand pool. We introduce a strategy for error-aware ML-driven discovery by limiting how far the GA travels away from the nearest ANN training points while maximizing property (i.e., spin-splitting) fitness, leading to discovery of 80% of the leads from full chemical space enumeration. Over a 51-complex subset, average unsigned errors (4.5 kcal/mol) are close to the ANN ’s baseline 3 kcal/mol error. By obtaining leads from the trained ANN within seconds rather than days from a DFT- driven GA, this strategy demonstrates the power of ML for accelerating inorganic material discovery.
Show more

8 Read more

genomic and behavioural evolution in the artificial ecosystem simulation EcoSim

genomic and behavioural evolution in the artificial ecosystem simulation EcoSim

more realistic simulation. The use of IBM in ecology and evolution has been re- viewed by Grimm in 1999 and Lomnicki 1999 [18]. DeAngelis and Mooij presented another review study which focused on how the IBM field developed [15]. DeAn- gelis categorized the different directions along which to study individual variation in IBM into five different directions ”(a) spatial variability, local interactions and movement; (b) life cycle and ontogenetic development; (c) phenotypic variability, plasticity and behavior; (d) differences in experience and learning; and (e) genetic variability and evolution.” He also grouped the IBM systems into seven major study groups; movement through space, formation of patterns among individu- als, foraging and population dynamics, species interactions, local competition and community dynamics, evolutionary processes, management related processes. A book by Grimm and Railsback (2005) [19] provides a set of guidelines for building, testing, and analyzing individual-based models, updated in [20]. IBM has been used in many areas in ecology including forest ecology (e.g. [21]), fisheries and marine life (e.g. [22]), conservation biology and spatial heterogeneity (e.g. [23]). Many ecological IBM systems were not designed to be general platforms that could capture different aspects in ecology and evolution but rather these models answer specific question in their narrow domain. More group of evolutionary IBMs that were designed as platforms studying evolutionary behavior, emergence, adaptation and complexity are mention below.
Show more

144 Read more

Key Evolution In the Fields of Data Analysis and Artificial Intelligence

Key Evolution In the Fields of Data Analysis and Artificial Intelligence

As artificial intelligence devices end up being extra ubiquitous, or notable in specific industries, three skills requires observe. To start with, as day-to-day interactions along with artificial intelligence become the standard for lots of people, an essential understanding of making use of data as well as these devices will end up being an important resource needed through individuals of all ages and histories. Offering crucial principles in artificial intelligence at school may help ensure this. Second of all, to ensure that a stable of markets and also careers possess the absorbent capability to use machine learning in manner ins which work for them, brand-new devices are needed to have to create a pool of informed consumers or professionals. Thirdly, further support is needed to create state-of-the-art abilities in artificial intelligence. There is actually presently higher need for people along with innovative capabilities, along with experts in the business being actually extremely searched for, and also extra information to raise this talent swimming pool are critically needed to have. ‗ No regrets' intervene building electronic education and also updated users are going to likewise aid prep the UK for achievable improvements in the employment garden, as the fields of machine learning, artificial intelligence, and also robotics create. There is a huge range of prospective gain from additional uptake of artificial intelligence around industry markets, as well as the financial effects of this technology might participate in a core part in assisting to deal with the UK's efficiency space. Companies of all dimensions around sectors need to have access to appropriate support that aids them to understand the worth of data and also machine learning to their operations. To comply with the need for artificial intelligence all over industry markets, the UK is going to need to have to assist an active maker learning sector, which capitalises on the UK's strength around, as well as its family member international one- upmanships. The UK's start-up atmosphere has nurtured a number of prominent effectiveness stories in artificial intelligence, as well as key factor needs to
Show more

