Technical Abstracts 1995

TR 95-01
LEARNING RULES TO PREDICT RODENT CARCINOGENICITY OF NON-GENOTOXIC COMPOUNDS
Yongwon Lee, Bruce G. Buchanan, Donald M. Mattison, Gilles Klopman and Herbert S. Rosenkranz

Abstract: The results of short-term assays (induction of chromosomal aberrations and sister-chromatid exchanges, oncogenic transformations and cellular toxicity) together with MTD (maximum tolerated dose) values and physical chemical properties of non-genotoxic (i.e. Salmonella non-mutagens) carcinogens and non-carcinogens were submitted to RL, an inductive learning program. RL was able to learn rules that correctly predicted between 70 and 80% of non-genotoxic chemicals. This is a marked improvement over current predictions using only the results of short-term assays and exceeds the predictions of human experts that used the whole spectrum of acute and subchronic toxicity results as well as human knowledge and intuition.

TR 95-02
A NETWORK-BASED FRAMEWORK TO SUPPORT STREAM SYNCHRONIZATION IN DISTRIBUTED MULTIMEDIA SYSTEMS
Taieb Znati, Robert Simon and Brian Field

Abstract: The recent advances in data processing and desktop computing, coupled with the advent of high speed communication networks, paved the way to the development of large-scale multimedia computing and communications systems. The presentation of multimedia information requires synchronization between various streams, to insure that logically related data are presented concurrently at the appropriate times. The main challenge in developing distributed multimedia systems is the development of efficient schemes to achieve temporal and intermedia synchronization. In this paper, a communication scheme to facilitate synchronization among multiple multimedia streams is presented. The basic tenet of the scheme is that by bounding the end-to-end delay on the co ordinated streams and limiting jitter, synchronization can be guaranteed. The fundamental characteristics of the proposed scheme include guaranteed support for on-time delivery of packets, jitter control, efficient network admission control mechanisms, and efficient network load based per node delay assignment.

TR 95-03
OPTIMAL PLACEMENT OF LOAD-STORE PLACEMENT OPERATIONS FOR ARRAY ACCESSES IN LOOPS
Rastislav Bodik and Rajiv Gupta

Abstract: The performance of scientific programs on superscalar processors can be significantly degraded by memory references that can frequently arise due to load and store operations associated with array references. Therefore, register allocation techniques have been developed for allocating registers to array elements whose values are repeatedly referenced over one or more loop iterations. However, straightforward techniques for the placement of load, store and register-to-register shift operations can result in the introduction of fully/partially redundant and dead operations.

In this paper we present techniques for determining the placement of loads, stores, and register-to-register shift operations at points such that placement of fully/partially redundant and dead operations is avoided. We avoid redundant and dead operations that are exposed by examining a single loop iteration as well as multiple loop iterations. The placement of loads and stores is optimal in that the number of loads and stores encountered along each path through the loop is minimal for the given branching structure and ordering of code statements. The placement of shifts minimizes the average number of shifts per loop iteration. An analysis framework is developed to efficiently solve the various data flow problems required by the above optimizations. This framework enables the analysis of the behavior of multiple congruent references in concert over multiple loop iterations.

TR 95-04
ADAPTIVE LOOP TRANSFORMATIONS FOR SCIENTIFIC PROGRAMS
Rajiv Gupta and Rastislav Bodik

Abstract: To facilitate the efficient execution of scientific programs, parallelizing compilers apply a wide range of loop transformations. In some situations it may not be possible to determine the applicability and usefulness of the transformations at compile-time. In this paper we develop adaptive versions of various loop transformations which lend themselves to efficient application at run-time. Instead of explicitly applying a transformation at compile-time, we generate adaptive code which is able to behave like the transformed code if so desired at run-time. Using run-time information a decision is made: to apply or not to apply a transformation; or to select a particular transformation from alternate transformations at run-time. Therefore an adaptive program can be viewed as expressing multiple ways of executing a program. The selection of the particular execution is based upon run-time information.

An alternative approach for exploiting run-time information generates multiversion programs by explicitly creating different versions of the program and selecting the appropriate version for execution based upon run-time information. We demonstrate that adaptive programs offer a more practical and powerful alternative to multiversion programs because adaptive programs: avoid the potentially exponential growth in code size associated with multiversion programs; realize a large fraction of the speedups achievable by the corresponding multiversion programs; and can achieve program executions that cannot be expressed using multiversion programs.

