10 Virtual reality and adaptive

Communications Through Virtual Technology:

Identity Community and Technology in the Internet Age

Edited by G. Riva and F. Davide, IOS Press: Amsterdam, 2001 - ? 2001, 2002, 2003

10 Virtual reality and adaptive

technology

Richard WALKER

Abstract. In modern societies and economies human beings are faced with

forms of “artificial complexity which are very different from those to which the

forms of complexity they face in more traditional societies. To handle their inter-

action with Complex Artificial Systems they require new technological tools.

These tools, it is argued, could be based on the same “adaptive technologies”

which we have used to assist us in our interactions with natural complexity, in

particular, Artificial Neural Networks, Evolutionary Computing, the Ecologi-

cal/Embodied approach to artificial cognition as well as swarm computing. The

essay suggests ways in which these technologies could be used to produce practi-

cal systems for Adaptive Artificial Perception and for the bottom-up simulation

of Complex Artificial Systems. The human interface to such a system, it is ar-

gued, could be provided by integration with soft Virtual Reality Technologies.

The paper describes a hypothetical “Market Browser” based on this strategy.

Contents

10.1 Natural and artificial complexity (150)

10.2 Adaptation as a strategy to manage complexity (151)

10.3 Adaptive Technologies (152)

10.3.1 Artificial Neural Networks (152)

10.3.2 Evolutionary computing (154)

10.3.3 Ecological and “embodied” approaches (154)

10.3.4 Collective or “Swarm” computing (156)

10.3.5 Learning from the current generation of adaptive technologies 156 10.3.6 Bottom-up simulation (157)

10.4 Technologies to handle artificial complexity (157)

10.4.1 Initial requirements for handling artificial complexity (157)

10.4.2 Artificial perception (158)

10.4.3 Data mining as primitive artificial perception (159)

10.4.4 Weaknesses in current data mining software/ further

requirements for artificial perception (159)

10.4.5 Adaptive Artificial Perception (160)

10.4.6 Simulating the behavior of artificial complex system (161)

10.5 Integrating Adaptive and Virtual Reality technologies (162)

10.6 A “Market Browser” using Adaptive Virtual Reality (163)

10.7 Conclusions – future prospects (164)

10.8 References (164)

10.1 Natural and artificial complexity

It is often claimed, that modern societies and economies are more complex than the socie-ties and economies of the past. This is, or should be, a controversial point. What is cer-tainly true however is that the forms of complexity modern human beings have to deal with are qualitatively different with respect to more “traditional” forms.

In traditional societies the main forms of complexity encountered by individuals are complexity in the physical, biological and social worlds. The complex systems humans have to deal with include predators, prey, the weather, animal and human disease and – perhaps a tougher challenge than anything else - complex social and political structures. None of this is simple. Yet in practice biological intelligence has proved to be remarkably effective in dealing with natural complex systems. Human beings (and other animals) rou-tinely produce adaptive behaviors which today’s artificial systems are very far from matching.

Modern societies however are different. In these societies individuals have to deal not just with natural but also with artificial complex systems such as markets, large techno-logical systems, and the World Wide Web. These systems are not necessarily more com-plex than natural complex systems. In cognitive terms however they pose new challenges. In particular:

? Human beings have poor sensory access to artificial systems. Our eyes allow us to evaluate the body language of an adversary (or a mate) or to judge and predict the movements of a predator but we have no sensory system equipped to sense a negative trend in market demand or a subtle malfunction in a nuclear power station. To detect such events we first need to process large sets of numerical and/or textual data. Hu-man beings have no inbuilt cognitive system to perform this task

? The rules governing the behavior of artificial systems can change very rapidly. The laws determining the behavior of natural complex system change – when they change at all - over periods of time far longer than the life of individuals. Whereas the dy-namic of a market or a nuclear power station can change from year to year – some-times from second to second.

Human beings and animals relate to natural complex systems directly via biological sensory-motor systems. With unnatural complex systems, on the other hand, they require technological interfaces to help them perceive the characteristics and behavior of these and to think about the consequences of their own actions. In other words, to deal with this new kind of Virtual Reality human beings require Virtual Reality Technology (VRT). Where what is meant is not just 3-D displays or data gloves but a complete set of tools fa-cilitating interaction with an increasingly virtual world. The tools which exist today, data mining software for instance, are still extremely primitive. In this paper I will describe ways in which it might be possible to create a new generation of Virtual Reality tools based on what I will call “Adaptive Technology”.

In Section 10.2 of this essay I will argue that from a mathematical point of view natural and artificial complexity are essentially similar. This implies that the techniques we have developed to handle natural complexity may have additional applications in Virtual Real-ity. In Section 10.3 I will briefly review some of the relevant technologies. These include Artificial Neural Networks, Evolutionary Computation, so-called Ecological and Embodied Approaches to Artificial Cognition and what, for want of a better name, I will call “collective” or “swarm” computing. I will also briefly describe the “bottom-up simulation” strategy which researchers have used to develop these technologies. In Section 10.4 I will speculate about ways in which we can use these different technologies

to handle artificial complexity. In section 10.5 I will suggest a possible strategy for integrating and conventional VRT. Section 10.6 describes a hypothetical “Market Browser” incorporating this strategy. Section 10.7 concludes.

10.2 Adaptation as a strategy to manage complexity

Despite their obvious differences natural and artificial complex systems have at least two key points in common.

? Both classes of system are, by definition, analytically intractable. There exists no simple mathematical formula or engineering solution which allows us to precisely predict their behavior.

? Both classes of system display statistical regularities in their behavior which humans and animals can detect and exploit. The way in which naturally intelligent systems acquire this ability is via adaptation.

Biologists portray adaptation as a process of exploration on an adaptive landscape[1] (see Figure 10.1). Imagine that an organism’s cognitive system can be described by n pa-rameters. The ability of the organism to adapt to a given environment is described by an additional variable. In this set up every possible “configuration” of the organism can be pictured as a point on an n+1 dimensional “landscape” with valleys standing for unfit con-figurations of the organisms and peaks representing configurations which are effective at performing the task. Adaptation can be modeled as “hill climbing”. “Hill-climbing” or-ganisms explore their environment by systematically moving up-hill, thereby improving their fitness. Sometimes exploration involves whole populations whose trajectory across the landscape lasts for many generations; in other cases an individual climbs adaptive peaks via “learning”. Biological adaptation nearly always uses both evolution and learning. No-one, to date, has proposed a better way of generating systems with the ability to inhabit complex environments. It is for this reason that researchers trying to build autonomous ro-bots have founded much of their work on adaptive technologies which attempt to simulate some of the adaptive methods used by biological intelligence.

Figure 10.1. Hill climbing on an adaptive landscape

Before examining these technologies, however, it is useful to return to our initial dis-tinction between natural and artificial complexity. The description of adaptation given above is an abstract one. In this picture the ability of a population or an individual to adapt to a specific environment depends exclusively on the topological characteristics of the landscape[2]. It can be easily shown that adaptation is much easier on “smooth” land-scapes – where all paths lead to a single “fitness peak” - than on “rugged” ones – where it is difficult for organisms to locate truly optimal configurations. It follows that unless artifi-cial complex systems produce adaptive landscapes which are topologically different from those produced by natural complex systems – and we have no reason to suspect they do - the same processes of adaptation should work on both. This implies we can apply similar adaptive technologies in both domains.

10.3 Adaptive Technologies

Over the last forty years scientists and technologists have developed a number of different techniques which model or simulate the way biological organisms use learning or evolu-tion to adapt to their environments.

In what follows I will concentrate on four specific areas of development, namely:

? Artificial Neural Networks

? Evolutionary Computing

? The Ecological/Embodied approach to artificial cognition

? Collective or Swarm computing

As we will see at the end of this section all of these areas share a common epistemo-logical approach. All are based on the attempt to understand the dynamics of complex natural systems using what I will call bottom-up simulation. Later on in this essay I will argue that we can use the same approach when trying to penetrate the workings of artificial complex systems.

10.3.1 Artificial Neural Networks

Artificial Neural Networks (ANNs) were originally proposed by McCullough and Pitts in the early 1940s [3], and further developed by Rosenblatt with his Perceptron model in the late 1950s [4],.ANN models represent an attempt to incorporate knowledge about the workings of biological neurons and synapses in (extremely simplified) models of animal perceptual systems. ANNs perform computations using a network of processing units, conventionally known as “neurons”. Neurons are interconnected via “synapses” (see figure 10.2). Each “neuron” is characterized by a state; each synapse by a “weight”. The state of a neuron at time t is a (usually non-linear) function of the states of the neurons to which it is connected at time t-1 and the weights of its connections to these neurons.

In most ANNs a subset of neurons are defined to be “input neurons”; a second subset are “output neurons”. When the network performs a computation the experimenter pro-vides the system with an input by defining the states of the input neurons. The network then computes the values of all neurons to which these neurons are connected. This process is repeated recursively until it is possible to determine the state of the output neurons. This is the result of the computation.

Different classes of network are characterized by different patterns of connectivity. So-called feed-forward networks are based on a layered architecture where the input neurons represent the top layer of the network and output neurons the bottom layer. All connections

flow from the top of the network towards the bottom. In so-called Recursive Networks on the other hand [5] there are no restrictions on the direction of connections. Both feed-for-ward and recursive networks can be fully or sparsely connected.

Figure 10.2. A simple feed-forward ANN – the Perception

A key characteristic of ANNs is their ability to “learn” – a form of single-generation adaptation. Given a set of predefined examples (e.g. a set of vectors representing the goods purchased by individual customers in a store, sets of diagnostic data from a machine) certain kinds of ANN can “learn” to classify input vectors into categories defined by the experimenter (e.g. to assign the customer to a market segment; to categorize a malfunction of the machine). This is called “supervised learning”. Other classes of ANN perform “unsupervised learning” developing their own spontaneous categorizations.

