Introduction

This chapter discusses the relationship between workplace studies and requirements engineering for computer systems, focusing primarily on user requirements. Superficially, the two activities have much in common. Though an exact definition of the term ‘requirements engineering’ is still to be agreed upon, various perspectives reveal that requirements engineers must understand a domain well enough to determine requirements for computer systems intended for use in that setting. For example, Davis (1993) proposes that requirements engineering is the analysis, documentation and ongoing evolution of both user needs and the external behaviour of the system to be built; Loucopoulos and Karakostas (1995: vii) suggest that ‘requirements engineering deals with activities which attempt to understand the exact needs of the users of a software intensive system and to translate such needs into precise and unambiguous statements which will subsequently be used in the development of the system’. Determining the needs of users will necessarily involve understanding the activities certain users are performing in order to accomplish their work, most particularly as the new technology might be intended to replace, transform or support existing work practices. Indeed, further definitions have sought to make such a perspective explicit in order to focus the debate on both social and technical concerns in requirements. Thus, Goguen (1994: 1) states that ‘requirements are properties that a system should have in order to succeed in the environment in which it will be used’.

Introduction

Since the 1980s, ethnographic and cognitive studies of work have taken important steps forward. Powerful micro-level methodologies and theories, such as ethnomethodology, conversation analysis and distributed cognition, have been developed. However, a nagging question sometimes arises (e.g. Hughes et al., 1993; Rogers, 1997): what difference do these studies make in practice? In this vein, Grudin and Grinter (1995) write about ‘the ethnographers’ deep professional bias against intervention'.

After examining the problem in some detail, Rogers (1997) draws the following conclusion:

Introduction

Over the course of a number of investigations we have examined aspects of the organisation of engineering work under the rubric of ethnomethodology's programme of ‘studies of work’ (Garfinkel, 1986). Our investigations were initially undertaken with the purpose of contributing to an understanding of the design process through the close observation of design and development work under ‘industrial’ conditions, rather than, as was more usually the case with studies of design, located in contrived experimental settings (cf. Button and Sharrock, 1994, 1996, 1998; Sharrock and Button, 1997). The close observation of the day-to-day work of design in large organisational conditions could enhance the awareness of conditions and contingencies that would be part of the design process in complex arrangements of work. They could thus provide informational input to the circumstances under which tools to support design would be used and the practical requirements with which such tools would need to be articulated. These studies also have cogency for the area of Computer Supported Cooperative Work (CSCW). One of its concerns is the development of systems that will support large and complex ventures in coordinated work and the ‘industrial’ (hardware and software) design and development project is an example of a sizeable and complicated exercise in coordinated work.

Ethnomethodology's programme of work was itself developed as a corrective to the tendency of sociological studies titled as ‘studies of work’ to attend to almost everything that goes on in the workplace, except the work being done there.

Introduction

Computer system developers often speak of the ‘coupling’ of human intelligence with machine power in a single, interactive system that substantially enhances performance. But achieving this objective is not primarily a matter of deciding how to allocate functions between the machine components and the human elements, as much of the literature on human factors in expert and automated systems would have us believe. Without denying that this allocation problem has some heuristic relevance, the most important and vexing issue facing developers is how to build effective tools that take the fundamental differences between human action and machine operation into account.

Although several studies of human–machine interaction have demonstrated the significance of these differences for effective expert system design and deployment (e.g. Suchman, 1987; Hartland, 1993; Whalen, 1995a), and both cognitive science and the artificial intelligence (AI) research underpinning expert system design have been subjected to farranging criticism for their views on human action (Coulter, 1983, 1989; Winograd and Flores, 1986; H. Collins, 1990; Dreyfus, 1992; Button et al., 1995; Hutchins, 1995; Clancey, 1997), most artificial intelligence practitioners have continued to assume that machines can do, or can in principle be designed to do, what humans do. Accordingly, they remain focused on the allocation problem, and have been intrigued about the possibilities for designing expert applications that contain most, if not all, of the knowledge required to perform a task or solve a problem, with the ‘knowledgeability’ of the user confined largely to data entry and information retrieval procedures.