TR 95-05
INDUCTIVE POLICY: THE PRAGMATICS OF BIAS SELECTION
Foster John Provost and Bruce G. Buchanan

Abstract: This paper extends the currently accepted model of inductive bias by identifying six categories of bias and separates inductive bias from the policy for its selection (the inductive policy). We analyze existing "bias selection" systems, examining the similarities and differences in their inductive policies, and identify three techniques useful for building inductive policies. We then present a framework for representing and automatically selecting a wide variety of biases and describe experiments with an instantiation of the framework addres sing various pragmatic tradeoffs of time, space, accuracy, and the cost of errors. The experiments show that a common framework can be used to implement policies for a variety of different types of bias selection, such as parameter selection, term selection, and example selection, using similar techniques. The experiments also show that different tradeoffs can be made by the implementation of different policies, for example, from the same data different rule sets can be learned based on different tradeoffs of accuracy versus the cost of erroneous predictions.

TR 95-06
HARE: A HIERARCHICAL ALLOCATOR FOR REGISTERS IN MULTIPLE ISSUE ARCHITECTURES
David Berson, Rajiv Gupta and Mary Lou Soffa

Abstract: In this extended abstract we present HARE, a new hierarchical approach to register allocation and assignment for multiple issue load/store architectures. HARE makes extensive use of execution estimates and functional unit availability information to select values for spilling and to place the spill code in locations that minimize the increase in the overall execution time of the program. Incorporated in HARE is recursive allocation, a new hierarchical allocation technique that compares the costs of spilling registers in a code region with reducing the number of registers used in subregions by spilling additional values. HARE also makes improvements in the register allocation techniques of coalescing, rematerialization, and register copy instructions.

TR 95-07
USING MULTI-SEGMENTED VIRTUAL CHANNELS FOR RELIABLE DELIVERY OF MULTIMEDIA TRAFFIC
P.K. Chrysanthis and D. Mosse'

Abstract: With increasing demands for multimedia application by users, there is greater need to provide efficient and reliable access to distributed multimedia data. For multimedia data where multiple data streams need to be synchronized, reliability implies timeliness. Thus, reliable delivery of several data streams comprising a single datatype or used in a single application from several sources must be performed in a timely and synchronized fashion. In this paper, we propose a scheme for the establishment of multimedia delivery channels that support the timing and reliability requirements of multimedia applications while allowing for better utilization of the limited network bandwidth. The virtual, end-to-end channels from sources to destination consist of multiple segments that (a) offer alternative paths for fault tolerance, (b) support traffic splitting to cope with limited bandwidth and reduce contention per channel, (c) consider data processing and transformation requirements at specific, intermediate nodes, and (d) have a resource reservation scheme that is able to issue guarantees of timeliness and reliability.

TR 95-08
SUPPORTING SEMANTICS-BASED TRANSACTION PROCESSING IN MOBILE DATABASE APPLICATIONS
Gary D. Walborn and Panos K. Chrysanthis

Abstract: Advances in computer technologies and telecommunication methods have made mobile computing a reality. However, greater mobility implies a more tenuous network connection and a higher rate of disconnection. But disconnection need not be treated as failure. The part of a computation executing on a mobile host might continue executing while the mobile host is moving and not connected to the network. Because mobile hosts may require access to data shared by fixed and mobil hosts, it is necessary to support autonomous database operations on shared data while ensuring data consistency. In this paper, we examine the various schemes which exploit object semantics to facilitate autonomous operation and disconnected operation in mobile database applications. We define the class of fragmentable objects which may be split among a number of sites, operated upon independently at each site, and then recombined in a semantically consistent fashion. Such objects are also suited to distributed transaction processing and multidatabase systems. A number of objects with such characteristics are presented, including stacks, weak queues, and sets. Finally, an implementation of fragmentable stacks is presented and discussed.

TR 95-09
AN ESCROW METHOD TO SUPPORT DISCONNECTED MOBILE DATABASE OPERATIONS
Gary D. Walborn and Panos K. Chrysanthis

Abstract: Advances in computer technologies and telecommunication methods have made mobile computing a reality. However, greater mobility implies a more tenuous network connection and a higher rate of disconnection. But disconnection need not be treated as failure. The part of the computation executing on a mobile host might continue executing while the mobile host is moving and not connected to the network. This paper proposes a method to facilitate data management and transaction processing for disconnected mobile operations that involve access to a database shared by fixed and mobile hosts. The proposed method integrates the caching mechanisms of the CODA file system with the escrow transactional method and allows for "negotiation of terms" (reservations and deadlines for escrow items) in which both the fixed host system and the mobile system provide certain guarantees of performance.

