Search

The paper discusses algorithmic priority inversion in mission-critical machine inference pipelines used in modern neural-network-based perception subsystems and describes a solution to mitigate its effect. In general, priority inversion occurs in computing systems when computations that are "less important" are performed together with or ahead of those that are "more important." Significant priority inversion occurs in existing machine inference pipelines when they do not differentiate between critical and less critical data. We describe a framework to resolve this problem and demonstrate that it improves a perception system's ability to react to critical inputs, while at the same time reducing platform cost. [ABSTRACT FROM AUTHOR]

Time and state descriptions form the core of a simulation model representation. The historical influence of initial application areas and the exigencies of language implementations have created a muddled view of the time and state relationships. As a consequence, users of simulation programming languages work in relative isolation; model development, simulation application, model portability, and the communication of results are inhibited and simulation practice fails to contribute to the recognition of an underlying foundation or integrating structure. A model representation structure has been forged from a small set of basic definitions which carefully distinguish the state and time relationships. This paper focuses on the coordination of the time and state concepts using "object" as the link. In addition to clarifying the relationships, the structure relates the concept of ""state-sequenced simulation" to the variations in time flow mechanisms. In conclusion, some speculations are offered regarding alternative algorithms for time flow mechanisms. [ABSTRACT FROM AUTHOR]

This article investigates artificial intelligence (AI) fairness in decision-making specifically examining the limitations of this process. The article discusses eight inherent limitations in AI fairness including lack of fairness in human decision-making used to train the AI model, lack of a universal fairness definition, and limited power over the complete decision-making process.

The article discusses Explainable artificial intelligence (XAI) and the expectation of artificial intelligence (AI) end users to understand how AI works, including how algorithmic decision making occurs. An overview of AI end users' desire for information on AI methodology on reaching decisions, including in regard to the ethical nature of AI, is provided.

The ancient oriental game of Go has long been considered a grand challenge for artificial intelligence. For decades, computer Go has defied the classical methods in game tree search that worked so successfully for chess and checkers. However, recent play in computer Go has been transformed by a new paradigm for tree search based on Monte-Carlo methods. Programs based on Monte-Carlo tree search now play at human-master levels and are beginning to challenge top professional players. In this paper, we describe the leading algorithms for Monte-Carlo tree search and explain how they have advanced the state of the art in computer Go. [ABSTRACT FROM AUTHOR]

Recognizing objects in images is an active area of research in computer vision. In the last two decades, there has been much progress and there are already object recognition systems operating in commercial products. However, most of the algorithms for detecting objects perform an exhaustive search across all locations and scales in the image comparing local image regions with an object model. That approach ignores the semantic structure of scenes and tries to solve the recognition problem by brute force. In the real world, objects tend to covary with other objects, providing a rich collection of contextual associations. These contextual associations can be used to reduce the search space by looking only in places in which the object is expected to be; this also increases performance, by rejecting patterns that look like the target but appear in unlikely places. Most modeling attempts so far have defined the context of an object in terms of other previously recognized objects. The drawback of this approach is that inferring the context becomes as difficult as detecting each object. An alternative view of context relies on using the entire scene information holistically. This approach is algorithmically attractive since it dispenses with the need for a prior step of individual object recognition. In this paper, we use a probabilistic framework for encoding the relationships between context and object properties and we show how an integrated system provides improved performance. We view this as a significant step toward general purpose machine vision systems. [ABSTRACT FROM AUTHOR]

Autonomous helicopter flight is widely regarded to be a highly challenging control problem. As helicopters are highly unstable and exhibit complicated dynamical behavior, it is particularly difficult to design controllers that achieve high performance over a broad flight regime. While these aircraft are notoriously difficult to control, there are expert human pilots who are nonetheless capable of demonstrating a wide variety of maneuvers, including aerobatic maneuvers at the edge of the helicopter's performance envelope. In this paper, we present algorithms for modeling and control that leverage these demonstrations to build high-performance control systems for autonomous helicopters. More specifically, we detail our experiences with the Stanford Autonomous Helicopter, which is now capable of extreme aerobatic flight meeting or exceeding the performance of our own expert pilot. [ABSTRACT FROM AUTHOR]

This article discusses the recent areas of focus at the National Institute of Informatics (NII) in Japan. First, a look at its academic services including SINET6, a backbone network service across Japan and its data platform service, the NII Research Data Cloud. The article also details NII research in computer science with projects in innovative computer vision and engineerable artificial intelligence, as well as applied research projects including a COVID-19 pneumonia diagnosis tool.