Introduction

The investigations which we shall present here form part of a research programme whose objective is the design of computer systems in terms of support systems for users. This research programme was initiated in 1979 (after two decisive studies, one about nurses' activity, the other about human–computer interaction and data coding) by an investigation aimed at designing a system for a data collecting and coding station (Pinsky, 1979; Pinsky and Theureau, 1982). Since then, it has been developed in various work situations: offices, hospitals and other services, control of sequential and continuous industrial processes, air and urban traffic control, agriculture and fishery. Such a design involves helping the users to understand the situation and take action themselves, including the search for information, and relieving the users as much as possible (within technical and economic limitations) of the details of data supply and action, in so far as these are unnecessary for an understanding of the situation. The computer system is thus considered as an element of the support system among others, like documentation, training, organisation and other sources of information on the situation here and now.

This design approach is an alternative to the design in terms of cognitive prosthesis which emerged at the beginning of computerisation and is still dominant (see Woods and Roth, 1988, for its criticism). A computer system designed as a cognitive prosthesis is supposed to concentrate the intelligence of experts (hence the commercial name ‘expert systems’ for the most sophisticated of these systems).

Generating technology innovation: the context

I approach workplace studies with a viewpoint shaped by the research and development needs of a technology provider. I recognise the validity and the importance of other points of view, especially what is often (dismissively) called ‘the academic’ or ‘the theoretical’. I support and enjoy the work which is done in their name. But in the end, what drives my interest is the contribution which workplace studies should make to ensuring the commercial success of the work systems that my company might bring to market. Clearly, I do believe that they can and have made such a contribution. Moreover, because for a number of reasons I think this contribution could well be vital in the future, I want to see workplace studies not only continued but also supported and managed in ways which will maximise their impact on the whole design and commercialisation process. In all that follows, then, I am arguing from this strongly positive position. I want to promote workplace studies within research and development. I want to secure their representation and voice in the design process. And of course I want to maximise the opportunity which their presence in design might give for improving both the process itself and the artefacts so designed, that is, the products we take to market. None of these, though, should be taken as a demand that all our workplace studies, let alone all studies of the workplace, should be reconstituted according to the framework I shall outline. As I say, my concern here is with studies carried out as part of a clearly defined R&D process.

It is widely recognised that new technology has had a profound impact on our working lives. It is argued that the so-called digital revolution has transformed the ways in which we communicate with each other, how we store and handle information, the ways in which we classify people and events, how we calculate products and value, and how we market goods and services. Social scientists discuss new forms of organisation, the transformation of contemporary culture and, among many other things, globalisation and emergence of the network society. Despite our attempts to remain optimistic in the face of the inevitable pursuit for more advanced and sophisticated technologies, often at the expense of employment, there is a growing recognition that complex systems may not necessarily enhance work practice, human relations or even efficiency within organisations. Poor performance, mishaps and technological failures are encouraging those within industry to begin to reconsider the advantages of technology for technology's sake, and to recognise that there is more to the design and deployment of complex systems than simply identifying new functionalities for computer systems. There is also a corresponding change within academia, a recognition that for all our discussion of technology we actually know very little as to how ordinary people in their daily working lives go about using the tools of their trades.

Over the past few years, there have been a number of research projects concerned with the ways in which new tools and technologies feature in everyday organisational conduct. These investigations have come to be known as ‘workplace studies’.

While there is no question that workplace studies play a prominent role in Computer Supported Cooperative Work, the exact nature of this role has been a subject of much reflection and debate over the years. So far, the deliberation has been inconclusive, and, moreover, in the last few years a certain sense of disillusionment and even scepticism has arisen concerning the ways in which and the extent to which such studies in fact contribute to CSCW systems design.

Plowman et al. (1995), for example, have raised the question ‘what are workplace studies for?’ To investigate this issue they undertook a survey of a large part of the workplace studies published in the area of CSCW – altogether seventy-five papers – and found what they called a ‘paucity of papers detailing specific design guidelines’ (Plowman et al., 1995: 313). While they hesitated to conclude that ‘workplace studies do not produce specific design guidelines’, they did feel confident that the observed paucity ‘can be attributed to the lack of reported research which has developed to the stage of a system prototype’ (Plowman et al., 1995: 313). Discussing these observations, Plowman et al. (1995: 321) surmised that the reason for the apparent failure to bridge the gap is ‘a big discrepancy between accounts of sociality generated by field studies and the way information can be of practical use to system developers’.