11 Read more

Evolution of artificial intelligence languages a systematic literature review

Evolution of artificial intelligence languages a systematic literature review

Keywords Artificial Intelligence; Programming Language; Python; AI; LISP; PROLOG; JAVA; C++; EOLC; ADA 1. Introduction Artificial intelligence (AI) is concerned with intelligent behaviors in artifacts such as perception, reasoning, learning, communicating and acting in complex environment. AI is concerned about machines possessing the listed characteristics as well as humans can, or even better and faster. The physical symbol system hypothesis states that a physical symbol system has the necessary and sufficient means for general intelligent action. A physical symbol system is a machine like a digital computer that is capable of manipulating symbolic data- adding numbers, rearranging lists of symbols and replacing some symbols by others 1
Show more

26 Read more

Imitation learning in artificial intelligence

Imitation learning in artificial intelligence

In the never-ending quest for Artificial Intelligence (AI), we take example from our- selves and our own intellect, since we are what we believe to be the most intelligent species on the planet. Intelligence however, did not spontaneously come into ex- istence, but was the result of a painstaking process of evolution [Bjorklund, 2006; Wynn, 1985; Sternberg, 1982]; even more interestingly, scientists argue that there exist multiple intelligences and not just one [Gardner, 2011]. How those intelligences arose is a topic biologists, geneticists and psychologists researching human intelli- gence have been working on for more than a century; yet their beliefs and theories directly affect computer scientists working on AI. Our focus are machines: robots, software and hardware, artificial artifacts, upon which humanity is trying to instill intelligence and make them as smart as humans. Yet we cannot dismiss how hu- man intelligence arose as outside the scope of AI, not only because it may be very relevant to the actual processes we’re trying to recreate, in that there may be cru- cial information in the emerge of intelligence in Homininae, information that could make Artificial General Intelligence (AGI) a reality (AGI as in, an irrefutably in- telligent, sentient and self-aware technological singularity [Goertzel and Pennachin, 2007; Kurzweil, 2005]).
Show more

191 Read more

genomic and behavioural evolution in the artificial ecosystem simulation EcoSim

genomic and behavioural evolution in the artificial ecosystem simulation EcoSim

more realistic simulation. The use of IBM in ecology and evolution has been re- viewed by Grimm in 1999 and Lomnicki 1999 [18]. DeAngelis and Mooij presented another review study which focused on how the IBM field developed [15]. DeAn- gelis categorized the different directions along which to study individual variation in IBM into five different directions ”(a) spatial variability, local interactions and movement; (b) life cycle and ontogenetic development; (c) phenotypic variability, plasticity and behavior; (d) differences in experience and learning; and (e) genetic variability and evolution.” He also grouped the IBM systems into seven major study groups; movement through space, formation of patterns among individu- als, foraging and population dynamics, species interactions, local competition and community dynamics, evolutionary processes, management related processes. A book by Grimm and Railsback (2005) [19] provides a set of guidelines for building, testing, and analyzing individual-based models, updated in [20]. IBM has been used in many areas in ecology including forest ecology (e.g. [21]), fisheries and marine life (e.g. [22]), conservation biology and spatial heterogeneity (e.g. [23]). Many ecological IBM systems were not designed to be general platforms that could capture different aspects in ecology and evolution but rather these models answer specific question in their narrow domain. More group of evolutionary IBMs that were designed as platforms studying evolutionary behavior, emergence, adaptation and complexity are mention below.
Show more

144 Read more

Evolution of communities of software: using tensor decompositions to compare software ecosystems

Evolution of communities of software: using tensor decompositions to compare software ecosystems

