A knowledge value chain is a sequence of intellectual tasks by which knowledge workers build their employer's unique competitive advantage and/or social and environmental benefit. As an example, the components of a research and development project form a knowledge value chain. Productivity improvements in a knowledge value chain may come from knowledge integration in its original sense of data systems consolidation. Improvements also flow from the knowledge integration that occurs when knowledge management techniques are applied to the continuous improvement of a business process or processes. The term first started coming into common use around 1999, appearing in management-related talks and papers. It was registered as a trademark in 2004 by TW Powell Co., a Manhattan company. Knowledge value chain processes Knowledge acquisition Knowledge storage Knowledge dissemination Knowledge application
Weight initialization
In deep learning, weight initialization or parameter initialization describes the initial step in creating a neural network. A neural network contains trainable parameters that are modified during training: weight initialization is the pre-training step of assigning initial values to these parameters. The choice of weight initialization method affects the speed of convergence, the scale of neural activation within the network, the scale of gradient signals during backpropagation, and the quality of the final model. Proper initialization is necessary for avoiding issues such as vanishing and exploding gradients and activation function saturation. Note that even though this article is titled "weight initialization", both weights and biases are used in a neural network as trainable parameters, so this article describes how both of these are initialized. Similarly, trainable parameters in convolutional neural networks (CNNs) are called kernels and biases, and this article also describes these. == Constant initialization == We discuss the main methods of initialization in the context of a multilayer perceptron (MLP). Specific strategies for initializing other network architectures are discussed in later sections. For an MLP, there are only two kinds of trainable parameters, called weights and biases. Each layer l {\displaystyle l} contains a weight matrix W ( l ) ∈ R n l − 1 × n l {\displaystyle W^{(l)}\in \mathbb {R} ^{n_{l-1}\times n_{l}}} and a bias vector b ( l ) ∈ R n l {\displaystyle b^{(l)}\in \mathbb {R} ^{n_{l}}} , where n l {\displaystyle n_{l}} is the number of neurons in that layer. A weight initialization method is an algorithm for setting the initial values for W ( l ) , b ( l ) {\displaystyle W^{(l)},b^{(l)}} for each layer l {\displaystyle l} . The simplest form is zero initialization: W ( l ) = 0 , b ( l ) = 0 {\displaystyle W^{(l)}=0,b^{(l)}=0} Zero initialization is usually used for initializing biases, but it is not used for initializing weights, as it leads to symmetry in the network, causing all neurons to learn the same features. In this page, we assume b = 0 {\displaystyle b=0} unless otherwise stated. Recurrent neural networks typically use activation functions with bounded range, such as sigmoid and tanh, since unbounded activation may cause exploding values. (Le, Jaitly, Hinton, 2015) suggested initializing weights in the recurrent parts of the network to identity and zero bias, similar to the idea of residual connections and LSTM with no forget gate. In most cases, the biases are initialized to zero, though some situations can use a nonzero initialization. For example, in multiplicative units, such as the forget gate of LSTM, the bias can be initialized to 1 to allow good gradient signal through the gate. For neurons with ReLU activation, one can initialize the bias to a small positive value like 0.1, so that the gradient is likely nonzero at initialization, avoiding the dying ReLU problem. == Random initialization == Random initialization means sampling the weights from a normal distribution or a uniform distribution, usually independently. === LeCun initialization === LeCun initialization, popularized in (LeCun et al., 1998), is designed to preserve the variance of neural activations during the forward pass. It samples each entry in W ( l ) {\displaystyle W^{(l)}} independently from a distribution with mean 0 and variance 1 / n l − 1 {\displaystyle 1/n_{l-1}} . For example, if the distribution is a continuous uniform distribution, then the distribution is U ( ± 3 / n l − 1 ) {\displaystyle {\mathcal {U}}(\pm {\sqrt {3/n_{l-1}}})} . === Glorot initialization === Glorot initialization (or Xavier initialization) was proposed by Xavier Glorot and Yoshua Bengio. It was designed as a compromise between two goals: to preserve activation variance during the forward pass and to preserve gradient variance during the backward pass. For uniform initialization, it samples each entry in W ( l ) {\displaystyle W^{(l)}} independently and identically from U ( ± 6 / ( n l + 1 + n l − 1 ) ) {\displaystyle {\mathcal {U}}(\pm {\sqrt {6/(n_{l+1}+n_{l-1})}})} . In the context, n l − 1 {\displaystyle n_{l-1}} is also called the "fan-in", and n l + 1 {\displaystyle n_{l+1}} the "fan-out". When the fan-in and fan-out are equal, then Glorot initialization is the same as LeCun initialization. === He initialization === As Glorot initialization performs poorly for ReLU activation, He initialization (or Kaiming initialization) was proposed by Kaiming He et al. for networks with ReLU activation. It samples each entry in W ( l ) {\displaystyle W^{(l)}} from N ( 0 , 2 / n l − 1 ) {\displaystyle {\mathcal {N}}(0,2/n_{l-1})} . === Orthogonal initialization === (Saxe et al. 2013) proposed orthogonal initialization: initializing weight matrices as uniformly random (according to the Haar measure) semi-orthogonal matrices, multiplied by a factor that depends on the activation function of the layer. It was designed so that if one initializes a deep linear network this way, then its training time until convergence is independent of depth. Sampling a uniformly random semi-orthogonal matrix can be done by initializing X {\displaystyle X} by IID sampling its entries from a standard normal distribution, then calculate ( X X ⊤ ) − 1 / 2 X {\displaystyle \left(XX^{\top }\right)^{-1/2}X} or its transpose, depending on whether X {\displaystyle X} is tall or wide. For CNN kernels with odd widths and heights, orthogonal initialization is done this way: initialize the central point by a semi-orthogonal matrix, and fill the other entries with zero. As an illustration, a kernel K {\displaystyle K} of shape 3 × 3 × c × c ′ {\displaystyle 3\times 3\times c\times c'} is initialized by filling K [ 2 , 2 , : , : ] {\displaystyle K[2,2,:,:]} with the entries of a random semi-orthogonal matrix of shape c × c ′ {\displaystyle c\times c'} , and the other entries with zero. (Balduzzi et al., 2017) used it with stride 1 and zero-padding. This is sometimes called the Orthogonal Delta initialization. Related to this approach, unitary initialization proposes to parameterize the weight matrices to be unitary matrices, with the result that at initialization they are random unitary matrices (and throughout training, they remain unitary). This is found to improve long-sequence modelling in LSTM. Orthogonal initialization has been generalized to layer-sequential unit-variance (LSUV) initialization. It is a data-dependent initialization method, and can be used in convolutional neural networks. It first initializes weights of each convolution or fully connected layer with orthonormal matrices. Then, proceeding from the first to the last layer, it runs a forward pass on a random minibatch, and divides the layer's weights by the standard deviation of its output, so that its output has variance approximately 1. === Fixup initialization === In 2015, the introduction of residual connections allowed very deep neural networks to be trained, much deeper than the ~20 layers of the previous state of the art (such as the VGG-19). Residual connections gave rise to their own weight initialization problems and strategies. These are sometimes called "normalization-free" methods, since using residual connection could stabilize the training of a deep neural network so much that normalizations become unnecessary. Fixup initialization is designed specifically for networks with residual connections and without batch normalization, as follows: Initialize the classification layer and the last layer of each residual branch to 0. Initialize every other layer using a standard method (such as He initialization), and scale only the weight layers inside residual branches by L − 1 2 m − 2 {\displaystyle L^{-{\frac {1}{2m-2}}}} . Add a scalar multiplier (initialized at 1) in every branch and a scalar bias (initialized at 0) before each convolution, linear, and element-wise activation layer. Similarly, T-Fixup initialization is designed for Transformers without layer normalization. === Others === Instead of initializing all weights with random values on the order of O ( 1 / n ) {\displaystyle O(1/{\sqrt {n}})} , sparse initialization initialized only a small subset of the weights with larger random values, and the other weights zero, so that the total variance is still on the order of O ( 1 ) {\displaystyle O(1)} . Random walk initialization was designed for MLP so that during backpropagation, the L2 norm of gradient at each layer performs an unbiased random walk as one moves from the last layer to the first. Looks linear initialization was designed to allow the neural network to behave like a deep linear network at initialization, since W R e L U ( x ) − W R e L U ( − x ) = W x {\displaystyle W\;\mathrm {ReLU} (x)-W\;\mathrm {ReLU} (-x)=Wx} . It initializes a matrix W {\displaystyle W} of shape R n 2 × m {\displaystyle \mathbb {R} ^{{\frac {n}{2}}\times m}} by any method, such as orthogonal initialization, t
User modeling
User modeling is the subdivision of human–computer interaction which describes the process of building up and modifying a conceptual understanding of the user. The main goal of user modeling is customization and adaptation of systems to the user's specific needs. The system needs to "say the 'right' thing at the 'right' time in the 'right' way". To do so it needs an internal representation of the user. Another common purpose is modeling specific kinds of users, including modeling of their skills and declarative knowledge, for use in automatic software-tests. User-models can thus serve as a cheaper alternative to user testing but should not replace user testing. == Background == A user model is the collection and categorization of personal data associated with a specific user. A user model is a (data) structure that is used to capture certain characteristics