ANNs have a number of desirable properties. The ability to learn from examples re-moves the need to write programs for problems which do not lend themselves to easy for-malization. The number of neurons required to implement a specific computation can often be surprisingly small. Many systems (particularly feed-forward networks) can perform their computations extremely fast. This does not mean however that ANNs are necessarily a valid model of biological cognition, still less that ANNs on their own can represent a complete solution to our problems in interacting with complex artificial systems

ANNs are known to suffer from a number of weaknesses. In particular, while research-ers have had great success in teaching ANNs to solve a number of relatively well-defined problems (e.g. recognition of phonemes in continuous speech [6], generation of continuous speech from text [7] generation of the past tense of English verbs [8]) it has not been pos-sible to develop ANNs with the ability to perform more complex cognitive tasks (e.g. automatic dictation). There are many reasons for this. Neural network architectures are designed by hand. While for simple cognitive tasks this can be relatively simple no one has yet developed an analytical method capable of discovering the optimal architecture for complex tasks. It is also known that at least for feed-forward networks the training problem is NP-complete[9]This implies that as the size of the network increases the time necessary to train it increases exponentially. Last but not least ANNs are dependent on the training input provided by the experimenter. This last characteristic contrasts with the ability of

Input Neurons Output Neurons

relatively simple living organisms to autonomously explore their environment and to “train” using the information they find there.

10.3.2 Evolutionary computing

An alternative “adaptive” technology, developed over the last thirty years, is so-called “evolutionary computing” ( a cover-all term encompassing Genetic Algorithms [10], Ge-netic Programming [11] and a number of other techniques). While ANNs model learning by individual organisms during a single lifetime, evolutionary computing borrows its basic principles from the much slower processes of Darwinian evolution.

In evolutionary computing, solutions to problems (or machines capable of generating solutions) are represented as the genomes of “organisms” belonging to a “population”. Genomes and organisms can be modeled in many different ways. In some instances the genome is a string of bits representing the weights of connections in an ANN; in other ANN-based models weights are represented as vectors of real numbers; in certain algo-rithms the genome is a table specifying a Finite State Machines; in so-called Genetic Pro-gramming it is a computer programs written in a high-level symbolic language.

Despite these differences the various techniques used by practitioners of evolutionary computing depend on what is, in essence, a single algorithm. All evolutionary techniques are based on three key principles: selection, replication and variation.

At the beginning of a computation based on evolutionary computing the system gener-ates a population of randomly generated genomes, specifying “digital organisms”. Every individual is evaluated for specific abilities specified by a “fitness formula”. Using the re-sults of this evaluation a sub-set of organisms is selected for replication. Selected organ-isms produce new copies of their genomes (“offspring”). Replication may be either sexual (two – occasionally more – parents contribute to the genome of the offspring) or asexual (offspring are clones of the parent). When copying genomes from “parents” to offspring the replication process introduces random “mutations” and “cross-overs” thereby intro-ducing variation within the population. Offspring are, in other words, similar but not iden-tical to their parents. The process of selection, replication and variation is repeated for many generations. In each generation it is the “fittest organisms” which produce the most offspring. In this way the mean fitness of the population increases steadily, exactly as the fitness of biological organisms is supposed to increase during Darwinian evolution.

Like Artificial Neural Networks evolutionary computing algorithms have been success-fully applied to a broad range of different domains. These range from the optimization of throughput on a manufacturing line [12] to the generation of school time tables [13] and the automated creation of electronic circuits [14]. Unfortunately evolutionary computing also shares many of the weaknesses of ANNs. Like ANNs it has proved hard to “scale up” evolutionary algorithms to tackle “hard problems”. Just like ANNs the time necessary to “evolve” a system increases exponentially with the complexity of the function the system is supposed to perform. And evolutionary algorithms, like ANNs, depend critically on in-put from the designer. It is the designer who defines the kind of organism and the genomic coding technique to be used for any specific problem. And again it is the designer who chooses the problems – unlike living systems – who to some extent are able to choose their own.

10.3.3 Ecological and “embodied” approaches

The techniques described above belong to what can be called the “information processing paradigm”. In this paradigm cognition is modeled as a mathematical function mapping an

“input space” to an “output space”. In the opinion of many modern psychologists this ap-proach excludes key features of biological cognition. Biological organisms, they point out, are physical organisms inhabiting environmental niches with specific physical properties. This has a number of implications for the workings of their cognitive systems. In particu-lar:

? Biological organisms are not abstract mathematical problem solvers. An animal, for instance, does not need to classify every possible image which might appear on its retina but only those images which are in some way of ecological importance (images of food, predators, mates, obstacles etc.).

? Biological organisms do not need to compute a complete, “objective” representation of their environment. Very often they exploit very simple signals as triggers for com-plex behavior. A fledgling, to cite just one example, will open its beak in response to any shadow passing at a certain angle above its line of sight [15]. Given the ecology of the fledgling’s nest there is no need to perform complex information processing to discover whether or not the shadow represents the fledgling’s parent.

? Physical organisms and environments contain large amounts of implicit information which the organism can exploit in performing its functions. To program a virtual ro-bot to shift a brick from the top of a pile and place it on the floor the programmer has to compute every movement of the brick. In a physical robot it is enough to ensure that the brick looses its balance. Gravity will do the rest.

? When organisms move around their environment the input they receive from their sensors changes continuously as a function of their position. Living organisms, unlike ANNs or evolutionary organisms, automatically select their own input every time they move.

? And above all organisms are autonomous. No living organism has to resolve an ab-stract problem set by an external experimenter. All it has to do is survive and repli-cate. If an animal or a population finds it hard to live in a particular environment it can often colonize another. At least to some extent living animals choose their own adaptive problems.

Recent years have seen the development of a new school of psychological thinking which has attempted to incorporate these insights into their models [16]. In order to achieve this goal researchers in “ecological” and “embodied intelligence” model not only the “brain” but the sensors feeding the brain its information, the motor systems it uses to move and the physical environment it inhabits. In some cases these models take the form of “animats” or software simulations. In others researchers have hand-designed physical robots or have used hybrid techniques to “evolve” computers on a computer and “validate” the resulting machines in a physical world [17]. The behaviors exhibited by these real or simulated machines have included walking (with six legs) [18], obstacle avoidance [19], imitation of the mating behavior of the male cricket [20] and simulating the behavior of laboratory rats in classic animal psychology experiments [21] There can be little doubt that the machines produced by researchers in “evolutionary robotics”, while still relatively primitive, have a far greater ability to deal with the complexity of the physical world than any ANN or evolutionary algorithm.

And yet even here there are problems in evolving models or robots with the ability to tackle the “hard” problems which animals successfully resolve in their everyday lives. Most researchers working in the field would agree that current robots have roughly the same level of intelligence as a simple insect. It seems, in other words, that even with rap-idly growing computational power there are severe, perhaps fundamental obstacles to the

creation of genuinely complex systems using the technologies described in this and the previous paragraphs.

10.3.4 Collective or “Swarm” computing

In the ANNs and evolutionary computing and often even in work by the ecologi-cal/embodied school the emphasis is on “individual computational units”. Complexity theorists have noted however that a knowledge of the workings of individual computation units is usually insufficient to explain the complex functionality of large complex systems. It is impossible, for example, to predict the properties of the genome from the sequences of individual genes or those of the brain from the characteristics of a neuron or those of an ant nest from the limited behavioral repertoire of individual ants. Complex behavior is usually, perhaps always, an “emergent” result of interactions among (relatively) simple compo-nents. There is strong evidence that major transitions in biological (and cultural) evolution (e.g. the emergence of the eukariotic cell, the evolution of insect societies, the emergence of new forms of human social organization following the Neolithic revolution) have often been due not to major change in individuals but to new forms of aggregation among them . It follows that if we wish to design systems exhibiting genuinely complex behavior the best strategy might be to build systems in which large populations of individual units interact, both on evolutionary and single generation timescales. This has been called swarm intelli-gence.

In a Swarm system complexity derives from the interaction among individual computa-tional units In Ray’s Tierra [22], for instance, a population of “computer programs” – written in an evolution-friendly Assembly Language – compete to absorb CPU cycles. Ray has observed how, in the course of this process individual programs acquire the ability to exploit other programs, as “subroutines”. The system spontaneously evolved parasitism and even hyperparasitism (parasites which live off other parasites). L. Steels has created “societies” of robots which spontaneously develop their own languages to describe objects with which they interact [23], M. Dorigo et al. have developed “ant colony techniques” in which populations of “artificial ants” spontaneously find solutions to complex optimization problems [24]; other researchers have developed “artificial immune systems” in which arti-ficial anti-bodied “compete” to attack “antigens” from the outside world [25]. Interestingly some of the techniques developed in the field of collective computing have been shown to have important practical applications in dealing with artificial complexity. In particular Ant Colony techniques have suggested new routing algorithms for telecommunications while artificial immune systems are applied to detect and respond to attacks on the security of computer networks.

10.3.5 Learning from the current generation of adaptive technologies

The current generation of adaptive technologies have limited (though real) practical appli-cations. They nonetheless provide important conceptual insights which we can apply in de-signing technologies to handle artificial complexity.

? The design of systems to handle complexity is an exploratory rather than an analyti-cal process. When trying to build such systems it is much often easier to “train” or “evolve” a system than to design it from scratch. System designers have only a very limited ability to hand-design systems to resolve arbitrary problems. Successful adaptive systems are themselves the product of adaptation.

