Bruce A. MacDonald

A framework for knowledge acquisition through techniques of concept learning

An integrative framework is developed for describing concept learning techniques that makes it possible to evaluate their relevance to knowledge engineering. The framework provides a general basis for relating concept learning to knowledge acquisition and is a starting point for the development of formal design rules. First, concept learning is framed in the context of knowledge acquisition. Then the general forms of input and concept representation such as logic, functions and procedures are discussed. Next, methods of biasing the search for a suitable concept are described and illustrated including: background knowledge, conceptual bias, composition bias, and preference orderings. Finally, modes of teacher interaction are reviewed, including the nature of examples given and the method of presenting them. The framework is illustrated by applying it to the better-documented concept learning systems. >

Programming by example: The human face of AI

It is argued that “human-centredness” will be an important characteristic of systems that learn tasks from human users, as the difficulties in inductive inference rule out learning without human assistance. The aim of “programming by example” is to create systems that learn how to perform tasks from their human users by being shown examples of what is to be done. Just as the user creates a learning environment for the system, so the system provides a teaching opportunity for the user, and emphasis is placed as much on facilitating successful teaching as on incorporating techniques of machine learning. If systems can “learn” repetitive tasks, their users will have the power to decide for themselves which parts of their jobs should be automated, and teach the system how to do them — reducing their dependence on intermediaries such as system designers and programmers.This paper presents principles for programming by example derived from experience in creating four prototype learners: for technical drawing, text editing, office tasks, and robot assembly. A teaching metaphor (a) enables the user to demonstrate a task by performing it manually, (b) helps to explain the learners limited capabilities in terms of a persona, and (c) allows users to attribute intentionality. Tasks are represented procedurally, and augmented with constraints. Suitable mechanisms for attention focusing are necessary in order to control inductive search. Hidden features of a task should be made explicit so that the learner need not embark on the huge search entailed by hypothesizing missing steps.

Recent laser radar field-test results gathered with the rapid optical beam steering (ROBS) system

We present a summary of recent imagery and tracking measurements made with the ROBS laser radar system operating at White Sands Missile Range. The ROBS instrument utilizes a 0.5 meter aperture optical system, 3 - 5 micron imaging cameras, and a coherent CO2 laser radar for range and Doppler measurements. The optical system is based on a roving fovea design, which enables signal target tracking over large angles at high track update rate, and rapid retargeting between multiple targets. Results of several field measurements are presented.

Adaptive robot training by programming and guiding

The simplest and most common robot training method is operator-guiding, but this method is inflexible and limited to fixed sequences. Explicit textual robot programming has enjoyed many advances in the last decade and current research implementations are powerful. But explicit programming is complex and programmers must be specially trained. Mixed systems attempt to unite the facility of guiding with the power of programming. However, existing mixed systems exhibit mismatches in the interaction between programming and guiding, because they ignore underlying dependencies between the programmer and operator. An operator may (1) be given insufficient visual cues; (2) lack the required dexterity; or (3) be required to add unforeseen movement sequences to the program. This paper presents the design and implementation of ART (adaptive robot trainer), a prototypical mixed robot training system that eliminates or corrects deficiencies in visual cues and dexterity, and additionally improves the guiding and programming components. Mixed systems and assembly tasks are analysed, to give an effective representation of task state, which in turn motivates the design of ARTs language to automate much of the program-guiding interaction. The language allows the programmer to express assembly operations and object-feature relationships in a natural way while providing the system with the information necessary to maintain the task state. The representation also enables the correction of guiding errors, flexibility in the guiding protocol and the generation of meaningful messages to prompt operator actions. These principles in the design and implementation pave the road to more instructable, capable robots.

Missile and aircraft field test data acquired with the rapid optical beam steering (ROBS) sensor system

The ROBS instrument has recently acquired unique imagery of a missile intercepting an airborne drone target. We present a summary of that mission. We also present imagery of three airborne targets collected while the ROBS instrument simultaneously tracked all three aircraft. The recent test data highlights the capability of the ROBS instrument for autonomous acquisition, tracking, and imaging of multiple targets under field test conditions. We also describe improvements to the optical system currently underway.

Specifying Procedures to Office Systems

This paper surveys current practice, research, and future prospects for communicating procedures to office computer systems, placing special emphasis on robustness and suitability for the casual user. Users of existing systems who have to specify procedures must generally resort to some kind of command language. Explicit forms programming languages, perhaps based on ideas of logic programming which suppress control structure, offer better prospects. In the more distant future, knowledge-based techniques utilizing models of office semantics may address the more general issues of problem-solving in the office. Programming by example is a promising method for specifying procedures but presents difficulties with editing, conditionals, iteration, variables and data structures. These can be partially overcome by using several example sequences or having users provide control information explicitly through a well-engineered interactive interface.

Task learning from instruction: an application of discourse processing techniques to machine learning

A good teacher will provide crucial information about a new task, rather than simply performing examples with no elaboration. Machine learning paradigms have ignored this form of instruction, concentrating on induction over multiple examples, or knowledge-based generalization. This paper presents a model of supervised task learning designed to exploit communicative acts. Instruction is viewed as planned explanation, and plan recognition is applied to the problem at both domain and discourse levels, and extended to allow the learner to have incomplete knowledge. The model includes a domain level plan recognizer and a discourse level plan recognizer that cues a third level of plan structure rewriting rules. The rewriter may add new domain operator schemata. Details are given of an example in which a robot apprentice is instructed in the building of arches.

The Other Side of the Coin: Teaching Artificial Learning Systems

The burgeoning technology of machine learning is beginning to provide some insight into the nature of learning and the role of teaching in expediting the learning process. A number of systems that learn concepts and procedures from examples have been described in the research literature. In general these require a teacher who not only has an analytical understanding of the problem domain, but also is familiar with some of the internal workings of the learning system itself. This is because the learner is performing a search in concept space which is generally quite intractable, but for the teachers selection of guiding examples.