A Hybrid Architecture for Situated Learning of Reactive Sequential Decision Making

In developing autonomous agents, one usually emphasizes only (situated) procedural knowledge, ignoring more explicit declarative knowledge. On the other hand, in developing symbolic reasoning models, one usually emphasizes only declarative knowledge, ignoring procedural knowledge. In contrast, we have developed a learning model CLARION, which is a hybrid connectionist model consisting of both localist and distributed representations, based on the two-level approach proposed in [40]. CLARION learns and utilizes both procedural and declarative knowledge, tapping into the synergy of the two types of processes, and enables an agent to learn in situated contexts and generalize resulting knowledge to different scenarios. It unifies connectionist, reinforcement, and symbolic learning in a synergistic way, to perform on-line, bottom-up learning. This summary paper presents one version of the architecture and some results of the experiments.     

Information problem‐solving skills in small virtual groups and learning outcomes

This study investigated the frequency of use of information problem‐solving (IPS) skills and its relationship with learning outcomes. During the course of the study, 40 teachers carried out a collaborative IPS task in small virtual groups in a 4‐week online training course. The status of IPS skills was collected through self‐reports handed in over the course of the 4 weeks. Learning was evaluated by means of open‐ended questionnaires before and after the group task. Three types of knowledge learning were evaluated: declarative, procedural and situational. Teachers exhibited a recurrent use of all skills during the whole collaborative task, although periodic use differed from week to week. Results showed a relationship between some IPS skills and declarative and procedural knowledge. The skills that were statistically significant were share information, read peer's information and analyse information. Implications for learning support and instruction are discussed.     

An Embedded Microprocessor for Intelligent Control

The conventional approach for the implementation of the knowledge base of a planning agent, on an intelligent embedded system, is solely of software nature. It requires the existence of a compiler that transforms the initial declarative logic program, specifying the knowledge base, to its equivalent procedural one, to be programmed to the embedded systems microprocessor. This practice increases the complexity of the final implementation (the declarative to sequential transformation adds a great amount of software code for simulating the declarative execution) and reduces the overall systems performance (logic derivations require the use of a stack and a great number of jump instructions for their evaluation). The design of specialized hardware implementations, which are only capable of supporting logic programs, in an effort to resolve the aforementioned problems, introduces limitations in their use in applications where logic programs need to be intertwined with traditional procedural ones in a desired application. In this paper, we exploit HW/SW codesign methods to present a microprocessor, capable of supporting hybrid applications using both programming approaches. We take advantage of the close relationship between attribute grammar (AG) evaluation and knowledge engineering methods to present a programmable hardware parser that performs logic derivations and combine it with an extension of a conventional RISC microprocessor that performs the unification process to report the success or failure of logic derivations. The extended RISC microprocessor is still capable of executing conventional procedural programs, thus hybrid applications can be implemented. The presented implementation increases the performance of logic derivations for the control inference process (experimental analysis yields an approximate 1000% – 10 times increase in performance) and reduces the complexity of the final implemented code through the introduction of an extended C language called C-AG that simplifies the programming of hybrid procedural-declarative applications.     

Autonomous learning of sequential tasks: experiments and analyses

Presents a learning model CLARION, which is a hybrid model based on the two-level approach proposed by Sun. The model integrates neural, reinforcement, and symbolic learning methods to perform on-line, bottom-up learning (i.e., learning that neural to symbolic representations). The model utilizes procedural and declarative knowledge (in neural and symbolic representations, respectively), tapping into the synergy of the two types of processes. It was applied to deal with sequential decision tasks. Experiments and analyses of various ways are reported that shed light on the advantages of the model.     

Analyzing CAD competence with univariate and multivariate learning curve models

Understanding how learning occurs, and what improves or impedes the learning process is of importance to academicians and practitioners; however, empirical research on validating learning curves is sparse. This paper contributes to this line of research by collecting and analyzing CAD (computer-aided design) procedural and cognitive performance data for novice trainees during 16-weeks of training. The declarative performance is measured by time, and the procedural performance by the number of features used to construct a design part. These data were analyzed using declarative or procedural performance separately as predictors (univariate), or a combination of declarative or procedural predictors (multivariate). Furthermore, a method to separate the declarative and procedural components from learning curve data is suggested.     

The instructor's gaze guidance in video lectures improves learning

Instructor behaviour is known to affect learning performance, but it is unclear which specific instructor behaviours can optimize learning. We used eye‐tracking technology and questionnaires to test whether the instructor's gaze guidance affected learners' visual attention, social presence, and learning performance, using four video lectures: declarative knowledge with and without the instructor's gaze guidance and procedural knowledge with and without the instructor's gaze guidance. The results showed that the instructor's gaze guidance not only guided learners to allocate more visual attention to corresponding learning content but also increased learners' sense of social presence and learning. Furthermore, the link between the instructor's gaze guidance and better learning was especially strong for participants with a high sense of social connection with the instructor when they learned procedural knowledge. The findings lead to a strong recommendation for educational practitioners: Instructors should provide gaze guidance in video lectures for better learning performance.     