about an individual user, and a user profile is the actual representation in a given user model. The process of obtaining the user profile is called user modeling. Therefore, it is the basis for any adaptive changes to the system's behavior. Which data is included in the model depends on the purpose of the application. It can include personal information such as users' names and ages, their interests, their skills and knowledge, their goals and plans, their preferences and their dislikes or data about their behavior and their interactions with the system. There are different design patterns for user models, though often a mixture of them is used. Static user models Static user models are the most basic kinds of user models. Once the main data is gathered they are normally not changed again, they are static. Shifts in users' preferences are not registered and no learning algorithms are used to alter the model. Dynamic user models Dynamic user models allow a more up to date representation of users. Changes in their interests, their learning progress or interactions with the system are noticed and influence the user models. The models can thus be updated and take the current needs and goals of the users into account. Stereotype based user models Stereotype based user models are based on demographic statistics. Based on the gathered information users are classified into common stereotypes. The system then adapts to this stereotype. The application therefore can make assumptions about a user even though there might be no data about that specific area, because demographic studies have shown that other users in this stereotype have the same characteristics. Thus, stereotype based user models mainly rely on statistics and do not take into account that personal attributes might not match the stereotype. However, they allow predictions about a user even if there is rather little information about him or her. Highly adaptive user models Highly adaptive user models try to represent one particular user and therefore allow a very high adaptivity of the system. In contrast to stereotype based user models they do not rely on demographic statistics but aim to find a specific solution for each user. Although users can take great benefit from this high adaptivity, this kind of model needs to gather a lot of information first. == Data gathering == Information about users can be gathered in several ways. There are three main methods: Asking for specific facts while (first) interacting with the system Mostly this kind of data gathering is linked with the registration process. While registering users are asked for specific facts, their likes and dislikes and their needs. Often the given answers can be altered afterwards. Learning users' preferences by observing and interpreting their interactions with the system In this case users are not asked directly for their personal data and preferences, but this information is derived from their behavior while interacting with the system. The ways they choose to accomplish a tasks, the combination of things they takes interest in, these observations allow inferences about a specific user. The application dynamically learns from observing these interactions. Different machine learning algorithms may be used to accomplish this task. A hybrid approach which asks for explicit feedback and alters the user model by adaptive learning This approach is a mixture of the ones above. Users have to answer specific questions and give explicit feedback. Furthermore, their interactions with the system are observed and the derived information are used to automatically adjust the user models. Though the first method is a good way to quickly collect main data it lacks the ability to automatically adapt to shifts in users' interests. It depends on the users' readiness to give information and it is unlikely that they are going to edit their answers once the registration process is finished. Therefore, there is a high likelihood that the user models are not up to date. However, this first method allows the users to have full control over the collected data about them. It is their decision which information they are willing to provide. This possibility is missing in the second method. Adaptive changes in a system that learns users' preferences and needs only by interpreting their behavior might appear a bit opaque to the users, because they cannot fully understand and reconstruct why the system behaves the way it does. Moreover, the system is forced to collect a certain amount of data before it is able to predict the users' needs with the required accuracy. Therefore, it takes a certain learning time before a user can benefit from adaptive changes. However, afterwards these automatically adjusted user models allow a quite accurate adaptivity of the system. The hybrid approach tries to combine the advantages of both methods. Through collecting data by directly asking its users it gathers a first stock of information which can be used for adaptive changes. By learning from the users' interactions it can adjust the user models and reach more accuracy. Yet, the designer of the system has to decide, which of these information should have which amount of influence and what to do with learned data that contradicts some of the information given by a user. == System adaptation == Once a system has gathered