? Successful adaptive systems are always adapted to the requirements and characteris-tics of a specific environment. This implies, on the one hand, that we should not ex-pect to be able to create general purpose technologies capable of dealing with every possible kind of artificial complexity, on the other that in many cases it may be pos-sible to exploit specific characteristics of individual environments so as to greatly simplify the computational tasks the technology is asked to perform.

? Successful adaptive systems display a high degree of autonomy. They “live” in their environment, make their own decisions how to move around the environment, select their own inputs and to a large extent choose their own problems.

? Successful adaptive systems are themselves complex. Swarm systems comprising large numbers of interacting components have emergent behaviors which cannot be predicted from the properties of these components. It can often be useful to design for emergence.

10.3.6 Bottom-up simulation

The adaptive technologies developed by workers in the fields of ANN, evolutionary com-puting, the ecological/embodied school and swarm computing are all based on attempts to simulate the working of complex systems found in nature. ANNs are a primitive model of the brain (or of perceptual systems within the brain); evolutionary computing is inspired by Darwinian evolution; the ecological/embodied school attempts to reproduce the way in which whole, embodied organisms interact with a physical environment; swarm computing simulates the complex interactions found in eco-systems, ant-nests and other animal socie-ties.

Viewed from a highly abstract point all the techniques share a similar bottom-up strat-egy which can be briefly summarized as follow:

? The system designer views the system to be simulated as a collection of interacting subsystems. Brains are viewed as a collection of neurons; evolutionary algorithms and swarm computing use populations of “organisms”; the ecological/embodied ap-proach use elements from both of these views.

? The simulation effort concentrates on reproducing the (relatively simple) behavior of individual sub-systems (neurons, organisms etc.) in their interactions with other sub-systems.

? Specific system behaviors are produced by general purpose “training algorithms” de-signed to modify the interactions among sub-systems (neural networks, swarm sys-tems) and/or to select subsystems with desirable characteristics.

? Complex behaviors are “shaped” one step at a time. Systems with the ability to generate a specific behavior are used as “building blocks” to build other more com-plex behaviors.

As will be seen later on in this essay these techniques are applicable to the simulation of artificial as well as natural complex systems.

10.4 Technologies to handle artificial complexity

10.4.1 Initial requirements for handling artificial complexity

From the point of view of human beings Complex Systems are black boxes. Humans do not know what is going on in the head of the tiger as it waits for them in the bushes; work-

ers have only limited insight into the workings of the equipment they operate; traders lack the knowledge to penetrate the mechanisms which determine the rise or fall of stock prices. When human beings attempt to predict, manage or control the behavior of a complex sys-tem the data they have to work with is mainly data about the behavior of the system. Al-though a pilot has no knowledge of micro-fractures within an engine or a strut she can sense if the plane is vibrating in an unusual way; a manager knows next to nothing about the psychology of individual consumers but can tell if they begin to shift their purchasing patterns. To interact effectively with a complex system it is necessary to analyze its be-havior and decide one’s own best reactions. To do this however we do not need data but information. From the point of view of survival it is not important to note the exact ge-ometry of a tiger’s spots; it is vital to distinguish a tiger from other, less dangerous ani-mals.

Cones and rods in the human retina provide a very detailed low-level description of patterns of light, shadow and color in the visual field. In order to be useful this data has to be categorized and organized into forms which can be used to guide behavior. In natural cognitive systems this is the role of perception. Perception transforms sensor- level de-scriptions of reality into information which the organism can use to further its goals. Com-plex artificial systems pose similar problems. Modern information systems contain ex-tremely high volumes of data. We can record changes in hundreds or thousands of stock or bond prices minute by minute or even second by second; stores, banks and credit card companies record details of every individual customer transaction; modern machinery can produce vast volumes of diagnostic data; telecommunications networks keep track of every individual call or, in theory, even of individual data packets. Yet this is low level data. To be useful it needs to be transformed into a form suitable for processing by the human cog-nitive apparatus. In short we need artificial perception.

10.4.2 Artificial perception

The role of perception in natural cognitive systems is to “map” the high-dimensional, noisy input from sensors onto simpler “low-dimensional” representations (for instance a two di-mensional image, a phoneme, a smell) which can be used as a basis for decision-making. The perceptual systems we need to deal with artificial complexity have to perform the same tasks identifying and signaling significant information (e.g. customer preferences, system malfunctions or overloads, trends in stock prices) present in large volumes of noisy, incomplete and largely irrelevant data.

When researchers seek to simulate animal perceptual systems they often use Artificial Neural Networks. Like biological perceptual systems the networks they use have large numbers of input neurons and a much smaller number of “outputs”. When the network is “trained” on a given data set it “learns” to “categorize” the data. Hand-written characters, for instance, can be classified according to the letters they represent. Frequency spectra from mass spectrographs can be used to identify specific chemical compounds. In some cases output neurons are organized into “maps” [26]. If one input activates a specific neu-ron on the map similar inputs will activate neurons which are close to this neuron; inputs which are dissimilar will activate neurons which are further away.

Artificial Neural Networks are an extremely effective and parsimonious way of per-forming artificial perception. There is, what is more, nothing which restricts their applica-tion to natural environments. From the very early days of Neural Network research they have been used not just to gain insight into biological cognitive systems but also for practi-cal engineering applications (e.g. in the field of Optical Character Recognition). There can be very little doubt that Artificial Neural Networks will play a key role in systems of Arti-

ficial Perception designed to facilitate our interactions with artificial environments. The first of these applications are, in fact already with us, in the form of so-called “data mining software”.

10.4.3 Data mining as primitive artificial perception

Data Mining [27] is a technique to extract commercially or scientifically useful informa-tion from data collected for other purposes. Using data mining a supermarket can analyze data collected at the till to detect associations between the purchases of certain products and thus to optimize the layout of the store. A disease management team in a hospital can analyze data about symptoms, treatments and outcomes to select the best treatment proto-col for a specific clinical condition. The till or the hospital records system provide raw data. The data mining system generates information. It can thus be seen as a form of artifi-cial perception.

Data Mining software uses an eclectic range of techniques including ANNs. In each case the analyst provides the system with a set of “training data” including a set of predic-tors and a target variable. The system then attempts to generates a model (e.g. a trained ANN) with the ability to predict the value of he target variable.

Let us suppose that a telecommunications operator wishes to predict the probability that a particular customer will “churn”, canceling her contract and choosing a different opera-tor. A typical operator will possess large volumes of data describing the characteristics (age, sex, profession, address) and behavior (number of calls, destinations of calls, weekly and daily distribution of calls, requests to customer care, technical problems, delays in payment) of past customers who have “churned” and of those who have remained loyal to the company. In order to predict the behavior of its current customers the company can use the data about the past behavior of customers to train an ANN to predict whether or not a customer will churn. Once this has been achieved the company can then feed the network with data about its current customers, read out the predicted probability of churn. This is the kind of information which it is possible to produce with current state-of-the-art artifi-cial perception.

10.4.4 Weaknesses in current data mining software/ further requirements for artificial

perception

Using the concepts developed by the ecological/embodied school it is easy to see the dif-ferences between the primitive artificial perception just described and genuine “natural perception”. Four of these are of particular importance.

? Natural perception is sensitive, above all, to change and movement. What matters most to the gazelle is not the bushes but the unexpected movement in the bushes which might signal the presence of a predator. The artificial perception provided by current data mining systems (and other forms of data analysis) is, on the contrary, es-sentially static. The software identifies unchanging relationships among predictor and target variables. No mechanism is provided for alerting the user to the unexpected. In many application areas this is a critical weakness. In the specific case of telecommu-nications, for example, the appearance of a new competitor, a new technology or a new pricing mechanism can herald extremely rapid changes in market dynamics.

? Natural perception adapts to changes in the environment. As an organism develops it learns to recognize the features of the environment which are most important to its survival. Changes in the environment are matched by changes in the behavior of the perceptual system. Currrent data mining systems, on the other hand, use a single set

of static training data. The ANN is trained once. Tacitly it is assumed that future be-havior will be determined by the same factors that have determined behavior in the past.

? In natural perception the organism actively selects the information which is most relevant to its current needs – choosing from a much larger volume of information generated by the perceptual system. It is the presence of this apparently superfluous information – and the ability to select information that allows biological organisms to respond to novelty in the environment. In contrast data mining software produces predictive models for single variables, or very small sets of variables selected by an analyst. It is the analyst who chooses the questions. The software gives the answers.

This means that if the analyst has not chosen the right questions the software is of very little use.

All of which suggests that more effective systems for artificial perception should:

? detect potentially significant changes in the environment;

? adapt to changes in the user needs and/or in the underlying dynamic governing the behavior of the environment;

? present the user with a broad range of different information, allowing her to select the information most relevant to her current needs.

No system currently present on the market comes close to meeting these requirements. We can nonetheless imagine ways in which imaginative use of adaptive technologies – and of the lessons of the ecological/embodied school - could help us to produce systems which are much closer to meeting the requirements than those currently in use.

10.4.5 Adaptive Artificial Perception

Current Data Mining uses a static training set to produce a single model to represent the relationship between predictor variables and a single target variable. From what we know of adaptive technology it is not difficult to imagine changes in this paradigm which might enable us to implement new forms of Adaptive Artificial Perception

? In data mining systems the “training set” is a static data file. In future systems on the other hand the data available to the system could be updated on a continual basis. All analytical activities would thus be based on up to date information.