This article discusses how the combination of humans and artificial intelligence could target misinformation and polarization, by examining two scenarios- one where computing methods can assist human fact-checkers and one where human search methods can assist in search engine algorithms for personalization. Topics include the Australian Communication and Media Authority research on misinformation in Australia, whether filter bubbles are created by personalization attempts of algorithms, and the various computational methods that can assist fact-checking to prevent misinformation.

This paper suggests a bidirectional relationship between state-space and problem- reduction representations. It resents a formalism based on multiple-input and multiple-output operators which provides a basis for viewing the two types of representations in this manner. A representation of the language recognition problem which is based on the Cocke parsing algorithm Is used as an illustration. A method for representing problems in first-order logic in such a way that the inference system employed by a resolution-based theorem prover determines whether the set of clauses is interpreted in the state-space mode or in the problem-reduction mode is presented. The analogous concepts in problem-reduction and theorem proving, and the terminology used to refer to them, are noted. The relationship between problem-reduction, input resolution, and linear resolution is is discussed. [ABSTRACT FROM AUTHOR]

The article focuses on the search for algorithms that can reduce the time needed for multiplication problems within the fields of machine learning (ML), deep learning (DL), and artificial intelligence (AI). Topics discussed include the history of matrix multiplication and the data science group DeepMind and their ML approach AlphaTensor.

The article presents three principles concerning five challenges to the establishment of trustworthy artificial intelligence (AI), including the development computer languages and algorithms for data modeling, trustworthy-by-construction design, and complex machine learning (ML) representations.

The article examines trust in artificially intelligent (AI) systems as well as their regulation by various governments including those of the European Union, U.S., and China. The author predicates trust in AI on its perceived safety and reliability and calls for the continued human oversight of AI systems ("human-in-the-loop" or "society-in-the-loop"). Applications of artificial intelligence are outlined including military strategy, law enforcement, and healthcare.

The article utilizes the adaptive learning computer system AlphaZero in order to examine how variations of the game of chess would proceed at what is referred to as a superhuman level. According to the article, AlphaZero marks a significant milestone in the field of artificial intelligence (AI) and the use of algorithms in computer chess.

The article discusses the concept of topic novelty and its impact upon accuracy in the news and journalism, and whether artificial intelligence (AI) algorithms can counter fake news. The article discusses AI, accuracy, and fake news during the COVID-19 pandemic and an experiment performed on Amazon Mechanical Turk (AMT).

The article discusses the development and relative effectiveness of anti-surveillance cosmetic makeup designed to elude facial-recognition technology. It also examines the rapid pace of computer vision development, facial-detection algorithms, and how law enforcement agencies utilize facial recognition systems.

Commonsense reasoning remains a major challenge in AI, and yet, recent progresses on benchmarks may seem to suggest otherwise. In particular, the recent neural language models have reported above 90% accuracy on the Winograd Schema Challenge (WSC),22 a commonsense benchmark originally designed to be unsolvable for statistical models that rely simply on word associations. This raises an important question--whether these models have truly acquired robust commonsense capabilities or they rely on spurious biases in the dataset that lead to an overestimation of the true capabilities of machine commonsense. To investigate this question, we introduce WinoGrande, a large-scale dataset of 44k problems, inspired by the original WSC, but adjusted to improve both the scale and the hardness of the dataset. The key steps of the dataset construction consist of (1) large-scale crowdsourcing, followed by (2) systematic bias reduction using a novel AfLite algorithm that generalizes human-detectable word associations to machine-detectable embedding associations. Our experiments demonstrate that state-of-the-art models achieve considerably lower accuracy (59.4%-79.1%) on WinoGrande compared to humans (94%), confirming that the high performance on the original WSC was inflated by spurious biases in the dataset. Furthermore, we report new state-of-the-art results on five related benchmarks with emphasis on their dual implications. On the one hand, they demonstrate the effectiveness of WinoGrande when used as a resource for transfer learning. On the other hand, the high performance on all these benchmarks suggests the extent to which spurious biases are prevalent in all such datasets, which motivates further research on algorithmic bias reduction. [ABSTRACT FROM AUTHOR]