information about a user it can evaluate that data by preset analytical algorithm and then start to adapt to the user's needs. These adaptations may concern every aspect of the system's behavior and depend on the system's purpose. Information and functions can be presented according to the user's interests, knowledge or goals by displaying only relevant features, hiding information the user does not need, making proposals what to do next and so on. One has to distinguish between adaptive and adaptable systems. In an adaptable system the user can manually change the system's appearance, behavior or functionality by actively selecting the corresponding options. Afterwards the system will stick to these choices. In an adaptive system a dynamic adaption to the user is automatically performed by the system itself, based on the built user model. Thus, an adaptive system needs ways to interpret information about the user in order to make these adaptations. One way to accomplish this task is implementing rule-based filtering. In this case a set of IF... THEN... rules is established that covers the knowledge base of the system. The IF-conditions can check for specific user-information and if they match the THEN-branch is performed which is responsible for the adaptive changes. Another approach is based on collaborative filtering. In this case information about a user is compared to that of other users of the same systems. Thus, if characteristics of the current user match those of another, the system can make assumptions about the current user by presuming that he or she is likely to have similar characteristics in areas where the model of the current user is lacking data. Based on these assumption the system then can perform adaptive changes. == Usages == Adaptive hypermedia: In an adaptive hypermedia system the displayed content and the offered hyperlinks are chosen on basis of users' specific characteristics, taking their goals, interests, knowledge and abilities into account. Thus, an adaptive hypermedia system aims to reduce the "lost in hyperspace" syndrome by presenting only relevant information. Adaptive educational hypermedia: Being a subdivision of adaptive hypermedia the main focus of adaptive educational hypermedia lies on education, displaying content and hyperlinks corresponding to the user's knowledge on the field of study. Intelligent tutoring system: Unlike adaptive educational hypermedia systems intelligent tutoring systems are stand-alone systems. Their aim is to help students in a specific field of study. To do so, they build up a user model where they store information about abilities, knowledge and needs of the user. The system can now adapt to this user by presenting approp
Hubert Dreyfus's views on artificial intelligence
Hubert Dreyfus was a critic of artificial intelligence research. In a series of papers and books, including Alchemy and AI (1965), What Computers Can't Do (1972; 1979; 1992) and Mind over Machine (1986), he presented a skeptical and cautious assessment of AI's progress and a critique of the philosophical foundations of the field. Dreyfus' objections are discussed in most introductions to the philosophy of artificial intelligence, including Russell & Norvig (2021), a standard AI textbook, and in Fearn (2007), a survey of contemporary philosophy. Dreyfus argued that human intelligence and expertise depend primarily on yet-to-be understood informal and unconscious processes rather than symbolic manipulation and that these essentially human skills cannot be fully captured in formal rules. His critique was based on the insights of modern continental philosophers such as Merleau-Ponty and Heidegger, and was directed at the first wave of AI research which tried to reduce intelligence to high level formal symbols. When Dreyfus' ideas were first introduced in the mid-1960s, they were met in the AI community with ridicule and outright hostility. By the 1980s, however, some of his perspectives were rediscovered by researchers working in robotics and the new field of connectionism—approaches that were called "sub-symbolic" at the time because they eschewed early AI research's emphasis on high level symbols. In the 21st century, "sub-symbolic" artificial neural networks and other statistics-based approaches to machine learning were highly successful. Historian and AI researcher Daniel Crevier wrote: "time has proven the accuracy and perceptiveness of some of Dreyfus's comments." Dreyfus said in 2007, "I figure I won and it's over—they've given up." == Dreyfus' critique == === The grandiose promises of artificial intelligence === In Alchemy and AI (1965) and What Computers Can't Do (1972), Dreyfus summarized the history of artificial intelligence and ridiculed the unbridled optimism that permeated the field. For example, Herbert A. Simon, following the success of his program General Problem Solver (1957), predicted that by 1967: A computer would be world champion in chess. A computer would discover and prove an important new mathematical theorem. Most theories in psychology will take the form of computer programs. The press dutifully reported these predictions of the imminent arrival of machine intelligence. Dreyfus felt that this optimism was unwarranted and, in 1965, argued forcefully that predictions like these would not come true. He would eventually be proven right. Pamela McCorduck