? In the current generation of systems training is “batch” process conducted at periodi-cal intervals. In future systems data would be analyzed on a continuous basis – en-suring that information concerning new trends in the environment becomes available as rapidly as possible.

? Current systems usually take little or no account of the dates and times at which data was recorded. In a new generation of system all data would be characterized by the time at which it was recorded. Systems would be provided with automatic mecha-nisms to detect short or longer term changes in critical variables. Predicting change would be treated as a significant goal of the system

? Current systems capture regularities in the data set in a single “model”. Future sys-tems are likely to borrow from the evolutionary paradigm allowing the “evolution” of multiple competing models. Data Mining has already developed techniques for auto-matically measuring the reliability of models. These techniques could be used to measure the “fitness” of the models produced by competing agents. As the environ-ment changes older models will be replaced newer, more reliable ones. Adaptation to change will be automatic

? The models produced by present data mining software predict the value of a single target variable. While a new generation of systems would obviously retain this ability the goal will be to present the user with a broad range of different information, de-scribing as many aspects as possible of the environment with which she is interact-ing. The user will be able to explore the environment, selecting the information of greatest interest to her. The fitness formula used to evaluate competing models will be designed in such a way as to reward the models producing information of the greatest interest to the user. In this way the user will become “part of the loop” guid-ing the development of the system in the same way as researchers in artificial robot-ics “shape” the behavior of their creatures in successive rounds of evolution.

? The models created by current data mining software are monolithic entities: a single ANN is responsible for all computational work performed by the model. Such models may include tens or even hundreds of predictor variables many of which are later shown to have no influence on the value of the target. As we have seen the time nec-essary to train an ANN grows exponentially with the size of the network. It is not therefore surprising that the training of this kind of model is a computation-intense task which often requires many hours. Future systems, I would suggest, are likely to take an alternative approach incorporating insights from Swarm Computing. In future systems, based on multiple competing agents, individual agents may be relatively simple with the ability to use only a limited range of information from the environ-ment (a limited number of predictor variables). The time required to train such agents would be short. At the same time however each agent would view the other agents as part of its environment. In this model every agent would use the categorizations and predictions produced by other agents as input for its own categorizations and predic-tions, just as the programs in Ray’s Tierra use other programs as sub-routines. Com-plexity (and the ability to react to complexity) would arise not from the sophistication of individual agents but from their interactions. As the system evolved it would de-velop the ability to perceive information at ever higher levels of abstraction.

With the exception of the proposed use of Swarm Computing – which might require a significant effort in basic research – the ideas just described require no fundamental breakthroughs in theory or technology. In the majority of cases all that is required is imaginative use of technologies and algorithms which already exist. It is not hard to predict that in coming years adaptive technologies will play a key role in new, flexible systems of artificial perception designed to help humans in their interaction with complex artificial systems.

10.4.6 Simulating the behavior of artificial complex systems

Artificial perception can provide human operators with the ability to detect, classify and react to significant information hidden in very large volumes of data. It would however be a mistake to believe that this is the only contribution which adaptive technology can make to our interaction with complex artificial systems.

An important (if somewhat over-valued) characteristic of human cognition is the ability to plan. Human beings have the ability to “mentally simulate” the consequences of their actions. By examining alternative scenarios they can choose tactics and strategies likely to lead to favorable outcomes. The human brain is well equipped to plan actions in the natural environments where human beings and their ancestors evolved. Human beings are, for in-stance, extremely skilled in predicting the movements of physical objects, the behavior of animals or that of other human beings. But when dealing with artificial complexity this

ability can break down. A city planner may build a new road hoping to reduce traffic bur-den and find that actual traffic increases; the operator of a complex technological system may make a tiny adjustment to a single parameter and find the system spiraling out of con-trol. To effectively manage artificial complex systems we need technological tools to help us predict their behavior. To achieve this goal we have to simulate the working of the sys-tems we are attempting to manage. This brings us back to the underlying research strategy of adaptive technology and to its use of bottom-up simulation.

Artificial complex systems, like their natural counterparts, are made up of large num-bers of interacting sub-systems. Financial markets are constituted by large numbers of buyers and sellers; patterns of traffic in a city or on a highway are determined by the inter-actions between large numbers of vehicles; the spread of human and animal disease is the result of interactions between hosts and disease vectors and the ways these move across geographical space. In all of these cases it is possible to build bottom-up simulations ap-plying the same basic strategies developed by researchers in adaptive technology. Today bottom up simulation is already used in the planning of urban transport systems, the evaluation of exotic financial instruments and environmental impact assessments. As the cost of computing power continues to fall the range of domains where bottom up simula-tion can provide an effective management tool can be expected to increase steadily. In short in the future adaptive technology will provide us not only with technological tools to perceive the state and dynamics of complex artificial systems but with new strategies to simulate and predict their behavior.

10.5 Integrating Adaptive and Virtual Reality technologies

In the previous sections I have tried to show that adaptive technologies have significant potential for assisting us in our interactions with complex artificial systems. At this point however the reader has the right to ask what has all this has to do with Virtual Reality – the central theme of this book.

It is true that complex artificial systems are all, in a weak sense, “virtual” – a share price or a network malfunction does not have the same perceptual impact as a tiger or a fall from a great height. But Virtual Reality – in the normal usage of the word – means more than this. When we talk about Virtual Reality what we are normally talking about are technolo-gies which allow the user to explore a Virtual World – which she can sense with her visual, auditory and perhaps her tactile and olfactory systems. This sort of technology appears, at least at first sight, to have very little relationship to “artificial simulation” or “bottom-up simulation”. To perceive the link it is necessary to take a closer look at the tools human beings will need if they are to interact effectively with complex artificial systems.

The ultimate goal of a human being interacting with an artificial environment is to achieve the same kind of intuition and fluidity of action which humans and animals display in their interactions with natural systems. This is what VRT achieves when a pilot flies a virtual jet fighter or a surgeon performs a virtual operation or a child explores a dinosaur-infested island. The problem with the kind of virtual environment I have been describing here is that it has no physical counterpart – even in the imagination. Taken in its raw form there is nothing for Virtual Reality to visualize and nowhere for the user to explore. The role of adaptive technology, I have argued, is to provide us with tools to map this kind of abstract, high dimensional world into the kind low dimensional representation which the human cognitive system is capable of processing and to simulate the way in which the world might react to particular actions by the user. But here again the idea of a low-dimen-sional representation is an abstract one. A key element is missing. Adaptive technology

does not allow us to “see” the information it provides; it lacks a “human interface”. Virtual Reality, I would suggest can help to fill this gap.

First of all a proviso. It is obvious that not all aspects of Virtual Reality Technology are relevant to the kinds of system we have been discussing in this paper. I can, for example, see few applications for “immersive” virtual reality, stereoscopic vision and 3D sound and even fewer for data gloves and artificial olfaction. It is likely that in the design of systems to help us interact with complex artificial systems the most important contribution of Vir-tual Reality will not be hard technologies such as advanced 3D rendering but soft tech-niques such as ergonomics and user-interface design.

To conclude, therefore, I would like to sketch out an example of how Virtual Reality and Adaptive Technologies could be merged in an integrated system enabling human be-ings to interact with some kind of large artificial system. My example system – a “Market Browser” - will again be tailored to the needs of a large Telecommunications company – or more specifically to the needs of a marketing manager working for the company.

10.6 A “Market Browser” using Adaptive Virtual Reality

In our hypothetical system the goal of the marketing manager is to increase her company’s profits by taking the initiative in the market and by reacting rapidly and effectively to the initiatives of competitor companies. The “Market Browser” I will describe uses Adaptive Artificial Perception to generate the information and she needs in order to interact effec-tively with the market; a market simulator allows her to generate alternative scenarios for market development and to simulate the effects of different actions (changes in pricing plan, special promotions, advertising) on the behavior of consumers and competitors; the human interface is based on VRT.

As in most Virtual Reality Systems the user interface to the Browser is organized around a spatial metaphor. The market is portrayed to the user as a “market space” which she can visualize via maps. A map is a two or three dimensional representation of some particular aspect of the space. A Market Browser might include Geographical Maps showing physical locations where the company sells; a Traffic Pattern Map clustering customers in terms of different patterns of phone use; a Demographic Map providing an alternative clustering in terms of age, sex, area of residence, income and other demo-graphic variables. The Artificial Perceptual System will use the maps to present the user with a broad range of information, drawing her attention to factors which appear to have special predictive power for revenue, churn or other key indicators. The geographical map, for instance can be colored or textured to show the contribution of individual areas to com-pany revenues; special symbols can be used to indicate areas whose contribution to reve-nue is rising faster or slower than the general trend for the company; contours can be used to indicate quality of technical service in specific area, additional symbols can be used to show the presence of local points of sale or to alert the user to areas with unusual technical problems or with an usually high number of complaints to technical service. A large sym-bol can show the company’s current “center of gravity” with a trace indicating the way in which this is changing over time...

As in more conventional Virtual Reality systems the system is navigable. At all times the user is characterized by her position in “market space”. Navigational controls allow her to move round the space. At all times she can control her view of the space.Menu com-mands allow the user to decide which maps to display (multiple maps will be displayed on multiple windows or perhaps on multiple screens);zoom commands control the amount of detail she wishes to see.

A key aspect of the system is its ability to draw the user’s attention to new information of which she was not aware. New information discovered by the system is displayed either on the maps the user is currently viewing or (where the information is not relevant to the current map) on “tickers” and other instruments located on peripheral areas of the screen A novel aspect of the system is the sharing of control over the display of information between the Artificial Perceptual System and the user. When the system is used for the first time the Artificial Perception will autonomously decide the information that needs to be displayed. The user will then be able to eliminate information which she does not believe to be im-portant and explicitly request the display of information she needs. The Artificial Percep-tual System will use information about user choices to evaluate the “fitness” of agents (see 10.4.5) thereby “evolving” a population of agents suited to the user’s specific interests and requirements.