The identification of communities in software dependency networks that evolve over time is one of the main motivations of our work. Community detection in temporal, evolving or adaptive networks has largely attracted network scientists’ attention due to its important implications in the analysis of dynamical processes in complex networks, such as spreading and cascading dynamics, stability, synchronisation and robustness. Differ- ent types of methods and algorithms have been used, for example: the Louvain algorithm (Aynaud and Guillaume 2010), statistical null models (Bassett et al. 2013; Sarzynska et al. 2016), algorithms which exploit the historic community structure of the network (He et al. 2017; He and Chen 2015), Markov models (Rosvall et al. 2014), semidefinite programming (Tantipathananandh and Berger-Wolf 2011), gravitational relationship between nodes (Yin et al. 2017), and temporal matrix factorisation (Yu et al. 2017), amongst others. Machine learning techniques (Savi´c et al. 2019; Xin et al. 2017), genetic algorithms (Folino and Pizzuti 2014), consensus clustering (Aynaud and Guillaume 2010) and tensor fac- torisation (Araujo et al. 2014; Gauvin et al. 2014) have only recently been used for the detection of communities in temporal networks.
Show more

22 Read more

VISUAL SIMULATIONS, ARTIFICIAL ANIMALS AND VIRTUAL ECOSYSTEMS

VISUAL SIMULATIONS, ARTIFICIAL ANIMALS AND VIRTUAL ECOSYSTEMS

This review is about a field that does not traditionally belong to biological sciences. A branch of computer animation has its mission to create active self-powered objects living artificial lives in the theoretical biology zone. Selected work, of particular interest to biologists, is presented here. These works include animated simulations of legged locomotion, flexible-bodied animals swimming and crawling, artificial fish in virtual ecosystems, automated learning of swimming and the evolution of virtual creatures with respect to morphology, locomotion

8 Read more

CO-EVOLUTION OF BLOCKCHAIN AND ARTIFICIAL INTELLIGENCE

CO-EVOLUTION OF BLOCKCHAIN AND ARTIFICIAL INTELLIGENCE

Here also AI can provide more secure environment by integrating natural language processing, image recognition, and multi dimensional real time data transformation capabilities into Blockchain peer-to-peer linking. Also it provides better machine learning intelligence that provides more flexibility to the system.

8 Read more

Learning and decision-making in artificial animals

Learning and decision-making in artificial animals

the ability to form new nodes sometimes means the difference between life and death. The results obtained suggest that dynamic architectures are more powerful than static architectures. The animat model described was intended as a proof of concept. Clearly, it can be developed further in several directions. For instance, the structural learning rules could be improved and straightforward probabilistic rules that form memories at random moments could be added. Second, to enable symbolic learning and reasoning, a working memory and a rewrite engine could be added, possibly along the lines of (Strannegård et al., 2016). Third, the exploration techniques could be refined, e.g. by avoiding actions that are believed to bring strong punishment while exploring. Finally, the animat model needs to be tested in much more complex and challenging ecosystems. In particular, it must be tested in ecosystems with multiple animats.
Show more

29 Read more

FINDING THE RIGHT WAVELENGTH: SCALABLE ADVANCED LEARNING ECOSYSTEMS

FINDING THE RIGHT WAVELENGTH: SCALABLE ADVANCED LEARNING ECOSYSTEMS

That same instructor also worked with another professor in the design of a course in the OMSCS degree program, Knowledge-Based Artificial Intelligence. This course was the first of its kind to employ an AI-based tutor, “Jill” Watson, so-named because “she” was powered by IBM’s Watson (IBM, 2018) technology (Korn, 2016; Leopold, 2017). The course combined the Udacity (Udacity, 2018) MOOC platform with Georgia Tech’s Sakai LMS along with proprietary tools for peer feedback and for autograding the source code of computer programming projects. Again, the goal was to find that special combination of tools that the instructors believed would give students the best learning experience and therefore the best chance at success in the course.
Show more

12 Read more

Triple Helix and the evolution of ecosystems of innovation: the case of Silicon Valley

Triple Helix and the evolution of ecosystems of innovation: the case of Silicon Valley