explains Dreyfus' position: A great misunderstanding accounts for public confusion about thinking machines, a misunderstanding perpetrated by the unrealistic claims researchers in AI have been making, claims that thinking machines are already here, or at any rate, just around the corner. These predictions were based on the success of the cognitive revolution, which promoted an "information processing" model of the mind. It was articulated by Newell and Simon in their physical symbol systems hypothesis, and later expanded into a philosophical position known as computationalism by philosophers such as Jerry Fodor and Hilary Putnam. In AI, the approach is now called symbolic AI or "GOFAI". Dreyfus argued that "symbolic AI" was the latest version of the ancient program of rationalism in philosophy. Rationalism had come under heavy criticism in the 20th century from philosophers like Martin Heidegger and Edmund Husserl. The mind, according to modern continental philosophy, is not "rationalist" and is nothing like a digital computer. Cognitivism led early AI researchers to believe that they had successfully simulated the essential process of human thought, thus it seemed a short step to producing fully intelligent machines. Dreyfus' last paper detailed the ongoing history of the "first step fallacy", where AI researchers tend to wildly extrapolate initial success as promising, perhaps even guaranteeing, wild future successes. === Dreyfus' four assumptions of artificial intelligence research === In Alchemy and AI and What Computers Can't Do, Dreyfus identified four philosophical assumptions, at least one of which he deems necessary for AI to succeed. "In each case," Dreyfus writes, "the assumption is taken by workers in AI as an axiom, guaranteeing results, whereas it is, in fact, one hypothesis among others, to be tested by the success of such work." Dreyfus argues that AI would be impossible without accepting at least one of these four assumptions: The biological assumption The brain processes information in discrete operations by way of some biological equivalent of on/off switches. In the early days of research into neurology, scientists found that neurons fire in all-or-nothing pulses. Several researchers, such as Walter Pitts and Warren McCulloch, speculated with great confidence that neurons functioned similarly to the way Boolean logic gates operate, and so could be imitated by electronic circuitry at the level of the neuron. When digital computers became widely used in the early 50s, this argument was extended to suggest that the brain was a vast physical symbol system, manipulating the binary symbols of zero and one. Dreyfus was able to refute the biological assumption by citing research in neurology that suggested that the action and timing of neuron firing had analog components. But Daniel Crevier observes that "few still held that belief in the early 1970s, and nobody argued against Dreyfus" about the biological assumption. The psychological assumption The mind can be viewed as a device operating on bits of information according to formal rules. He refuted this assumption by showing that much of what we know about the world consists of complex attitudes or tendencies that make us lean towards one interpretation over another. He argued that, even when we use explicit symbols, we are using them against an unconscious and informal background including commonsense knowledge and that without this background our symbols cease to mean anything. This background, in Dreyfus' view, was not implemented in individual brains as explicit individual symbols with explicit individual meanings. The epistemological assumption All knowledge can be formalized. This concerns the philosophical issue of epistemology, or the study of knowledge. Even if we agree that the psychological assumption is false, AI researchers could still argue (as AI founder John McCarthy has) that it is possible for a symbol processing machine to represent all knowledge, regardless of whether human beings represent knowledge the same way. Dreyfus argued that there is no justification for this assumption, since so much of human knowledge is not symbolic or even expressible using formal constructs. The ontological assumption The world consists of independent facts that can be represented by independent symbols AI researchers (and futurists and science fiction writers) often assume that there is no limit to formal, scientific knowledge, because they assume that any phenomenon in the universe can be described by symbols or scientific theories. This assumes that everything that exists can be understood as objects, properties of objects, classes of objects, relations of objects, and so on: precisely those things that can be described by logic, language and mathematics. The study of being or existence is called ontology, and so Dreyfus calls this the ontological assumption. If this is false, then it raises doubts about what we can ultimately know and what intelligent machines will ultimately be able to help us to do. === Knowing-how vs. knowing-that: the primacy of intuition === In Mind Over Machine (1986), written (with his brother) during the heyday of expert systems, Dreyfus analyzed the difference between human expertise and the programs that claimed to capture it. This expanded on ideas from What Computers Can't Do, where he had made a similar