As described earlier the Artificial Perceptual System will work side by side with a “Market Simulator” – based on bottom-up simulation techniques- which will enable the user to generate new scenarios for market development and to explore the consequences of her company’s possible actions. The simulator will generate a dynamic market space char-acterized by the same variables which the Artificial Perceptual System uses to describe the actual market. The “Market Browser” will thus allow the user to use the same human inter-face to explore “real” and simulated reality.

10.7 Conclusions – future prospects

Adaptive technologies, as we have seen, can provide powerful tools to assist human beings in interacting with complex artificial systems as well as in thinking about and planning these interactions. The development of such tools is already feasible and requires no major theoretical or technological breakthroughs. In reality the main difficulty in the design of such tools is not technological but ergonomic. Classic adaptive technologies have no hu-man interface. VRT, I have argued, has the potential to supply such interfaces. In Section 10.6 I have described a hypothetical system based on this kind of integration. I leave it to the reader – and the next twenty years of technological development – to decide if the pic-ture I have painted is a realistic one.

10.8 References

[1] S. Wright, Evolution in Mendelian Populations, Genetics, (1931), 16-97.

[2] S. Kauffman, The Origins of Order, O.U.P., 1992.

[3] W.S. McCullough & W. Pitts, A logical calculus of the ideas immanent in nervous activity, Bulletin of

Mathematical Biophysics, 5 (1943) 115-133.

[4] F. Rosenblatt, The perceptron: a probabilistic model for information storage and organization in the

brain, Psychological Review, 65 (1958) 386-408.

[5] D.J. Amit, Modeling Brain Function, C.U.P., 1989

[6] T. Kohonen, The ‘neural’ phonetic Typewriter, Computer21 (1988) 11-22.

[7] T.J. Sejnowski & C.R.Rosenberg, NETtalk, a parallel network that learns to read aloud. John Hopkins

University Electrical and Computer Science Technical Report, JHU/EECS-86/01, 1986.

[8] D.E. Rumelhart & J.L. McLelland, On Learning the Past tense of English Verbs, in Parallel

Distributed Processing Vol 2., J.L. McClelland, D.E. Rumelhart and the PDP Research Group (ed.), MIT Press, 1986.

[9] J.S. Judd, Neural Network Design and the Complexity of Learning, MIT Press, 1990.

[10] M. Mitchell, An Introduction to Genetic Algorithms, MIT Press, 1998.

[11] J.R. Koza, Genetic Programming, on the programming of computers by means of natural selection,

MIT Press, 1992.

[12] R. Dupas, G. Cavory & G. Goncalves, Real world applications: optimizing the throughput of a

manufacturing production line using a Genetic Algorithm, GECCO 99, Proceedings, Morgan Kauffman, 1999

[13] C. Fernandes, J.P. Caldeira, F. Melicio & A. Rosa, Evolutionary Algorithm for School Timetabling,

GECCO-99, op. cit.

[14] A. Thompson, Silicon Evolution in J.R. Koza et al., eds., Genetic Programming 1996: Proceedings of

the First Annual Conference (GP96), 640-656, Springer Verlag, 1996

[15] N. Tindbergen, The Study of Instinct, O.U.P., 1951

[16] A. Clark, Being There, Putting brain, body and world together, MIT Press, 1997

[17] O. Miglino, H.H. Lund .& S. Nolfi S., Evolving Mobile Robots in Simulated and Real Environments,

Artificial Life, 2(4) (1996) 417-434

[18] R.D. Beer, Intelligence as Adaptive Behavior, Academic Press, 1990

[19] R. Brooks, A robust layered control system for a mobile robot, IEEE Journal of Robotics and

Automation, (1986), 14-23 reprinted in R. Brooks, Cambrian Intelligence, MIT Press, 1999

[20] H.H. Lund,., B. Webb & J. Hallam, Robot Attracted to the Cricket Species Gryllus bimaculatus. in P.

Husbands and I. Harvey, eds, Proceedings of Fourth European Conference on Artificial Life, pages 246--255, Cambridge, MA,. MIT Press, Bradford Books, 1997

[21] R. Walker & O. Miglino, Replicating Experiments in "Detour Behavior" with Artificially Evolved

Robots: An A-Life Approach to Comparative Psychology, Lecture Notes in Artificial Intelligence 1674, Springer , 1999

[22] T.S. Ray, An Approach to the Synthesis of Life in Artificial Life II, 371-408, MIT Press, 1992

[23] L. Steels, Synthesizing the origins of language and meaning using co-evolution, self-organization and

level formation in Hurford J. (ed), The Evolution of Human Language, Edinburgh University Press, 1997

[24] M. Dorigo, V. Maniezzo & A. Colorni, The Ant System, Optimization by a colony of cooperating

agents. IEEE Transactions on Systems, Man and Cybernetics Part B, 26,1, 29-41, 1996

[25] S. A. Hofmeyr & S. Forrester, Immunity by design, an artificial immune system, GECCO 99, pp

1289-1296, op. cit., 1999

[26] T. Kohonen, Associative Memory, Springer, 1976

[27] A. Berson, S. Smith, K. Thearling, Building Data Mining Applications for CRM, McGraw-Hill, 2000

翻译技巧和经验第17期Virtual reality 该如何翻译

近一个时期来，Virtual Reality (VR)一词在报刊等新闻媒体上频频出现，十分活跃，从而也引出了对VR 该怎么好的问题。据《光明日报》载，VR 的译名计有：“虚拟实在”（1996 年,10 月28 日）“临境”、“虚实”、“电象”、“虚拟境象”以及“灵境”（1997 年1 月16 日）；此外，还有最常见的“虚拟现实”（如《文汇报》1997 年3 月12 日“小辞典”）。 对VR 究竟怎么理解和怎么翻译，笔者不敢妄下结论。不过，从最近一期外刊上读到的一篇文章（载于 Jane's DEFENCE'97，作者为 Ian Strachan，Editor，Jane's Simulationand Training Systems）在这两方面都可给人以不小的启发。为帮助说明问题，现将此文中有关的部分摘录于下。文章标题和引语是： VIRTUALLY REAL Virtual reality is rapidly becoming the training tool o f the 21stCentury - but what exactly is it? 将此翻译过来大致是： 可说是真的Virtual reality 正在迅速成为二十一世纪的训练器具--但它确切的含义是什么呢？ 文中在讲到 virtual 一词的定义时是这样说的：One dictionary definition of "virtual" is"something which is unreal but can be considered as being real for some purposes." 文章接着对 VR 作了几种实质性的释义： Cyber-something? To some, VR is putting on a Cyber-helmet (whatever that is) and involvestotal immersion (whatever that means) in cyberspace (where ver that is). Collimation? To some, VR is a view through a collimated display system. Tactile sensors? To some, VR implies tactile simulation as well a visual syst em. 对以上这些解释或释义，Strachan 认为它们都是对的：Which of these interpretations of VR is right? Ibelieve that they all are.（对VR 的这些诠释中哪个是对的？我相信它们都是对的。）为什么都对？Strachan 说那是因为 English is a living language and insistence on any preciseinterpretation of VR is narrow and not in accordance with what has already become commo nusage.（英语是一门活的语言，而坚持对 VR 作任何精确的诠释是狭隘观念，也不符合约定俗成。） 通过以上介绍，我们似乎能够得出这样的结论：各位专家学者所给的译名都是对的。因为，汉语也是一门活的语言，坚持对 Virtual Reality 作任何精确的翻译是狭隘观念。但是，另一方面，则仍有必要从以上译名中挑选出一两个最好的来，或者也可以另起炉灶，如果还有更好的话。那么，我们认为“虚拟现实”和“虚拟实在”与原文更加“形似”。其它几个译名虽各有千秋，但总嫌“神似”有余，而“形似”不足。词和词语的翻译同句子和篇章一样应力求“神形兼具”。在这种情况下，找到一个最理想的译名实属必要。 借鉴以上众多译名，并综合Strachan 所做的分析和解释，经过反复比较和推敲，我们认为 VirtualReality 视情可译为“拟真技术”、“拟真”或“虚拟真实”。这几个译名与所介绍的那些在含义上大致相同而表述略有不同，或许是最为得体的。理由如下： 第一，这么译十分符合英文词语的本义。Virtual 在这里意为 in effect, though not in fact; not such infact but capable of being consider

(完整版)初级语法总结(标准日本语初级上下册)