TR 95-10
AN ICONIC QUERY PROCESSING FACILITY FOR MOBILE USERS
Antonio Massari and Panos K. Chrysanthis

Abstract: Advances in computer technologies and telecommunication methods are making mobile computing a reality. In this paper, we present a query processing facility that supports the exploration and query of databases in mobile computing environment based on the manipulation of icons. Icons are particularly suitable for mobile computing since they can be manipulated without typing. The facility requires no special knowledge of the location and the content of the underlying database nor understanding of the details of the database schema, hence it is equally useful to both unsophisticated and expert users. Its iconic interface visualizes the whole database within each object, expressing classes of objects as generalized attributes of the object, and its iconic query language involves no path specification in composing a query. The query facility provides metaquery tools that assist in the formulation of complete queries in an incremental manner using incons and without involving access to the actual data in the database. By not requiring constant access and caching of the actual data, it is able to effectively cope with the inherent limitations in memory on the mobile computer, disconnections of the mobile computer for substantial periods, restricted communication bandwidth, and high monetary cost of wireless communication.

TR 95-11
SCALING UP INDUCTIVE LEARNING WITH MASSIVE PARALLELISM
Foster J. Provost and John M. Aronis

Abstract: Machine learning programs need to scale up to very large data sets for several reasons, including increasing accuracy and discovering infrequent special cases. Current inductive learners perform well with sets of hundreds or several thousand examples. However, in some cases up to a million or more examples may be necessary to learn important special cases with confidence. These tasks are infeasible for current learning programs running on sequential machines. We discuss the need for very large datasets, Then we show how massive parallelism can be used to scale up learning methods to millions of examples by taking advantage of the inherent parallelism in rule-learning search. We describe an implementation of an inductive learning program on the CM-2 Connection Machine, show that it scales up to millions of examples, and show that it can uncover special-case rules that other learning programs would miss. The parallel version is preferable to the sequential version, for example sets larger than about 10K examples. When learning from a very large public-health database (3.5 million examples), the parallel rule-learning system uncovered a surprising relationship, which has led to considerable follow-up research and an article based on the discovered relationship to be submitted to a public health journal.

TR 95-12
EFFICIENTLY CONSTRUCTING RELATIONAL FEATURES FROM BACKGROUND KNOWLEDGE FOR INDUCTIVE MACHINE LEARNING
John M. Aronis and Foster J. Provost

Abstract: Most existing inductive learning systems form concept descriptions in propositional languages from vectors of basic features. However, many concepts are characterized by the relationships of individual examples to general domain knowledge. We describe a system that constructs relational terms efficiently to augment the description language of standard inductive systems. In our approach, examples and domain knowledge are combined into an inheritance network, and a form of spreading activation is used to find relevant relational terms. Since there is an equivalence between inheritance networks and relational databases, this yields a method for exploring tables in the database and finding relevant relationships among data to characterize concepts. We also describe the implementation of a prototype system on the CM-2 parallel computer and some experiments with large data sets.

TR 95-13
ICONIC INDEXING OF VERY LARGE SPATIAL DATABASES USING THE SIGMA-TREE
Erland Jungert and Shi-Kuo Chang

Abstract: The next generation pictorial information systems, including among others geographical information systems (GIS) and medical picture archive and communication systems (PACS), must be able to handle extremely large amount of data coming from a variety of sources, e.g. high speed sensors. The resulting databases are required to handle and maintain many Tera-bytes of data. Therefore, efficient retrieval methods that can support different types of querying and reasoning systems need to be developed. In this paper, an approach for iconic indexing is proposed, which is based on a variation of symbolic projection called the sigma-tree. Furthermore, a set of operators, called alpha-operators, which carry qualitative position-, size- and shape-related information, are introduced as the means to support qualitative spatial reasoning on such large data sets.

TR 95-14
A FRAMEWORK OF A VISUAL LANGUAGE WITH DYNAMIC SPECIFICATION
Jun'ichi Miyao and Shi-Kuo Chang

Abstract: Behavior of complex software may be changed depending on external environment such as user profile, available resources, input data, time, and so on. Design of such dynamic software in conventional languages needs extra effort to keep consistency among behavior for all states.

In this paper, we will propose an object oriented visual language for dynamic software and its visualized specification with dynamics presentation. The language supports dynamic behavior according to both time-dependent and time-independent external states based on dynamic objects. In the visualized specification, we present uniform dynamics presentation for all elements in the language based on three concepts of snapshot, overlap and sweep. The proposed language and its specification methodology will make design of dynamic software consistent and reliable.

