I think this machine -may be of interest in view of its connection ...:h the problems of trial-and-error learning, forgetting and feedback _ ems. The maze can be changed _ any desired mantler by rearranging the partitions between the twen --:five squares. In the maze there is a sensing finger, which can feel the -.:titions of the maze as it comes against them. This finger is moved .- The goal is mounted on a pin which can be slipped into a jack _ any of the twenty-five squares.

It is no pure coincidence that an organic chemist rather than a computer specialist is writing this foreword.

Harold Shephard Samuel D. Thurman William T. Lake JOINDER OF CLAIMS, COUNTERCLAIMS, AND CROSS-COMPLAINTS: SUGGESTED REVISION OF THE CALIFORNIA PROVISIONS. Research in artificial intelligence, a branch of computer science, has illuminated our capacity to use computers to model human thought processes. In this Article we will argue that the time has come for serious interdisciplinary work between lawyers and computer scientists to explore the computer's potential in law. Interdisciplinary work between the lawyer and the computer scientist has floundered on the misconceptions that each has of the other's discipline. As a result, no one has yet attempted computer programs incorporating complex techniques of legal reasoning. Even efforts in legal information retrieval have been hampered by these misconceptions. In retrieval, lawyers have viewed the computer as, at most, a storehouse from which cases and statutes might be retrieved by skillfully designed indexing systems.

The nature of the business doesn't matter; in every business computers have made numerous changes in record keeping, process control, and decision-making. And there will be more. One of the most important trends in computing is making computers behave intelligently. The software underneath this intelligent behavior is called an expert system, sometimes also called a knowledgebased system, or knowledge system. An expert system is a computer program that reasons about a problem in much the same way, and with about the same performance, as specialists. This chapter is about the trend toward using expert systems: what it means, how it's possible, and how to think about it. There have been lead articles about this in Fortune, Business Week, and Newsweek; most Fortune-SOO companies are using expert systems; many are establishing research and development groups for them; even staid IBM is marketing expert systems tools and using them internally. Bruce G. Buchanan I 129 There are many reasons why companies want to build an expert system. Most of them are based on the premise that: Expertise is a scarce resource. And the corollary (by Murphy's Law): Even when there is enough expertise, it is never close enough to the person who needs it in a hurry. Because this is true, almost by definition, an expert system containing some of the knowledge of a company's specialists may have several benefits.. There are several examples of expert systems working in various problem areas. At present, they are used more as "intelligent assistants" than as replacements for technicians or experts. That is, they help people think through difficult problems and may provide suggestions about what to do, without taking over every aspect of the task. Although the problems are quite different they can be categorized into two major classes problems of interpretation and problems of construction. Interpretive problem examples include Schlumberger's Dipmeter Advisor, which replicates the expertise of some of their company-wide specialists who interpret data from clients' oil wells and then sell the expert system's interpretations around the world.

The aim of this note is to discuss the reasons why temporal logic, specifically Linear Temporal Logic [7] might not be suitable to check whether the specifications of a system comply with a set of normative requirements. The debate whether it is possible to use temporal logic for the representation of norms is not a novel one (see for example [9]), and while the argument had settled for a while, the past decade saw a resurgence of the topic with many works in the fields of normative multi-agents and business process compliance advocating temporal logic as the formalism to express normative constraints on agent behaviours and process executions.

Recent literature on online learning has focused on developing adaptive algorithms that take advantage of a regularity of the sequence of observations, yet retain worst-case performance guarantees. A complementary direction is to develop prediction methods that perform well against complex benchmarks. In this paper, we address these two directions together. We present a fully adaptive method that competes with dynamic benchmarks in which regret guarantee scales with regularity of the sequence of cost functions and comparators. Notably, the regret bound adapts to the smaller complexity measure in the problem environment. Finally, we apply our results to drifting zero-sum, two-player games where both players achieve no regret guarantees against best sequences of actions in hindsight.

We propose a topic modeling approach to the prediction of preferences in pairwise comparisons. We develop a new generative model for pairwise comparisons that accounts for multiple shared latent rankings that are prevalent in a population of users. This new model also captures inconsistent user behavior in a natural way. We show how the estimation of latent rankings in the new generative model can be formally reduced to the estimation of topics in a statistically equivalent topic modeling problem. We leverage recent advances in the topic modeling literature to develop an algorithm that can learn shared latent rankings with provable consistency as well as sample and computational complexity guarantees. We demonstrate that the new approach is empirically competitive with the current state-of-the-art approaches in predicting preferences on some semi-synthetic and real world datasets.

Kernel ridge regression (KRR) is a standard method for performing non-parametric regression over reproducing kernel Hilbert spaces. Given $n$ samples, the time and space complexity of computing the KRR estimate scale as $\mathcal{O}(n^3)$ and $\mathcal{O}(n^2)$ respectively, and so is prohibitive in many cases. We propose approximations of KRR based on $m$-dimensional randomized sketches of the kernel matrix, and study how small the projection dimension $m$ can be chosen while still preserving minimax optimality of the approximate KRR estimate. For various classes of randomized sketches, including those based on Gaussian and randomized Hadamard matrices, we prove that it suffices to choose the sketch dimension $m$ proportional to the statistical dimension (modulo logarithmic factors). Thus, we obtain fast and minimax optimal approximations to the KRR estimate for non-parametric regression.

This paper introduces and develops a novel variable importance score function in the context of ensemble learning and demonstrates its appeal both theoretically and empirically. Our proposed score function is simple and more straightforward than its counterpart proposed in the context of random forest, and by avoiding permutations, it is by design computationally more efficient than the random forest variable importance function. Just like the random forest variable importance function, our score handles both regression and classification seamlessly. One of the distinct advantage of our proposed score is the fact that it offers a natural cut off at zero, with all the positive scores indicating importance and significance, while the negative scores are deemed indications of insignificance. An extra advantage of our proposed score lies in the fact it works very well beyond ensemble of trees and can seamlessly be used with any base learners in the random subspace learning context. Our examples, both simulated and real, demonstrate that our proposed score does compete mostly favorably with the random forest score.