1.名［场所］名［物/人］力*笳◎去歹/「仮歹 名［物/人］总名［场所］笳◎去歹/「仮歹 意思：“ ~有~ ”“~在~”，此语法句型重点在于助词“心‘ 例：部屋忙机力?笳◎去歹。 机总部屋 2.疑问词+哲+动(否定) 意思：表示全面否定 例：教室忙疋沁o 3?“壁①上”意思是墙壁上方的天棚，“壁才是指墙壁上 例：壁 4. ( 1)名［时间］动 表示动作发生的时间时，要在具体的时间词语后面加上助词“V”，这个一个时间点 例：森总心比7時V 起吉去To 注意：只有在包含具体数字的时间时后续助词“V”，比如“ 3月14 日V, 2008年V”；星期后面可加V,比如“月曜日V” ,也可以不加V;但是“今年、今、昨日、明日、来年、去年”等词后面不能加V。 此外：表示一定时间内进行若干次动作时，使用句型“名［时间］V 名［次数］+动” 例：李1週間V2回7°-^^行吉去T。 (2)名［时间段］动：说明动作、状态的持续时间，不能加“ V” 例：李毎日7時間働^^L^o (PS: “V”的更多用法总结请看初级上第15课) ( 3)名V 【用途】【基准】 表示用途时，前接说明用途的名词，后面一般是使"去T等动词 表示基准时，前名词是基准，后面一般是表示评价的形容词。 例：--乙①写真总何V使"去T力、。「用途」 --申請V使"去T。「用途」 乙①本总大人V易L^^ToL力'L、子供V总難L^^To 「基准」 X —X—力*近乙買⑴物V便利^To 「基准」 (4)动(基本形) OV 【用途】【基准】：使用与上述( 3)一样 例：乙O写真求一卜总申請T^OV 使"去T。 ^OV>^3>^ 買“物T^OV 便利^To (5)小句1 (简体形)OV，小句2:名/形動+肚+OV 表示在“小句1 ”的情况下发生“小句2”的情况不符合常识常理，翻译为“尽管…还是…，虽

北师大版高中英语教案--unit4lesson4

Unit4 lesson4 Virtual Tourism 夏英明 Teaching aims: 1, Ss learn to use reading skills. 2, Ss learn to use “do” and “make”. 3, Ss learn to listen and find out main facts in an article. Key and difficult points: 1, New words and phrases: offer, suggest, guide, volcano, suburb, harbour, central, settle, be located on, also, as well as, too,. 2, Help Ss develop the awareness of culture. 3, Help Ss use the reading strategies. Teaching methods: 1, Ss-centered method. 2, Read-to-learn method. 3, Listen-to-learn method. Teaching aids: 1, Multimedia to show photos, table. 2, Blackboard to rewrite the key words and phrases. Teaching steps: 1, Greet the C as usual. 2, Warm-up Have Ss discuss something about China in terms of population, capital, famous sights, climate and so on. Location Population Features History Cli Show the C some photos about Auckland and help Ss to learn some new words: guide, volcano, suburb, harbour, locate, central, Maori, settle. 3, Lead-in Show the C 5 sentences, have Ss read the article and judge them true or false. Call Ss to answer one by one, if he/she says it false, T will ask the S to give the

firmware升级流程

Firmware升级流程 一．概述 AN5116-02系统中，EC2以及各种ONU的芯片firmware版本均由TEKNOCUS公司提供，且更新比较频繁。目前最新中试的版本为R140，而工程是普遍使用的为R105版本，因此，工程上的AN5116-02系统迫切需要升级以满足新的功能需求和维护要求。一般情况下，我们可通过图形网管命令直接对局/远端的芯片firmware升级。参考拓扑图如下： 二．升级操作方法 升级的对象主要有EC2的firmware和ONU的firmware两项，每一项目又分为三个子项：boot程序，app程序和personality程序（boot可能不是必需的）。升级时，我们不关心三个子项的名称，仅关心这些程序是属于EC2的，还是ONU的，如果我们要升级的子项是EC2上的，则应在图形网管上选择升级“EC2 firmware”;反之，如果是ONU 上的子项升级，则应在图形网管上选择升级“ONU firmware”。另外，boot，app,personality这三个子项一般是严格按照顺序来升级的，即先升级boot，然后是app，最后是personality。升级时，待升级的对象必须在位，如果是ONU，它应该是已经授权了的状态。 下面介绍整个升级流程。 假设目前网管具有以下条件： 网管服务器ip地址为10.26.1.2/16 欲使用的FTP服务器的用户名为test , 密码为test 文件目录为D:\ ONU的授权号为5，其对应在线的EC2槽位号为2。 首先准备好FTP服务器和要升级的所有文件： 1.打开FTP server 软件，并设置用户名和密码均为test，文件存放目录为D:\ 。 2.准备好升级文件。根据归档提供的*.tkf文件编辑好ONU的personality文件 （编辑方法见文档后的附录），并任意改好文件名，如ec2_https://www.wendangku.net/doc/003482570.html,f , ec2_https://www.wendangku.net/doc/003482570.html,f , ec2_https://www.wendangku.net/doc/003482570.html,f ，onu_https://www.wendangku.net/doc/003482570.html,f , onu_https://www.wendangku.net/doc/003482570.html,f , onu_https://www.wendangku.net/doc/003482570.html,f，

标准日本语初级语法总结(上)

第一部分——名词 名词：名词是词性的一种，也是实词的一种，是指待人、物、事、时、地、情感、概念等实体或抽象事物的词。在日文中的充当句子成分时可以做主语，宾语，目的语等。与数量词、代词等构成体言。在单词方面完全由汉字组成的词汇如：説明せつめい 等很多是音读即类似汉语发音，当然也有训读的时候（海うみ ），甚至一个词有音读和训读两种念法（紅葉こうよう 和紅葉もみじ ）。由汉字和假名组成的词语大多训读念法例：お知らせ 名词无变形但有时态的差别： 名词 简 体 敬 体 现在式 过去式 现在式 过去式 肯定 名词+だ 例：学生だ 名词+だった 学生だった 名词+です 学生です 名词+でした 学生でした 否定 名词+ではない 学生ではない 名词+ではなかった 学生ではなか った 名词+ではありません 学生ではありません 名词+ではありませんでした 学生ではありませんでした 另外名词的中顿形：名词+で、～～～。 名词的推量形，表示推测或向对方确认：でしょう。 初级上名词相关时态、句型： 敬体： １．现在肯定：N です わたしは 王です。「３」２．现在否定：N ではありません わたしは 日本人では ありません。「３」 ３．过去肯定：Ｎでした 前の 会社は 日系企業でした。「１１」 ４．过去否定：Ｎでは ありませんでした 先週，大阪は いい 天気では ありませんでした。「１１」 简体： １．现在肯定：Ｎだ 今日は 日曜日だ。「１８」 ２．现在否定：Ｎではない クリスマスは 祝日では ない。「１８」 ３．过去肯定：Ｎだった 恋人からの 誕生日の プレゼンは ネクタイだった。「１８」 ４．过去否定：Ｎではなかった 昨日は 休みでは なかった。「１８」

北师大英语高一必修二 第四单元 lesson4 说课教案

Lesson 4 Virtual Tourism教学设计（陶佳）（2011一等奖） 课题：M 2-Unit 4-Lesson 4 Virtual Tourism（第一课时） 指导思想与理论依据： 高中英语课程的总目标是培养学生的综合语言运用能力，其中包括语言技能、语言知识、情感态度、学习策略和文化意识。在此基础上，特别注重提高学生用英语提取信息、处理信息、分析和解决问题的能力，用英语进行思维和表达的能力；形成跨文化交际的意识和基本的跨文化交际能力。本节课在阅读过程中，通过表格的设计，培养学生用英语提取信息、处理信息的能力，在之后的小组口语活动中，用所学到的相关知识进行英语的口头作文，这就培养了学生用英语进行思维和表达的能力。 同时，新课标倡导面向全体学生，以“主动参与，乐于探究，交流与合作”为主要特征的学习方式。在本节课中，从始至终贯彻“以学生为主体”的原则，首先让学生在阅读过程中独立完成各项任务，使学生能够完全自己参与，而不是依赖他人来获取有效信息，从而提高了学生的阅读技能，发展了学习策略；其次，在小组活动中，通过对学生提出详细的具体要求，使所有的学生都能够积极的参与进去，与小组的其他成员进行交流与合作，从而培养了学生团队合作的精神，形成了积极的情感态度。 此外，新课标要求教师“要善于结合教学实际的需要，灵活的和有创造性的使用教材，对教材的内容、编排顺序和教学方法等方面进行适当的取舍或调整”。本节课在结合教学需要和学生实际水平的前提下，将第四课与Communication Workshop 进行了整合与删改，正是体现了新课标的要求。 教学内容分析： 本单元为北师大版第二模块的第一个单元，语言材料难度不是很大。本单元的话题是“Cyberspace”，由“Tomorrow's World”“Websites”“Virtual Reality”“Virtual Tourism”“Communication Workshop”(Writing an Internet page & planning a scientific experiment)和“Culture Corner”(New Zealand Fact File)六部分组成。其中，第一、三课为“Language Focus”，第二、四课为“Skill Focus”，第五部分侧重说和写的输出，第六部分的文化广角侧重文化知识的输入。在本单元的教学安排上，教师根据学生的实际水平进行了适当的合并与删减，本单元的第四课是介绍一个地方的阅读课，而本单元的Communication Workshop中正有让学生完成一篇介绍一个地方的写作内容，教师将两部分进行了合并，同时删除了Communication Workshop中说的部分。 本节课的教学内容便是合并后的第四课与Communication Workshop的读说课，教材中第四课的文章生词量不大，文章结构清楚，简单易懂，学生很容易总结出文章结构和相关有用的单词、短语，从而为下一步的写作打下基础。教材中要求学生最后完成的是一个Internet page的设计，在本节课中这个要求被弱化，只要求学生完成介绍一个地方的口头写作内容，在课后作业中，要求学生将课上的口头作文落实到笔头上。