Abstract

We study the dynamics of on-line learning with time-correlated patterns. In this, we make a distinction between “small” networks and “large” networks. “Small” networks have a finite number of input units and are usually studied using tools from stochastic approximation theory in the limit of small learning parameters. “Large” networks have an extensive number of input units. A description in terms of individual weights is no longer useful and tools from statistical mechanics can be applied to compute the evolution of macroscopic order parameters. We give general derivations for both cases, but in the end focus on the effect of correlations on plateaus. Plateaus are long time spans in which the performance of the networks hardly changes. Learning in both “small” and “large” multi-layered perceptrons is often hampered by the presence of plateaus. The effect of correlations, however, appears to be quite different: they can have a huge beneficial effect in small networks, but seem to have only marginal effects in large networks.

Introduction

On-line learning with correlations

The ability to learn from examples is an essential feature in many neural network applications (Hertz et al., 1991; Haykin, 1994). Learning from examples enables the network to adapt its parameters or weights to its environment without the need for explicit knowledge of that environment. In on-line learning examples from the environment are continually presented to the network at distinct time steps. At each time step a small adjustment of the network's weights is made on the basis of the currently presented pattern. This procedure is iterated as long as the network learns.

July 1997 saw the start of a six month international research programme entitled Neural Networks and Machine Learning, hosted by the Isaac Newton Institute for Mathematical Sciences in Cambridge. During the programme many of the world's leading researchers in the field visited the Institute for periods ranging from a few weeks up to six months, and numerous younger scientists also benefited from a wide range of conferences and tutorials.

Amongst the many successful workshops which ran during the six month Newton Institute programme, the one week workshop on the theme of Online Learning in Neural Networks, organized by David Saad, was particularly notable. He succeeded in assembling an impressive list of speakers whose talks spanned essentially all of the major research issues in on-line learning. The workshop took place from 17 to 21 November, with the Newton Institute's purpose-designed building providing a superb setting.

This book resulted directly from the workshop, and comprises invited chapters written by each of the workshop speakers. It represents the first book to focus exclusively on the important topic of on-line learning in neural networks. On-line algorithms, in which training patterns are treated sequentially, and model parameters are updated after each presentation, have traditionally played a central role in many neural network models. Indeed, on-line gradient descent formed the basis of the first effective technique for training multi-layered networks through error back-progagation. It remains of great practical significance for training large networks using data sets comprising several million examples, such as those routinely used for optical character recognition.

Abstract

In this article we will examine online learning with an annealed learning rate. Annealing the learning rate is necessary if online learning is to reach its optimal solution. With a fixed learning rate, the system will approximate the best solution only up to some fluctuations. These fluctuations are proportional to the size of the fixed learning rate. It has been shown that an optimal annealing can make online learning asymptotically efficient meaning that asymptotically it learns as fast as possible. These results are until now only realized in very simple networks, like single–layer perceptrons (section 3). Even the simplest multilayer network, the soft committee machine, shows an additional symptom, which makes straightforward annealing uneffective. This is because, at the beginning of learning the committee machine is attracted by a metastable, suboptimal solution (section 4). The system stays in this metastable solution for a long time and can only leave it, if the learning rate is not too small. This delays the start of annealing considerably. Here we will show that a non–local or matrix update can prevent the system from becoming trapped in the metastable phase, allowing for annealing to start much earlier (section 5). Some remarks on the influence of the initial conditions and a possible candidate for a theoretical support are discussed in section 6. The paper ends with a summary of future tasks and a conclusion.

Introduction

One of the most attractive properties of artificial neural networks is their ability to learn from examples and to generalize the acquired knowledge to unknown data.

Abstract

An adaptive on-line algorithm extending the learning of learning idea is proposed and theoretically motivated. Relying only on gradient flow information it can be applied to learning continuous functions or distributions, even when no explicit loss function is given and the Hessian is not available. Its efficiency is demonstrated for drifting and switching non-stationary blind separation tasks of accoustic signals.

Introduction

Neural networks are powerful tools that can capture the structure in data by learning. Often the batch learning paradigm is assumed, where the learner is given all training examples simultaneously and allowed to use them as often as desired. In large practical applications batch learning is experienced to be rather infeasible and instead on-line learning is employed.

