OXFORD 1994 Oxford University Press, Walton Street, Oxford 0X2 6DP Oxford New York Athens Auckland Bangkok Bombay Calcutta Cape Town Dar es Salaam Delhi Florence Hong Kong Istanbul Karachi Kuala Lumpur Madras Madrid Melbourne Mexico City Nairobi Paris Singapore Taipei Tokyo Toronto and associated companies in Berlin lbadan Published in the United States by Oxford University Press Inc., New York 0 E. K. Furukawa, D. Michie, and S. Muggleton, 1994 All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, without the prior permission in writing of Oxford University Press. Enquiries concerning reproduction outside those terms and in other countries should be sent to the Rights Department, Oxford University Press, at the address above. This book is sold subject to the condition that it shall not, by way of trade or otherwise, be lent, re-sold, hired out, or otherwise circulated without the publisher's prior consent in any form of binding or cover other than that in which it is published and without a similar condition including this condition being imposed on the subsequent purchaser. The founder of modern computational logic, J.A. Robinson, opens this volume with a chapter on the field's great forefathers John von Neumann and Alan Turing.

When you lose a game of chess to a computer then don't pretend that you didn't think at the game.

Machine Intelligence 1 (1967) (eds N. Collins and D. Michie) Oliver & Boyd, Edinburgh Machine Intelligence 2 (1968) (eds E. Dale and D. Michie) Oliver & Boyd, Edinburgh (1 and 2 published as one volume in 1971 by Edinburgh University Press) (eds N. Collins, E. Dale, and D. Michie) Machine Intelligence 3 (1968) (ed. CLARENDON PRESS - OXFORD 1991 Oxford University Press, Walton Street, Oxford 0X2 6DP Oxford New York Toronto Delhi Bombay Calcutta Madras Karachi Petaling Jaya Singapore Hong Kong Tokyo Nairobi Dar es Salaam Cape Town Melbourne Auckland and associated companies in Berlin lbadan Oxford is a trade mark of Oxford University Press Published in the United States by Oxford University Press, New York C J. E. Hayes, D. Michie, and E. Tyugu, 1991 All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, without the prior permission of Oxford University Press British Library Cataloguing in Publication Data Machine intelligence. ISBN 0-19-853823-5 Library of Congress Cataloging in Publication Data Machine intelligence 12: towards an automated logic of human thought /edited by J. E. Hayes, D. Michie, and It is a pleasure to contribute an introduction to this twelfth volume of the international Machine Intelligence series. My own work has, at times, cast me in the scientific roles of experimenter, instrumentation designer, and administrator.

They produce one set of rules from one set of data and have no memory which permits them to add to a knowledge base by further learning. Incremental learning systems remember the concepts which they have learned and can use them for further learning and problem solving. Some examples are, CONFUCIUS (Cohen 1978) and Marvin (Sammut 1981). These programs build a model of their task environment through successive learning experiences which require interaction with the environment. The task that we consider in this paper involves a program learning to control an agent in a reactive environment. This is an environment where changes occur in response to actions. Agents other than the learner may be present. As an agent accumulates experience, it constructs a world model or theory of behaviour which can be used to predict the outcome f Present address: Department of Computer Science, University of New South Wales, Sydney, Australia.

Machine Intelligence 1 (1967) (eds N. Collins and D. Michie) Oliver & Boyd, Edinburgh Machine Intelligence 2 (1968) (eds E. Dale and D. Michie) Oliver & Boyd, Edinburgh (1 and 2 published as one volume in 1971 by Edinburgh University Press) (eds N. Collins, E. Dale, and D. Michie). CLARENDON PRESS OXFORD 1988 Oxford University Press, Walton Street, Oxford 0X2 6DP Oxford New York Toronto Delhi Bombay Calcutta Madras Karachi Petaling Jaya Singapore Hong Kong Tokyo Nairobi Dar es Salaam Cape Town Melbourne Auckland and associated companies in Berlin lbadan Oxford is a trade mark of Oxford University Press Published in the United States by Oxford University Press, New York J. E. Hayes, D. Michie, and J. Richards 1988 All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, without the prior permission of Oxford University Press British Library Cataloguing in Publication Data Machine Intelligence. Richard J. 006.3 ISBN 0-19-853718-2 Library of Congress Cataloging in Publication Data Data available Typeset and printed in Northern Ireland at The Universities Press (Belfast) Ltd. Held at intervals in Scotland, the first seven International Machine Intelligence Workshops spanning the period of 1965-71 were involved in developing the new subject internationally--in those early days mainly as a mid-Atlantic phenomenon.

Reprinted with permission from Analytical Chemistry, August 1976, Vol.48, No.9, pp.1368-1375.

In this paper we discuss some reasons why temporal logic might not be suitable to model real life norms. To show this, we present a novel deontic logic contrary-to-duty/derived permission paradox based on the interaction of obligations, permissions and contrary-to-duty obligations. The paradox is inspired by real life norms.

Quantum machine learning has received significant attention in recent years, and promising progress has been made in the development of quantum algorithms to speed up traditional machine learning tasks. In this work, however, we focus on investigating the information-theoretic upper bounds of sample complexity - how many training samples are sufficient to predict the future behaviour of an unknown target function. This kind of problem is, arguably, one of the most fundamental problems in statistical learning theory and the bounds for practical settings can be completely characterised by a simple measure of complexity. Our main result in the paper is that, for learning an unknown quantum measurement, the upper bound, given by the fat-shattering dimension, is linearly proportional to the dimension of the underlying Hilbert space. Learning an unknown quantum state becomes a dual problem to ours, and as a byproduct, we can recover Aaronson's famous result [Proc. R. Soc. A 463:3089-3144 (2007)] solely using a classical machine learning technique. In addition, other famous complexity measures like covering numbers and Rademacher complexities are derived explicitly. We are able to connect measures of sample complexity with various areas in quantum information science, e.g. quantum state/measurement tomography, quantum state discrimination and quantum random access codes, which may be of independent interest. Lastly, with the assistance of general Bloch-sphere representation, we show that learning quantum measurements/states can be mathematically formulated as a neural network. Consequently, classical ML algorithms can be applied to efficiently accomplish the two quantum learning tasks.

Most standard algorithms for prediction with expert advice depend on a parameter called the learning rate. This learning rate needs to be large enough to fit the data well, but small enough to prevent overfitting. For the exponential weights algorithm, a sequence of prior work has established theoretical guarantees for higher and higher data-dependent tunings of the learning rate, which allow for increasingly aggressive learning. But in practice such theoretical tunings often still perform worse (as measured by their regret) than ad hoc tuning with an even higher learning rate. To close the gap between theory and practice we introduce an approach to learn the learning rate. Up to a factor that is at most (poly)logarithmic in the number of experts and the inverse of the learning rate, our method performs as well as if we would know the empirically best learning rate from a large range that includes both conservative small values and values that are much higher than those for which formal guarantees were previously available. Our method employs a grid of learning rates, yet runs in linear time regardless of the size of the grid.

In this work we introduce a new fast incremental gradient method SAGA, in the spirit of SAG, SDCA, MISO and SVRG. SAGA improves on the theory behind SAG and SVRG, with better theoretical convergence rates, and support for composite objectives where a proximal operator is used on the regulariser. Unlike SDCA, SAGA supports non-strongly convex problems directly, and is adaptive to any inherent strong convexity of the problem. We give experimental results showing the effectiveness of our method.