TR 95-15
AN INFORMATION BASED BAYESIAN APPROACH TO HISTORY-TAKING
Giuseppe Carenini, Stefano Monti and Gordon Banks

Abstract: Effective history-taking systems need to dynamically reduce the number of questions to ask. This can be done either categorically or probabilistically, by exploiting previous patient's answers. In this paper, we propose a probabilistic information-based history-taking strategy that combines synergetically two information-content measures for reducing the number of questions asked. We have applied this strategy to an existing history-taking system and some preliminary results seem to confirm our initial intuitions.

TR 95-16
INDUCTION OF RULES FOR BIOLOGICAL MACROMOLECULE CRYSTALLIZATION
Daniel Hennessy, Vanathi Gopalakrishnan, Bruce G. Buchanan, Johm M. Rosenberg, and Devika Subramanian

Abstract: X-ray crystallography is the method of choice for determining the 3-D structure of large macromolecules at a high enough resolution. The rate limiting step in structure determination is the crystallization itself. It takes anywhere between a few weeks to several years to obtain macromolecular crystals that yield good diffraction patterns. The theory of forces that promote and maintain crystal growth is preliminary, and crystallographers systematically search a large parameter space of experimental settings to grow good crystals. There is a wealth of experimental data on crystal growth most of which is in paper laboratory notebooks. Some of the data has been gathered in electronic form, e.g., the Biological Macromolecular Crystallization Database (BMCD) which is a repository of successful experimental conditions for growing over 800 different macromolecules (Gilliland 1987). Crystallographers are in need of computational tools to gather and analyze past data to design new crystal growth trails. We are building the Crystallographer's Assistant (CA) to help crystallographers record and maintain experimental context in electronic form, offer suggestions on experimental conditions that are likely to be successful, and provide explanations for failed experiments. As an initial step in this project, we have applied RL, an inductive learning program, to the BMCD. In this paper, we report initial experiments and findings in applying RL to the BMCD. From the point of view of crystallography, we have discovered possibly significant new empirical relationships in crystal growth. From the point of view of machine learning, our work suggests refinements of existing methods for incorporating detailed domain knowledge into inductive analysis techniques.

TR 95-17
AN INTELLIGENT SYSTEM FOR COMPARING PROTEIN STRUCTURES
Ethan Benatan

Abstract: An approach to protein structure comparison is presented which uses techniques of artificial intelligence (AI) to generate a mapping between two protein structures. The approach proceeds by first identifying the seed of a possible mapping, and then searching for ways to extend the seed by incorporating corresponding elements from the two proteins. Correspondence is judged using heuristic functions which assess the similarity of the structural environments of the elements. The search can be guided by separately encoded knowledge. A prototype has been implemented which is able to rapidly create mappings with a high degree of accuracy in test cases.

TR 95-18
EXPLORING ALTERNATIVE BIASES PRIOR TO LEARNING IN SCIENTIFIC DOMAIN
Bruce G. Buchanan and Yongwon Lee

Abstract: Before machine learning can be applied to a new scientific domain, considerable attention must be given to finding appropriate ways to characterize the learning problem. A central idea guiding our work is that we must make explicit more of the elements of a program's bias and understand the criteria by which we prefer one bias over another. We illustrate this exploration with a problem to which we have applied the RL induction program, the problem of predicting whether or not a chemical is a likely carcinogen.

TR 95-19
BUILDING A MEDICAL MUTIMEDIA DATABASE SYSTEM TO INTEGRATE CLINICAL INFORMATION: AN APPLICATION OF HIGH-PERFORMANCE COMPUTING AND COMMUNICATIONS TECHNOLOGY
Henry J. Lowe, Bruce G. Buchanan, Gregory F. Cooper and John K. Vries

Abstract: The rapid growth of diagnostic-imaging technologies over the past two decades has dramatically increased the amount of nontextual data generated in clinical medicine. The architecture of traditional, text-oriented, clinical information systems has made the integration of digitized clinical images with the patient record problematic. Systems for the classification, retrieval, and integration of clinical images are in their infancy. Recent advances in high-performance computing, imaging, and networking technology now make it technologically and economically feasible to develop an integrated, multimedia, electronic patient record. As part of The National Library of Medicine's Biomedical Applications of High-Performance Computing and Communications program, we plan to develop Image Engine, a prototype microcomputer-based system for the storage, retrieval, integration, and sharing of a wide range of clinically important digital images. Images stored in the Image Engine database will be indexed and organized using the Unified Medical Language System Metathesaurus and will be dynamically linked to data in a text-based, clinical information system. We will evaluate Image Engine by initially implementing it in three clinical domains (oncology, gastroenterology, and clinical pathology) at the University of Pittsburgh Medical Center.