高中英语 unit5 virtual reality-reading教案 上海牛津版S2A

Unit5 Virtual Reality Reading教案 一、章节分析（Reading section ） （一）综述 本章节讲述虚拟现实(VR)在各个领域的运用并分析其利弊。由于此主题较新,并与学生日常生活的电脑网络的使用有关,学生们对此应该是比较熟悉也颇有兴趣的。 因此，教师应充分利用学生的兴趣来教授本课，并进行适当拓展。 本课的任务有两个： 1.学生通过对课文的学习。掌握一些核心词汇，例如：imaginary, realistic, security, image等。 2.通过学习课文，了解如何运用想象写说明文，为writing部分做一定的铺垫和 准备。 （二）阅读目标 1知识目标 学习课文中重点词、词组、句型和语法。 2能力目标 通过阅读了解虚拟现实在各个领域的使用以及其他相关知识。 3情感目标 正确判断电脑、网络以及虚拟现实对日常生活的利弊影响。 （三）教学方法 采用任务型教学法组织教学，通过听说，讨论等具体活动，达到教学效果。 （四）重点和难点 1词汇学习 1)核心词汇 artificial

commit imaginary realistic security image amazing 2)拓展词汇 inspect manufacture architect amazing glide head-set inspect medium virtual fantasy 3)词组和短语 look down upon go back in time come true thanks to introduce … into reach out just for entertainment before long

标准日本语初级上册语法总结(改：动词原形)

日语初级上册语法总结 ㈠日语常用的词汇分类及用法： 1 名词：在句子中作主语，谓语，宾语，定语（名词+の名词）。 2 形容词：定语，谓语。 3 形容动词：定语，谓语。 4 动词：定语，谓语。 5 副词：可做状语，修饰动词，形容词，形容动词。 6 助词：相当中文里的助词，用于说明一个句子或一个词，与其它句子或词的关系。 ㈡动词的分类及「て形」、「ない形」、「た形」的变形规则。 ㈣上册所学语法中与「て」「ない」「た」相关的语法。 ①「て」：～ています、～てから、～てもいいです、～てください、～てはいけません、～ても 补充：动词的「て」形表示动作的先后顺序，以及动作行为的方式方法。 例えば：顔を洗って学校へ行きます。 歩いて駅へ行きます。 ②「ない」：～なくてもいいです、～ないでください、～ないほうがいいです、～なければならない 补充：动词的「ない」形表示否定。 例えば：会社へ行かない。 ③「た」：～たことがあります、～たほうがいいです、～たり～たりします、～たら 补充：动词的「た」形表示过去时。 例えば：フランスへ行った。

㈢名词，形容词，形容动词，动词的简体及敬体变形 ㈤常见助词用法的归纳总结。 １「は」用与提示主语，像「には、では、へは、からは、までは等」属于助词的重叠使用。 起加强语气或提示为主题的作用。 例えば：田中さんは日本人です。 教室には学生がいます。 2「が」提示主语和描述状态的作用。 常用「が」的情况有：1其后为形容词。 2表示自然现象。 3其前为疑问词。 4整句中的一小部分的主语。 5另外自动词前也用「が」来提示而不用「を」。 例えば：天気がいいです。 空は青いです。 誰がいますか。 私は足が痛い。 電気が付いている。 3「も」表示后项事物和前项事物一样。相当于中文的[也]。 例えば：陳さんは中国人です。 李さんも中国人です。

高中英语名师备课易20万个名师教学视频链接

20万个名师教学视频，特惠价只要9.9元，即可拥有终身观看特权，语文数学英语物理化学历史体育班会说课等等应有尽有，一网打尽，淘宝抢购地址： https://www.wendangku.net/doc/003482570.html,/item.htm?spm=a1z10.1.w137712-940 9678669.2.7B6Xjn&id=42393892739 专辑名称： ．．．．．高中英语优秀课教学视频 专辑所含视频 写作指导高中英语 雷蕾——sports and me 高中英语 石门中学狮山校区，王跃副校长英语示范课 石门中学狮山校区，高中英语，2010年南海区首席教师主题沙龙活动 高一英语《Money》桂城中学小塘分校林晓玲课例 北师大英语模块一unit 3 lesson 2 parties高中英语 北师大英语模块一unit 3 lesson 2 parties 高中英语 北师大版模块3 Unit8 Lesson2 Extreme Sports 高中英语 World News 教学课例高中英语 Virtual Tourism 高中英语 Unit9 Lesson4 Car Culture The Road to Destruc 高中英语 Unit 22 Lesson 1 Global Warming 阅读课第二课时高中英语 Reading and Internent 高中英语 module 4 之lesson one unit 10 Car Culture 高中英语 Culture corner the culture of money 高中英语 21世纪报刊阅读高中英语 专辑名称： ．．．．．高中英语广东名师课堂教学展示 专辑所含视频 A taste of English humour--整节课例_高中英语广东名师课堂教学展示视频 Body language--整节课例(1)_高中英语广东名师课堂教学展示视频 Body language--整节课例_高中英语广东名师课堂教学展示视频 Computers(Grammar)--整节课例_高中英语广东名师课堂教学展示视频 Fashionable Decisions--整节课例_高中英语广东名师课堂教学展示视频 Festivals around the world--整节课例_高中英语广东名师课堂教学展示视频 cloning--整节课例_高中英语广东名师课堂教学展示视频 English around the World--整节课例_高中英语广东名师课堂教学展示视频 Making the news--整节课例_高中英语广东名师课堂教学展示视频 New Zealand--整节课例_高中英语广东名师课堂教学展示视频 Poems--整节课例_高中英语广东名师课堂教学展示视频 Sub junctive Mood--整节课例_高中英语广东名师课堂教学展示视频

Unit 4 Cyberspace Lesson 3 Virtual Reality 教学设计3-优质公开课-北师大必修2精品

Unit 4 Cyberspace Lesson 3 Virtual Reality Teaching aims: To understand the dialogue. To get specific information about the virtual reality holidays. To voice your opinion on a virtual university. Teaching difficulties: How to make the students take part in the class actively. Before the class. You are divided into two groups: Group A and Group B. If some student in your group volunteersto answer the question, your group will get a smile face. The group that gets more smile faces willbe a winner. OK? Teaching procedures: Ⅰ. Lead in First,pleaselookatthreegroupsofpicturestodecide:whichoneshowsrealsituation? Which one shows unreal situation?Real/ Unreal They are all unreal situations, but they make us feel as if we are in real situations. Do youthink so? They are called virtual reality. Today we will read a passage---Lesson3 Virtual Reality Please look at objectives by yourselves. Do you understand? ⅡReading Task1. Skip to get main idea. Try to answer the two questions. 3’ 1. How many people are there in the dialogue? Who are they?

新标准日本语语法点总结(初级上)

新標準日本語語法點總結（初級上） 1.名【场所】名【物/人】力* / 庭忙何力?笳◎去T力、。 部屋忙誰力"、去T力、。 2.名【物/人】总名【场所】笳◎去T /「岷T 図書館乙忙笳◎去T力、。 猫总椅子 3. 表示方位 上*9^隣 下L尢中忌 力、 前外 後6 九勺 4.疑问词+哲+动词（否定）教室V誰去乜人。 冷蔵庫忙何哲笳◎去乜人。 5.去T 去乜人 6.时间的表达方式 今何時何分IT力、。 今四時三十分IT。 毎日何時忙寝去T力、。 11時3 0分忙寝去T。 （叙述包含数字的时间后续助词V,例3月14日V, 但是今日、今、昨日、明日、毎日、去年等词后不加V,星期后一般加V,但也可以不加。午前中 試験总始去◎去T力、。 来週①木曜日IT。 （詢問時間用 ^^,當詢問的時間很具體時，在表示時間的詞語后加V,如何時V、何曜日V、何日V ）力、5……表示某动作发生在某个期间，也可以表示某移動動作的起點和終點森月曜日力、5水曜日去疋休注5。

表達時刻【何時何分】 1點一時7點七時 2點二時8點八時 3點三時9點九時

高一英语Unit 4 Cyberspace导学案

高一英语Unit 4 Cyberspace导学案高一英语Unit4 cyberspace导学案 Lesson4VirtualTourism Tips:Thelandscapebelongstothemanwholooksatit. 风景属于看风景的人 Learnintaims： 1)Tomastersomeimportantwordsandphrases. 2)Tolearnsomeimportantsentences. Learningimportanganddifficultpoints：mustthefollowingwords:retire;apply;applyfor;dependo n;congratulation;ontime learningprocess: StepⅠFinishthewords .Theseriousearthquakecausedthetotal_________ofa newtown. 2.Afewminuteslater,theoldman___________inthecrowds. 3.Heo______me300dollarsforthattelevision. 4.youareinformed thatyouhavebeenemployedasourcompany’ssalesmanager.

5.Hehas________formetoattendthemeeting. 6.Hisaccents________hewasastrangerhere. 7.Scientistsoftendos________research. 8.He________arequestfromtheoldman. 9.Itissaidthatthis_________willbeputintoprisonforse ttingfire. 0.Thereareallkindsof_____________forchildrenonchild ren’sDay. 1.TheHopep_________hasdoesalotofworktohelpthosepoor studentswiththeirstudies. 2.Theoldmanmadeaverygood_________atthemeetinglastwe ek. 3.Thisisah_______eventofchina. 4.Beijingisourfinal___________ 5.Iknowtheplacewell,soletmebeyourg________. 6.myhometownis__________bythechangjiangRiver. 7.Iwanttoownahouseatthe___________. 8.Theylivedinthe________

北师大版高中英语必修二 unit4 lesson3 Virtual Reality reading

Unit4 Cyberspace-Lesson3 Virtual Reality 一、学习目标 1.透彻理解课文内容。 2.对新学到的单词和句型等能学以致用。 3.学习条件状语从句。 二、重难点分析 1.对课文里长难语句的理解以及句子成分分析。 2.学习条件状语从句。 三、学习过程 Step 1: 预习课文 1.用15分钟阅读，粗通课文的大意，回答下面的问题。 （1）What’s the main idea of the text? A. The weekend life of Cathy and Tom. B. The life that students hope for living in the future C. The imagination of virtual reality’s holiday and study situation D. Imagination of how to live in the future （2）Which of the following is true? A. In the virtual reality holiday, we wouldn’t experience something directly. B. Tome will go camping with his families this weekend. C. Tom will send the website address to Cathy when he gets to school. D. It’s not hard for us to imagine the virtual reality studying situation. （3）From the text, we can infer that________ . A. Cathy has lots of work to do this weekend. B. In the future, we will not have to go to our destinations in the flesh at all. C. We will not only be able to travel around the world, but go to study in world famous universities we wanted to. D. Cathy dislikes to spend a long time traveling on planes during holiday.