While the number of seed and early stage deals is increasing—which is encouraging for new startups—the investment funnel in later stages is shrinking. As stated by J.S. Engel in the interview, we now see fewer deals but bigger share of the investment at ex- pansion and later stages, therefore, fewer companies are being funded with large amounts of money. We are also seeing less IPOs in the Silicon Valley. High valuations of some Silicon Valley Tech companies including Uber, Dropbox or Zenefits are chal- lenging their odds of a successful IPO, while their investors cannot cash out or even in- crease their investments. VC firms are also getting bigger with offices all around the U.S. Interestingly, those that did not have their headquarters in Silicon Valley back in 2008 they now do have them there. The number of deals made in 2016 by the most ac- tive firms is almost twice those in 2008, and more surprisingly two accelerators and some angel groups are on the top 15 in number of deals. As for the thematic areas of investment, the hottest ones in the U.S. in 2017 were artificial intelligence, cybersecu- rity and auto tech, but all of them saw a drop in invested dollars during the last quarter of the 2016.
Show more

21 Read more

An entropy model for artificial grammar learning

An entropy model for artificial grammar learning

The second measure to be included was the global associative strength one of Knowlton and Squire (1994, 1996). This is a meas- ure with a solid theoretical motivation (either as feature overlap or in terms of basic associative learning processes) and it has been explored in several AGL studies (e.g., Higham, 1997; Meulemans and van der Linden, 1997; Johnstone and Shanks, 1999; Pothos and Bailey, 2000; Lotz et al., 2009). Each test item has a global associative strength value, which is the average of the associative strength of all its bigrams and trigrams. The associative strength of a bigram or trigram is simply the average frequency with which it has been observed in training. For example, in computing the anchor associative chunk strength of string MSXVVR, we need to consider how frequently the following chunks appeared in training: MS, MSX, VR, VVR. A third measure, related to global associative strength, is anchor associative strength (e.g., see Knowlton and Squire, 1994, or Meulemans and van der Linden, 1997). The anchor associative chunk strength of a test item is computed in the same way as the global chunk strength, but taking into account only the bigrams and trigrams in the anchor positions of a string (the beginning and end of a string). The theoretical motivation for the anchor measure relates to the empirical finding that anchor chunks tend to be more salient to participants. For example, participants are more likely to identify NG strings if they violate the rules of the underlying finite state language in the anchor position (e.g., Reber and Allen, 1978).
Show more

14 Read more

mRNA display for the in vitro evolution of

artificial proteins and enzymes

mRNA display for the in vitro evolution of artificial proteins and enzymes

Artificial proteins and enzymes have the potential to aid in the production of pharmaceuticals and to facilitate basic biomedical research. Two methods currently exist for the development of artificial proteins: rational design and de novo selection. Rational design requires detailed knowledge of enzyme catalysis in order to design an enzyme active site in silico, and then introduce this active site into a protein. However, gaps in the understanding of protein folding and structure-function relationships make this approach challenging and far from routine. In contrast, laboratory evolution approaches to isolate artificial proteins and enzymes from libraries of variants are well established. In vitro selection techniques are powerful tools for the exploration of large areas of sequence space (up to 10 13 unique sequences) in the search for functional proteins and enzymes. mRNA display selection methods have only recently been developed, and the application of this technique for the engineering of de novo enzymes has not been fully explored. This thesis describes the establishment of an mRNA display platform for the selection and evolution of novel proteins and enzymes from large, high-diversity libraries. The synthesis of novel selection substrates are described that will facilitate the application of mRNA display to the selection of Diels-Alderase enzymes. A novel application of mRNA display is described for the solution-phase selection of protein- ligand pairs using interaction-dependent reverse transcription. Further development of this research could increase the throughput of ligand discovery to complement the pace at which new macromolecular targets of interest are being discovered.
Show more