TR 95-20
VQLM: A VISUAL QUERY LANGUAGE FOR MACROMOLECULAR STRUCTURAL DATABASES
Dawn Cohen, Kumar Vadaparty and Bill Dickinson

Abstract: Databases of macromolecular structures allow researchers to identify general principles of molecular behavior. They do this by providing a variety of data obtained under a number of different experimental conditions. Many new tools have been developed recently to aid in exploratory analysis of structural data. However, some queries of interest still require considerable manual filtering of data. In particular, studies attempting to make generalizations about complex arrangements of atoms or building blocks in macromolecular structures cannot be approached directly with existing tools. Such studies are frequently carried out on only a few structures or else require a labor-intensive process. To address this problem, we have developed a visual language. VQLM (Visual Query Language for Macromolecules). A query is formulated in this language by drawing an abstract picture of substructures to be searched for in the database and specifying constraints on the objects in them. To illustrate the usefulness of our language, we show how to encode a number of queries that were found scientifically interesting in the published literature in molecular biology. VQLM relies on VQL, a new database language, as its underlying engine for database retrieval and computation. We believe that VQLM will make macromolecular structural data more accessible to scientists, enabling faster and deeper data analysis.

TR 95-21
SIMD HYBERCUBE ALGORITHMS FOR COMPLETE EUCLIDEAN DISTANCE TRANSFORM
Henry Y.H. Chuang and Ling Chen

TR 95-23
A SYSTEMATIC DESIGN OF SYSTOLIC ARRAYS FROM AFFINE RECURRENCE EQUATIONS
Henry Y. H. Chuang and Ling Chen

Abstract: We present a method for synthesizing systolic arrays with non-uniform data flow from problem specifications expressed as affine recurrence equations (ARE). The method is systematic and applicable to a large class of AREs. It first uniformizes the recurrence equation into a piecewise uniform recurrence equation (PURE) where dependency is uniform in each segment, which is then mapped into a systolic array. We also show how to design control signals to govern the non-uniform data flow. Besides the design method, we present theoretical results concerning transformability of an ARE on a given domain into a PURE which has a systolic array realization. Techniques useful to obtain PUREs which result in simpler and more efficient systolic arrays are also given.

TR 95-24
THE ONLINE TRANSPORTATION PROBLEM
Bala Kalyanasundaram, and Kirk Pruhs

Abstract: We study the online transportation problem. We show that the halfopt-competitive ratio for the greedy algorithm is $\Theta( {\rm min}(m, \log C))$, where $m$ is the number of server sites, and $C$ is the total number of servers. We also present an algorithm {\sc Balance} that is a modification of the greedy algorithm and that has a halfopt-competitive ratio of $O(1)$.

TR 95-25
WHAT CONTINGENCIES TO PLAN FOR
Nilufer Onder

Abstract: Classical planning systems make two major assumptions about the world the agent inhabits. First, the agent is assumed to have perfect knowledge about both the current state of the world and the effect of all its actions. Second, the world is assumed to be static. Recent work on conditional planning aims at relaxing these assumptions. A plan constructed contains several branches, each corresponding to a possible outcome of an operation with unknown outcomes. On the other hand, existing conditional planners assume that the planner should plan for every contingency. In real life situations, many of the contingent events are not worth spending plan time on. The motivation of this work is to hypothesize, experiment with, and discuss the factors that affect the decision about which contingencies to plan for, and to construct a framework that allows for general-purpose planning that combines conditional planning with replanning.

TR 95-28, ISL-95-12
GENERATING PATIENT-SPECIFIC INTERACTIVE NATURAL LANGUAGE EXPLANATION
Giuseppe Carenini, Vibhu O. Mittal, and Johanna D. Moore

Abstract: Patient compliance is a significant problem and is strongly correlated with the patients' understanding of their condition and prescribed treatment. Since doctors typically do not have large amounts of time to educate patients, and impersonal, voluminous patient handouts are largely ineffective, we propose the use of a sophisticated computer-based information system to generate tailored, interactive handouts to communicate with patients. Our system uses text planning and user modeling techniques to generate natural language descriptions of migraine, its symptoms, triggering factors and prescriptions. The system is capable of handling follow-up questions requesting further information, and generating responses in the context of previously supplied information -- a capability unavailable in previous patient information systems. The system tailors its interaction to: (i) the class of migraine patients, (ii) the individual patient, and (iii) the previous dialogue. Preliminary evaluation of the system indicates that patients find it useful and informative. More extensive evaluation is in progress.