新版标准日本语初级上册语法总结

新版标准日语初级上册语法总结 ㈠日语常用的词汇分类及用法： 1 名词：在句子中作主语，谓语，宾语，定语（名词+の名词）。 2 形容词：定语，谓语。 3 形容动词：定语，谓语。 4 动词：定语，谓语。 5 副词：可做状语，修饰动词，形容词，形容动词。 6 助词：相当中文里的助词，用于说明一个句子或一个词，与其它句子或词的关系。㈡动词的分类及［ます形］「て形」、「ない形」、「た形」的变形规则。 动词的分类：ます形:て形: ない形: た形: ㈢名词，形容词，形容动词，动词的简体及敬体变形 ㈣上册所学语法中与「て」「ない」「た」相关的语法。 ㈤常见助词用法的归纳总结。 ㈥连词：连接句子于句子的词。 ㈦疑问词： ㈧副詞及接续词： 动词的分类： 动词「て形」「た形」的变形规则： １、一类动词： ①动词的最后一个假名以「うつる」结尾时，将它们改为「って」「た」 買う買って 立つ立って 終わる終わって

②动词的最后一个假名以「むすぶ」结尾时，将它们改为「んで」「た」 読む読んで 遊ぶ遊んで 死ぬ死んで ③动词的最后一个假名以「くぐ」结尾时，将它改为「いて」「た」 書く書いて 泳ぎぐ泳いで ④行く行って「た」 ⑤話す話して「た」 ２、二类动词:直接去掉加「て」「た」 食べる食べて出かける出かけて 鍛える鍛えて起きる起きて ３、三类动词：直接去掉「する」加「して」「た」。「来るー来（き）て」「た」。運動する運動して復習する復習して 買い物する買い物してチェックするチェックして 动词「ない形」的变形规则： １、一类动词：将动词「ます形」的最后一个假名改为其「あ」段假名。若动词「ます形」的最后一个假名以「い」结尾时不要将其改为「あ」，而要改为「わ」。 買う買わない 立つ立たない 読む読まない

firmware升级流程

版本均由TE R140，而工 系统迫切需要升级以满足新的功能需求和爱护要求。一样情形下，我们可通过图形网管命令直截了当对局/远端的芯片firmware升级。参考拓扑图如下： 升级操作方法 升级的对象要紧有EC2的firmware和ONU的firmware两项，每一项目又分为三个子项：boot程序，app程序和personality程序（boot可能不是必需的）。升级时，我们不关怀三个子项的名称，仅关怀这些程序是属于E C2的，依旧ONU的，如果我们要升级的子项是EC2上的，则应在图形网管上选择升级“EC2 firmware”;反之，如果是ONU上的子项升级，则应在图形网管上选择升级“ONU firmware”。另外，boot，app,personality这三个子项一样是严格按照顺序来升级的，即先升级boot，然后是app，最后是personality。升级时，待升级的对象必须在位，如果是ONU，它应该是差不多授权了的状态。 下面介绍整个升级流程。 假设目前网管具有以下条件： 网管服务器ip地址为10.26.1.2/16 欲使用的FTP服务器的用户名为test , 密码为test 文件名目为D:\ ONU的授权号为5，其对应在线的EC2槽位号为2。 第一预备好FTP服务器和要升级的所有文件： 打开FTP server 软件，并设置用户名和密码均为test，文件存放名目为D:\ 。

预备好升级文件。按照归档提供的*.tkf文件编辑好ONU的personality 文件（编辑方法见文档后的附录），并任意改好文件名，如ec2_https://www.wendangku.net/doc/003482570.html,f , e c2_https://www.wendangku.net/doc/003482570.html,f , ec2_https://www.wendangku.net/doc/003482570.html,f ，onu_https://www.wendangku.net/doc/003482570.html,f , onu_https://www.wendangku.net/doc/003482570.html,f , onu_https://www.wendangku.net/doc/003482570.html,f，分别放置于FTP对应的名目D:\下。注意文件名不要太长，应在网管能识别的范畴之类，一样来讲，包括后缀名在内，文件名不应超过16个字符。 然后升级EC2的firmware： 保证系统正常，网管连接正常。然后通过图形网管界面单击系统模块的GSW盘，依次选择配置----升级系统软件，将弹出一个对话框，第一填写对话框如下： 单击“升级系统软件”，并等待升级终止，EC2的boot程序即升级完毕。但目前确信不需要升级EC2的boot程序，可省去这步。 将上图对话框中的“文件名”项改为ec2_https://www.wendangku.net/doc/003482570.html,f，再单击“升级系统软件”，并等待升级终止，EC2的app程序即升级完毕。 将上图对话框中的“文件名”项改为ec2_https://www.wendangku.net/doc/003482570.html,f，再单击“升级系统软件”，并等待升级终止，EC2的pers程序即升级完毕。 EC2升级完成后，应登陆到系统的命令行网管中，在该EC2的debug 名目下执行restore olt 1 和restore olt 2命令。执行完毕后,两路EC2芯片将自行启动即可。 然后升级ONU的firmware： 保证系统正常，网管连接正常，ONU在线并差不多授权过。通过图形网管界面单击系统模块的GSW盘，依次选择配置----升级系统软件，将弹出一个对话框，第一填写对话框如下： 单击“升级系统软件”，并等待升级终止，ONU的boot程序即升级完毕。

新版标准日本语初级上册语法解释 第2课

新版标日初级·语法解释 第2课 1.これ/それ/あれは　[名]です 相当于汉语“这是/那是~”。 “これ”“それ”“あれ”是指代事物的词，相当于汉语“这、这个”“那、那个”。用法如下： （1）说话人与听话人有一点距离，面对面时： ·これ：距离说话人较近的事物 ·それ：距离听话人较近的事物 ·あれ：距离说话人和听话人都较远的事物 （2）说话人和听话人处于同一位置，面向同一方向时： ·これ：距离说话人、听话人较近的事物 ·それ：距离说话人、听话人较远的事物 ·あれ：距离说话人、听话人更远的事物 例：これは　本です。 それは　テレビです。 あれは　パソコンですか。 2.だれですか/何ですか 相当于汉语“~是什么？/~是谁？”。不知道是什么人是用“だれ”，不知道是什么东西时用“何”。句尾后续助词“か”，读升调。例：それは　何ですか。 あの人は　だれですか。 注意：“だれ”的礼貌说法是“どなた”。对方与自己是同辈、地位相当或地位较低时用“だれ”。对方比自己年长或地位高时用“どなた”。 例：吉田さんは　どなたですか。 3.[名]の[名]【所属】 助词“の”连接名词和名词，表示所属。 例：私のかぎ。 小野さんの傘。 4.この/その/あの[名]は　[名]です 相当于汉语“这个/那个~是~”。修饰名词时，要用“この”“その”“あの”。其表示的位置关系与“これ”“それ”“あれ”相同。例：このカメラは　私のです。 その傘は　小野さんのです。 あの車は　だれのですか。 5.どれ/どの[名] 三个以上的事物中，不能确定哪一个时用疑问词“どれ”“どの”。单独使用时用“どれ”，修饰名词时用“どの”。 例：森さんのかばんは　どれですか。 長島さんの靴は　どれですか。 私の机は　どの机ですか 扩展：100以下数字 0　れい/ぜろ 1　いち 2　に 3　さん 4　し/よん　 5　ご 6　ろく

Virtual Reality

Virtual Reality is a kind of computer simulation technique that is able to assist people experience virtual world by using special lens in head-sets. It can be applied in plenty of areas such as military training and medical science. For military training, with head-sets and some additional facilities, special situations like forests, desert and ruin can be virtually simulated. Then soldiers will feel that they are in real battlefield and complete tasks to improve themselves better. In terms of medical science, when facing dangerous operations, the data of patients’ tissues and organs can be input in the computer in advance. Then computer can create the structure of the patient in details. Surgeons wearing head-sets can practice on such simulation to get more familiar with the operation so as to handle unexpected dangerous condition well and avoid some mistakes. Virtual reality (VR) typically refers to computer technologies that use virtual reality headsets, sometimes in combination with physical spaces or multi-projected environments, to generate realistic images, sounds and other sensations that simulates a user's physical presence in a virtual or imaginary environment. A person using virtual reality equipment is able to "look around" the artificial world, and with high quality VR move about in it and interact with virtual features or items. VR headsets are head-mounted goggles with a screen in front of the eyes. Programs may include audio and sounds through speakers or headphones. VR systems that include transmission of vibrations and other sensations to the user through a game controller or other devices are known as haptic systems. This tactile information is generally known as force feedback in medical, video gaming and military training applications. Virtual reality also refers to remote communication environments which provide a virtual presence of users with through telepresence and telexistence or the use of a virtual artifact (VA). The immersive environment can be similar to the real world in order to create a lifelike experience grounded in reality or sci-fi. Augmented reality systems may also be considered a form of VR that layers virtual information over a live camera feed into a headset, or through a smart phone or tablet device.