The article criticizes the use of artificial intelligence in employee hiring. The algorithms utilized to screen resumes of applicants are characterized as inadequate and even discriminatory. The design of recruitment software is examined along with the alleged drive for "hyperefficiency" among human resource departments.

Generative adversarial networks are a kind of artificial intelligence algorithm designed to solve the generative modeling problem. The goal of a generative model is to study a collection of training examples and learn the probability distribution that generated them. Generative Adversarial Networks (GANs) are then able to generate more examples from the estimated probability distribution. Generative models based on deep learning are common, but GANs are among the most successful generative models (especially in terms of their ability to generate realistic highresolution images). GANs have been successfully applied to a wide variety of tasks (mostly in research settings) but continue to present unique challenges and research opportunities because they are based on game theory while most other approaches to generative modeling are based on optimization. [ABSTRACT FROM AUTHOR]

The article discusses the challenges associated with trusting the behavior of complex artificial intelligence (AI) algorithms used in mission-critical settings and specifying what makes a system intelligible. Topics include the ability to explain an AI system's decisions using techniques such as local approximation and vocabulary transformation, and the promotion of interactive explanation systems that can respond to a range of follow-up questions allowing explanation of a process.

The article examines the use of machine learning- and deep learning-based artificial intelligence tools to safeguard user privacy on computer networks of the future. Compliance with data privacy regulations such as the CCPA and GDPR is scrutinized and the application of AI-based approaches to telephone company networks is discussed.

The article discusses the design of autonomous, adaptive machines through evolutionary robotics that incorporate features based in biological systems as of August 2013. It suggests that useful industrial robots can be produced through simulators and then manufactured, avoiding challenges associated with the manual design of autonomous, adaptive robots. Topics include metaheuristic algorithms, developmental robotics, artificial intelligence, swarm robotics, dynamical systems theory, and challenges resulting from reality gaps.

The article discusses the search algorithm A* and the origin of its name in computer science and artificial intelligence (AI).

The article discusses parallel computation. The field of parallel computation is going through a period of unrest. A growing rift between theory and practice suggests that more realistic models of computation are needed. While most theoretical computer science is busy designing algorithms on, and exploring the properties of, the PRAM model of computation, machine designers are faced with constraints which have no counterpart in the theoretical models of computer science. Informally, a Spatial Machine is a finite set of finite processors moving and exchanging messages in 3D unbounded empty space. The Spatial Machine is equivalent in computational power to a Turing Machine and a restriction of the model in which processors are not allowed to move freely in space is equivalent to a Finite Automaton. Sequential (non) stationary Spatial Machines are equivalent to their parallel counterparts and demonstrate the power of processor mobility and the value of empty space (vacuum) as a computational resource.

The article discusses Decision Analysis (DA). A classic decision analysis (DA) problem is to select an optimal decision or sequence of decisions when faced with a numerable set of decision alternatives, a finite number of states of nature (scenarios) that may unfold in the future, and payoffs for each combination of these states of nature and decision alternatives. Three modes in decision analysis, table, formula and tree can be used to address this problem (for example [2, 18, 28]). These modes are equivalent, yielding the same solution for a given problem. To the user, the three modes may seem quite different, resulting in different degrees of difficulty to solve the problem, requiring different mathematical skills, and providing different insights into the input data, desired outputs and goals, and the solution. Nevertheless, most textbooks emphasize a single method for solving such problems, often ignoring or paying lip service to alternative methods (for example, in [18] the three modes are presented, each in a very incomplete way; in [2, 28] only the table method is presented completely, and tree method partially).