TR 95-29
DYNAMIC GENERATION OF FOLLOW UP QUESTION MENUS: FACILITATING INTERACTIVE NATURAL LANGUAGE DIALOGUES
Vibhu O. Mittal and Johanna D. Moore

Abstract: Most complex systems provide some form of help facilities. However, typically, such help facilities do not allow users to ask follow up questions or request further elaborations when they are not satisfied with the systems' initial offering. One approach to alleviating this problem is to present the user with a menu of possible follow up questions at every point. Limiting follow up information requests to choices in a menu has many advantages, but there are also a number of issues that must be dealt with in designing such a system. To dynamically generate useful embedded menus, the system must be able to, among other things, determine the context of the request, represent and reason about the explanations presented to the user, and limit the number of choices presented in the menu. This paper discusses such issues in the context of a patient education system that generates a natural language description in which the text is directly manipulable - clicking on portions of the text causes the system to generate menus that can be used to request elaborations and further information.

TR 95-32
DEALING WITH INCOMPATIBLE PRESUMPTIONS OF COMMIT PROTOCOLS IN MULTIDATABASE SYSTEMS
Yousef J. Al-Houmaily and Panos K. Chrysanthis

Abstract: A multidatabase system (MDBS) is a software system that is built on top of multiple pre-existing and heterogeneous database systems (LDBSs) to facilitate their interoperation. This paper discusses the issue of compatibility among atomic commit protocols (ACPs) in a MDBS environment. Specifically, it shows that supporting a visible prepare-to-commit state is not enough for a successful integration of ACPs in an operational fashion because the outcome of some transactions might have to be remembered forever. Therefore, we define an operational correctness criterion that allows terminated transactions to be forgotten and propose presumed any (PrAny), a two-phase commit (2PC) protocol variant that successfully integrates presumed abort (PrA) and presumed commit (PrC) in a MDBS. We also show how a MDBS can change the behavior of a LDBS that employs PrA to look as if it is employing PrC and vice versa, allowing the MDBS to dynamically adapt to the most appropriate 2PC variant at any given time.

TR 95-33
LEVERAGING A FELLOWSHIP IN MEDICAL INFORMATICS: FOCUS ON THE SOFTWARE
Bruce G. Buchanan

TR 95-34
UNDERSTANDING AND ASSISTING CAD USERS IN THE REAL WORLD
Suresh K. Bhavnani, Ulrich Flemming, Diana E. Forsythe, James H. Garrett, Jr., and Doris S. Shaw

Abstract: In spite of the rapid increase in functionality and resources provided by CAD systems, productivity growth expected from their use has been difficult to achieve. Although many surveys describe this "productivity puzzle", few studies have been conducted on actual CAD users to understand its causes. In an effort to understand CAD users, the primary author conducted a site visit of a federal architectural office and observed CAD users in their natural setting. This paper describes some preliminary results obtained from using ethnographic techniques developed by cultural anthropologists. The study revealed users who had leveled-off in their learning and experimentation and were using the CAD system in sub-optimal ways. The authors argue that this sub-optimal usage occurs because users have limited ways to learn with ease better or different ways of executing tasks. To encourage users to make better use of the system, the authors propose that CAD systems should provide active assistance, that is, intervene spontaneously with advice, assistance, and relevant information while the user interacts with the system. They conclude with some clues and difficulties pointed out by the study that need to be considered when developing such active assistance.

TR 95-36
ABSTRACTS OF POSTERS PRESENTED AT THE 1995 NORTH AMERICAN CONGRESS OF CLINICAL TOXICOLOGY ANNUAL MEETING
E. P. Krenzelok, T.D. Jacobsen and J. M. Aronis

TR 95-37
ABSTRACTS OF TALKS PRESENTED AT THE EUROPEAN ASSOCIATION OF POISONS CENTRES AND CLINICAL TOXICOLOGISTS SCIENTIFIC MEETING--OCCUPATIONAL AND ENVIRONMENTAL TOXICOLOGY
E. P. Krenzelok, F. J. Provost, T. D. Jacobsen, J. M. Aronis and B. G. Buchanan