In the on-line learning scenario only one example is given at a time and then discarded after learning. So it is less memory consuming and at the same time it fits well into more natural learning, where the learner receives new information at every moment and should adapt to it, without having a large memory for storing old data. Appart from easier feasibility and data handling the most important advantage of on-line learning is its ability to adapt to changing environments, a quite common scenario in industrial applications where the data distribution changes gradually over time (e.g. due to wear and tear of the machines). If the learning machine does not detect and follow the change it is impossible to learn the data properly and large generalization errors will result.

Artificial neural networks (ANN) is a field of research aimed at using complex systems, made of simple identical non-linear parallel elements, for performing different types of tasks; for review see (Hertz et al 1990),(Bishop 1995) and (Ripley 1996). During the years neural networks have been successfully applied to perform regression, classification, control and prediction tasks in a variety of scenarios and architectures. The most popular and useful of ANN architectures is that of layered feed-for ward neural networks, in which the non-linear elements (neurons) are arranged in successive layers, and the information flows unidirectionally; this is in contrast to the other main generic architecture of recurrent networks where feed-back connections are also permitted. Layered networks with an arbitrary number of hidden units have been shown to be universal approximators (Cybenko 1989; Hornik et al 1989) for continuous maps and can therefore be used to implement any function defined in these terms.

Learning in layered neural networks refers to the modification of internal network parameters, so as to bring the map implemented by the network as close as possible to a desired map. Learning may be viewed as an optimization of the parameter set with respect to a set of training examples instancing the underlying rule. Two main training paradigms have emerged: batch learning, in which optimization is carried out with respect to the entire training set simultaneously, and on-line learning, where network parameters are updated after the presentation of each training example (which may be sampled with or without repetition).

Abstract

We review our recent investigation of on–line unsupervised learning from high–dimensional structured data. First, on–line competitive learning is studied as a method for the identification of prototype vectors from overlapping clusters of examples. Specifically, we analyse the dynamics of the well–known winner–takes–all or K–means algorithm. As a second standard learning technique, the application of Sanger's rule for principal component analysis is investigated. In both scenarios the necessary process of student specialization may be delayed significantly due to underlying symmetries.

Introduction

Methods from statistical physics have been applied to the theory of adaptive systems with great success in recent years. Perhaps the most prominent example is the analysis of feedforward neural networks which can learn from example data. The statistical mechanics approach allows to investigate typical properties of very large systems on average over the randomness contained in the data. It complements results from computational learning theory and other disciplines.

Most of the investigations concern the supervised learning of a rule. For reviews of the field see for instance (Watkin et al. 1993; Opper and Kinzel, 1996). A particularly successful line of research was initiated in (Kinzel and Rujan, 1990; Kinouchi and Caticha, 1992) and aims at the analysis of the physics of on–line learning schemes (Amari, 1967 and 1993; Hertz et al., 1991). On–line learning is attractive from a practical point of view because it uses only the latest in the sequence of examples. Obviously storage needs and computational effort are reduced in comparison with batch– or off–line learning (Hertz et al., 1991).

Abstract

This paper presents two approaches to characterize the dynamics of weight space probability density starting from a master equation. In the first, we provide a class of algorithms for which an exact evaluation of the integrals in the master equation is possible. This enables the time evolution of the density to be calculated at each time step without approximation. In the second, we expand earlier work on the small noise expansion to a complete perturbation framework. As an example application, we give a perturbative solution for the equilibrium density for the LMS algorithm. Finally, we use the perturbation framework to review annealed learning with schedules of the form μ(t) = μ0/tp.

Introduction

Several of the contributions to this volume apply order parameter techniques to describe the dynamics of on-line learning. In these approaches, the description assumes the form of a set of ordinary differential equations that describe the motion of macroscopic quantities, the order parameters, from which observables such as generalization error can be computed directly, without intermediate averaging. The (typically non-linear) differential equations are usually solved numerically to provide the dynamics throughout the training. Recently, this author and colleagues have obtained analytic results from the order parameter equations applied to the late time (asymptotic) convergence of annealed learning (Leen, Schottky and Saad, 1998). However, most of the application of these techniques has involved numerical integration. Indeed, the ability to track the ensemble dynamics at all times by solution of ordinary differential equations establishes an attractive analysis tool.