The article discusses understanding of engineering drawings. A machine drawing consists of a description of one or more principal orthogonal views (projections) of an object. These projections can be used to reconstruct the 3D structure of the object in a variety of ways. The problem of solid reconstruction from engineering drawings deals with a subclass of the class of multiview line drawings, where the object world is not restricted to a finite set of prototypes. The interpretation of multiview line drawings has been investigated quite intensively [1, 7, 8, 12, 13, 14, 15, 16, 18]. For specific wireframe projections this is known as the "fleshing out projections" problem [19]. Scientists have developed an algorithm for computing the aspect graph for line drawings of polyhedral objects. There are several references for surveys of current approaches for solving the 3D object recognition problem. It presents a computational mechanism for the interpretation of line drawings by enabling a machine to reconstruct 3D object structures from their pictures drawn on a 2D plane.

The article discusses a new approach to string searching. String searching is a very important component of many problems, including text editing, bibliographic retrieval, and symbol manipulation. The string-matching problem consists of finding all occurrences of a pattern of length m in a text of length n. The problem is generalized allowing don't care symbols, the complement of a symbol, and any finite class of symbols. The main idea is to represent the state of the search as a number, and each search step does a small number of arithmetic and logical operations, provided that the numbers are large enough to represent all possible states of the search. The main properties of this class of algorithms are simplicity. The preprocessing and the search are very simple, and only bitwise logical operations, shifts and additions are used. The text does not need to be stored. All these properties indicate that this class of algorithm is suitable for hardware implementation. For these reasons, this new approach is a valuable contribution to all applications dealing with text searching.

We consider the effects of parallelizing branch-and-hound algorithms by expanding several live nodes simultaneously. It is shown that it is quite possible for a parallel branch-and-bound algorithm using n2 processors to take more time than one using n1 processors, even though n1<2. Furthermore, it is also possible to achieve speed- ups that are in excess of the ratio n2/n1. Experimental results with the 0/1-Knapsack and Traveling Salesman problems are also presented. [ABSTRACT FROM AUTHOR]

COKO III is a chess player written entirely in Fortran. On the IBM 360-65, COKO III plays a minimal chess game at the rate of .2 sec cpu time per move, with a level close to lower chess club play. A selective tree searching procedure controlled by tactical chess logistics allows a deploy merit of multiple minimal game calculations to achieve some optimal move selection. The tree searching algorithms are the heart of COKO's effectiveness, yet they are conceptually simple. in addition, an interesting phenomenon called a tree searching, catastrophe has plagued COKO's entire development just as it troubles a human player. Standard exponential growth is curbed to a large extent by the definition and trimming of the Fischer set. A clear distinction between tree pruning and selective tree searching is also made. Representation of the chess environment is described along with a strategical preanalysis procedure that maps the Lasker regions. Specific chess algorithms are described which could be used as a command structure by anyone desiring to do some chess program experimentation. A comparison is made of some mysterious actions of human players and COKO III. [ABSTRACT FROM AUTHOR]

A parallel processing algorithm for shrinking binary patterns to obtain single isolated elements, one for each pattern, is presented. This procedure may be used for counting patterns on a matrix, and a hardware implementation of the algorithm using large scale integrated technology is envisioned. The principal features of this method are the very small window employed (two-by-two elements), the parallel nature of the process, and the possibility of shrinking any pattern, regardless of the complexity of its configuration. Problems regarding merging and disconnection of patterns during the process as well as the determination of the maximum number of steps necessary to obtain a single isolated element from a pattern, are reviewed and discussed. An analogy with a neural network description, in terms of McCulloch-Pitts "neurons" is presented. [ABSTRACT FROM AUTHOR]

The article focuses on the use of variations of Monte Carlo tree search (MCTS) that may allow artificial intelligences (AI) to prevail against human opponents in the game Go. It states MCTS uses random sampling in decision making to quickly distinguish strong and weak moves in Go. It talks about the use of game-tree search that allows AIs to prevail in the games checkers and chess, but which is less viable in Go. It suggests that algorithms used in game playing have applications in other fields.