argument criticizing the "cognitive simulation" school of AI research practiced by Allen Newell and Herbert A. Simon in the 1960s. Dreyfus argued that human problem solving and expertise depend on our background sense of the context, of what is important and interesting given the situation, rather than on the process of searching through combinations of possibilities to find what we need. Dreyfus would describe it in 1986 as the difference between "knowing-that" and "knowing-how", based on Heidegger's distinction of present-at-hand and ready-to-hand. Knowing-that is our conscious, step-by-step problem solving abilities. We use these skills when we encounter a difficult problem that requires us to stop, step back and search through ideas one at time. At moments like this, the ideas become very precise and simple: they become context free symbols, which we manipulate using logic and language. These are the skills that Newell and Simon had demonstrated with both psy
AlphaGo Zero
AlphaGo Zero is a version of DeepMind's Go software AlphaGo. AlphaGo's team published an article in Nature in October 2017 introducing AlphaGo Zero, a version created without using data from human games, and stronger than any previous version. By playing games against itself, AlphaGo Zero: surpassed the strength of AlphaGo Lee in three days by winning 100 games to 0; reached the level of AlphaGo Master in 21 days; and exceeded all previous versions in 40 days. Training artificial intelligence (AI) without datasets derived from human experts has significant implications for the development of AI with superhuman skills, as expert data is "often expensive, unreliable, or simply unavailable." Demis Hassabis, the co-founder and CEO of DeepMind, said that AlphaGo Zero was so powerful because it was "no longer constrained by the limits of human knowledge". Furthermore, AlphaGo Zero performed better than standard deep reinforcement learning models (such as Deep Q-Network implementations) due to its integration of Monte Carlo tree search. David Silver, one of the first authors of DeepMind's papers published in Nature on AlphaGo, said that it is possible to have generalized AI algorithms by removing the need to learn from humans. Google later developed AlphaZero, a generalized version of AlphaGo Zero that could play chess and shōgi in addition to Go. In December 2017, AlphaZero beat the 3-day version of AlphaGo Zero by winning 60 games to 40, and with 8 hours of training it outperformed AlphaGo Lee on an Elo scale. AlphaZero also defeated a top chess program (Stockfish) and a top Shōgi program (Elmo). == Architecture == The network in AlphaGo Zero is a ResNet with two heads. The stem of the network takes as input a 17x19x19 tensor representation of the Go board. 8 channels are the positions of the current player's stones from the last eight time steps. (1 if there is a stone, 0 otherwise. If the time step go before the beginning of the game, then 0 in all positions.) 8 channels are the positions of the other player's stones from the last eight time steps. 1 channel is all 1 if black is to move, and 0 otherwise. The body is a ResNet with either 20 or 40 residual blocks and 256 channels. There are two heads, a policy head and a value head. Policy head outputs a logit array of size 19 × 19 + 1 {\displaystyle 19\times 19+1} , representing the logit of making a move in one of the points, plus the logit of passing. Value head outputs a number in the range ( − 1 , + 1 ) {\displaystyle (-1,+1)} , representing the expected score for the current player. -1 represents current player losing, and +1 winning. == Training == AlphaGo Zero's neural network was trained using TensorFlow, with 64 GPU workers and 19 CPU parameter servers. Only four TPUs were used for inference. The neural network initially knew nothing about Go beyond the rules. Unlike earlier versions of AlphaGo, Zero only perceived the board's stones, rather than having some rare human-programmed edge cases to help recognize unusual Go board positions. The AI engaged in reinforcement learning, playing against itself until it could anticipate its own moves and how those moves would affect the game's outcome. In the first three days AlphaGo Zero played 4.9 million games against itself in quick succession. It appeared to develop the skills required to beat top humans within just a few days, whereas the earlier AlphaGo took months of training to achieve the same level. According to Epoch.ai, training cost 3e23 FLOPs. For comparison, the researchers also trained a version of AlphaGo Zero using human games, AlphaGo Master, and found that it learned more quickly, but actually performed more poorly in the long run. DeepMind submitted its initial findings in a paper to Nature in April 2017, which was then published in October 2017. == Hardware cost == The hardware cost for a single AlphaGo Zero system in 2017, including the four TPUs, has been quoted as around $25 million. == Applications == According to Hassabis, AlphaGo's algorithms are likely to be of the most benefit to domains that require an intelligent search through an enormous space of possibilities, such as protein folding (see AlphaFold) or accurately simulating chemical reactions. AlphaGo's techniques are probably less useful in domains that are difficult to simulate, such as learning how to drive a car. DeepMind stated in October 2017 that