146 Read more

Probabilistic machine learning and artificial intelligence

Probabilistic machine learning and artificial intelligence

There are at least two reasons this is not the case [87]. First, as we have seen, Bayesian nonpara- metric models have essentially infinitely many parameters, so no matter how much data one has, their capacity to learn should not saturate, and their predictions should continue to improve. A second reason is that many large data sets are in fact large collections of small data sets. For example, in areas like personalised medicine and recommendation systems, there might be a large amount of data, but there is still a relatively small amount of data for each patient or client, respectively. To customise predictions for each person it becomes necessary to build a model for each person—with its inherent uncertainties—and to couple these models together in a hierarchy so that information can be borrowed from other similar people. We call this the personalisation of models, and it is naturally implemented using hierarchical Bayesian approaches such as hierarchical Dirichlet processes [36], and Bayesian multi-task learning [88, 89].
Show more

24 Read more

Artificial Intelligence for Videogames with Deep Learning

Artificial Intelligence for Videogames with Deep Learning

As mentioned before, Tensorflow works with the structure of a data flow graph. In this structure, the data is moved through cells containing different operations to be applied to the data. In the Figure 2.1 we can see a visual representation of one Tensorflow graph obtained with TensorBoard, an open source tool contained inside of Tensorflow’s library for obtaining visual representations of graphs. In that figure, we can see how the data, organized into tensors, is moved through cells which contain mathematical operations, suchs as matrix multiplications and additions. Moreover, we can observe that some of those cells, called layers, are composed of smaller cells. This is because Tensorflow is based on the idea that any kind of operations, independently from its complexity, can be expressed as a group of basic functions and primitive operations. This approach has many advantages for the development of Deep Learning models, because it allows to calculate the derivative of each operation individually and apply the chain rule, which is very useful for the loss functions based on gradient descent methods, a common approach in modern Deep Learning. It also helps to parallelize the operations between different GPUs/CPUs and to divide the model into different paths for different purposes.
Show more

94 Read more

Artificial Ontogenies: A Computational Model of the Control and Evolution of Development

Artificial Ontogenies: A Computational Model of the Control and Evolution of Development

An alternative explanation for the non-appearance of many ‘theoretically’ possi- ble morphologies has been phrased in terms of developmental constraints: “biases on the production of variant phenotypes or limitations on phenotypic variability caused by the structure, character, composition, or dynamics of the developmental system” (Maynard Smith et al., 1985, p. 266). Developmental constraints, it is argued, alter the structure of variation on which natural selection acts, and can therefore affect the probability of evolution proceeding in particular directions, irrespective of an adaptive gradient (Arthur, 2000). This is in contrast to the conventional view of evolution, which assumes that all directions of evolution are equally likely prior to the consideration of an adaptive gradient (Figure 2.2). It is important to note that ‘constraint’ in this context was not intended to be inter- preted solely in its negative sense, as forbidding certain evolutionary directions. Developmental constraints may also operate in a positive fashion, by rendering certain evolutionary directions easier to achieve. The term ‘developmental bias’ has been suggested as a more inclusive alternative, incorporating both positive and negative effects on the direction of evolution (Arthur, 2004a).
Show more

231 Read more

Colonisation of artificial substrates in chemosynthetic ecosystems

Colonisation of artificial substrates in chemosynthetic ecosystems

1. Reducing habitats The first contact with marine chemosynthetic environments occurred in 1977 with the discovery of deep-sea hydrothermal vents in the Galapagos Rift, and since then in other areas around the world oceans (Van Dover et al. 2002; Van Dover and Lutz 2004). Besides hydrothermal vents, several examples of chemosynthesis-based habitats are now known, including cold seeps and large organic falls. They are associated with fluid emissions rich in hydrogen sulphide originated by inorganic process in the seabed or by microbial- mediated sulphate reduction. Some times, methane produced by organic matter reduction via biogenic or thermogenic processes is also available. These ecosystems based in microbial chemoautotrophic production attracted the biologists’ attention, particularly because of their exuberant metazoans communities and the establishment of unusual symbioses (Tunnicliffe et al. 2003; Tyler et al. 2003).
Show more

78 Read more

Show all 10000 documents...