it had already started active work on attempting to use AlphaGo Zero technology for protein folding, and stated it would soon publish new findings. == Reception == AlphaGo Zero was widely regarded as a significant advance, even when compared with its groundbreaking predecessor, AlphaGo. Oren Etzioni of the Allen Institute for Artificial Intelligence called AlphaGo Zero "a very impressive technical result" in "both their ability to do it—and their ability to train the system in 40 days, on four TPUs". The Guardian called it a "major breakthrough for artificial intelligence", citing Eleni Vasilaki of Sheffield University and Tom Mitchell of Carnegie Mellon University, who called it an impressive feat and an “outstanding engineering accomplishment" respectively. Mark Pesce of the University of Sydney called AlphaGo Zero "a big technological advance" taking us into "undiscovered territory". Gary Marcus, a psychologist at New York University, has cautioned that for all we know, AlphaGo may contain "implicit knowledge that the programmers have about how to construct machines to play problems like Go" and will need to be tested in other domains before being sure that its base architecture is effective at much more than playing Go. In contrast, DeepMind is "confident that this approach is generalisable to a large number of domains". In response to the reports, South Korean Go professional Lee Sedol said, "The previous version of AlphaGo wasn’t perfect, and I believe that’s why AlphaGo Zero was made." On the potential for AlphaGo's development, Lee said he will have to wait and see but also said it will affect young Go players. Mok Jin-seok, who directs the South Korean national Go team, said the Go world has already been imitating the playing styles of previous versions of AlphaGo and creating new ideas from them, and he is hopeful that new ideas will come out from AlphaGo Zero. Mok also added that general trends in the Go world are now being influenced by AlphaGo's playing style. "At first, it was hard to understand and I almost felt like I was playing against an alien. However, having had a great amount of experience, I’ve become used to it," Mok said. "We are now past the point where we debate the gap between the capability of AlphaGo and humans. It’s now between computers." Mok has reportedly already begun analyzing the playing style of AlphaGo Zero along with players from the national team. "Though having watched only a few matches, we received the impression that AlphaGo Zero plays more like a human than its predecessors," Mok said. Chinese Go professional Ke Jie commented on the remarkable accomplishments of the new program: "A pure self-learning AlphaGo is the strongest. Humans seem redundant in front of its self-improvement." == Comparison with predecessors == == AlphaZero == On 5 December 2017, DeepMind team released a preprint on arXiv, introducing AlphaZero, a program using generalized AlphaGo Zero's approach, which achieved within 24 hours a superhuman level of play in chess, shogi, and Go, defeating world-champion programs, Stockfish, Elmo, and 3-day version of AlphaGo Zero in each case. AlphaZero (AZ) is a more generalized variant of the AlphaGo Zero (AGZ) algorithm, and is able to play shogi and chess as well as Go. Differences between AZ and AGZ include: AZ has hard-coded rules for setting search hyperparameters. The neural network is now updated continually. Chess (unlike Go) can end in a tie; therefore AZ can take into account the possibility of a tie game. An open source program, Leela Zero, based on the ideas from the AlphaGo papers is available. It uses a GPU instead of the TPUs recent versions of AlphaGo rely on.
Soterml
SoTerML (Soil and Terrain Markup Language) is a XML-based markup language for storing and exchanging soil and terrain related data. SoTerML development is being done within The e-SoTer Platform. GEOSS plans a global Earth Observation System and, within this framework, the e-SOTER project addresses the felt need for a global soil and terrain database. The Centre for Geospatial Science (Currently Nottingham Geospatial Institute) at the University of Nottingham has initiated the development since January 2009. Further development and maintenance is currently handled in National Soil Resources Institute (NSRI) at Cranfield University, UK. The role of CGS is within the development of the e-SOTER dissemination platform, which is based on INSPIRE principles. The SoTerML development included: 1. Development of a data dictionary for nomenclatures and various data sources (data and metadata). 2. Development of an exchange format/procedures from the World Reference Base 2006.
Journal of Experimental and Theoretical Artificial Intelligence
The Journal of Experimental and Theoretical Artificial Intelligence is a quarterly peer-reviewed scientific journal published by Taylor and Francis. It covers all aspects of artificial intelligence and was established in 1989. The editor-in-chief is Eric Dietrich (Binghamton University), the deputy editors-in-chief are Li Pheng Khoo (School of Mechanical & Aerospace Engineering, Nanyang Technological University) and Antonio Lieto (Department of Computer Science, University of Turin). == Abstracting and indexing == The journal is abstracted and indexed in: According to the Journal Citation Reports, the journal has a 2020/2021 impact factor of 2.340 .