Transitional feedback schedules during computer-based problem-solving practice

Feedback has a strong influence on effective learning from computer-based instruction. Prior research on feedback in computer-based instruction has mainly focused on static feedback schedules that employ the same feedback schedule throughout an instructional session. This study examined transitional feedback schedules in computer-based multimedia instruction on procedural problem-solving in electrical circuit analysis. Specifically, we compared two transitional feedback schedules: the TFS-P schedule switched from initial feedback after each problem step to feedback after a complete problem at later learning states; the TFP-S schedule transitioned from feedback after a complete problem to feedback after each problem step. As control conditions, we also considered two static feedback schedules, namely providing feedback after each practice problem-solving step (SFS) or providing feedback after attempting a complete multi-step practice problem (SFP). Results indicate that the static stepwise (SFS) and transitional stepwise to problem (TFS-P) feedback produce higher problem solving near-transfer post-test performance than static problem (SFP) and transitional problem to step (TFP-S) feedback. Also, TFS-P resulted in higher ratings of program liking and feedback helpfulness than TFP-S. Overall, the study results indicate benefits of maintaining high feedback frequency (SFS) and reducing feedback frequency (TFS-P) compared to low feedback frequency (SFP) or increasing feedback frequency (TFP-S) as novice learners acquire engineering problem solving skills.     

Engineering information integration using object‐oriented mediator technology

The mechanical product development process uses many different software systems to virtually simulate the behaviour of a design. The present work deals with flexible and efficient integration using object‐oriented mediator technology that provides transparent access to distributed engineering systems. The use of mediator technology is investigated for semi‐automatically integrating engineering information resident in computer aided design systems with a Common Object Request Broker Architecture based application programming interface. The purpose is to provide engineering analysis applications access to computer aided design system information and computational methods through a declarative query language. We conclude that the use of a declarative query language for developing engineering applications shows great potential in terms of flexibility, development productivity, performance, and ease of use, compared with using a procedural programming language. The work also shows new use of mediator technology, declarative queries, and active rules within engineering information integration that traditionally is accomplished using procedural programming. Copyright © 2004 John Wiley & Sons, Ltd.     

Competent information search in the World Wide Web: Development and evaluation of a web training for pupils

This paper describes the development and empirical evaluation of a web training for pupils (CIS-WEB, Competent Information Search in the World Wide WEB) which aims to convey prerequisite knowledge and skills that are necessary for a competent search for information on the web. The web training focuses on competent information handling and is based on two theoretical analyses. First, a conceptual analysis of information search from the perspective of media literacy research and information retrieval research was conducted and yielded a set of five pivotal content aspects that need to be covered by a web training. Each of these content aspects is characterized by declarative and procedural knowledge components which are necessary for the pursuit of a competent search for information on the web. Second, we conducted a task analysis which conceptualizes the search for information on the web as a problem-solving process and which allows to systematically distinguish between different types of information problems. In the empirical part of the paper two classroom studies are reported. In Study 1, the widespread training concept of a technically oriented Internet training for pupils was evaluated and it was shown that no substantial improvement of web searching skills can be expected from this type of treatment. In Study 2, it was shown that the web training CIS-WEB improves pupils’ declarative knowledge of the web as well as their search performance, thereby outperforming the conventional Internet training used in Study 1.     

Instructions and demonstration as media for training new users of Automatic Speech Recognition Devices

Abstract

Although automatic speech recognition (ASR) can provide a medium of controlling computers which is relatively easy to use, novice users often have problems with it during their initial practices. In this study, two methods for training subjects to use ASR are compared. One group of subjects received a short demonstration given by an experienced ASR user and the other group received verbal instructions on how to use the device. The results show that subjects given a demonstration achieved better performance than those given instructions (p< 0.005). This is explained by virtue of the fact that the successful use of ASR requires procedural knowledge which is better acquired through some form of practice than through instruction. It is concluded that a demonstration provides ‘practice by proxy’. ‘Task like’ forms of enrolment are discussed. It is suggested that although they can provide the possibility of practice, they are not applicable to all types of ASR use. A demonstration provides users with task familiarization, and an appropriate style of speech.     

Improving surgical pattern recognition through repetitive viewing of video clips

Previous field studies show that surgery residents and medical students have difficulty recognizing appropriate anatomic cues during laparoscopic surgery, causing delays in procedures and errors. Such observations led to the development of an anatomy recognition training intervention, specifically the use of an ordered set of video clips that show the main steps of a laparoscopic procedure. Each procedural step is shown several times in succession, with each repetition coming from a different surgery, thus exposing the learner to varied anatomy and crucial maneuvers. A factorial experiment with 30 medical students showed that the group using these ordered, perceptual learning modules had a significant increase in scores on questions assessing perceptual knowledge and procedural knowledge, with no corresponding increase for the control group who watched the videos from the same cases but in an unstructured format for the same amount of time (p< 0.05). Neither group showed improvement on strategic or declarative knowledge tests. The study suggests that ordered perceptual learning modules are a potential means for training perceptual and procedural knowledge in an effective, safe, and efficient manner, serving as a complement to other types of training methods that teach physical dexterity, strategic and declarative knowledge.     

Proprioception Variables as Determinants of Lapsed Time Estimation∗

Abstract

It was postulated that increasing positional and force cues would provide more discriminable stimuli to serve as references for the reproduction of a motor response. The levels and mode of the proprioceptive feedback were varied during the coincident positioning task of moving a carriage along a trackway in 2·0 sec. In a balanced design with 80 Ss, 2 values of positional cues, moving the carriage over 2·5 or 65·0 cm, and 2 values of force cues, minimal resistance and 18·9 kg, produced 4 different feedback combinations. Significant learning occurred during 30 trials with knowledge of results (KB), and the combinations involving movement produced the greatest accuracy. Throughout 30 trials without knowledge of results (NoKR) the estimates produced by the movement only and weight only groups became longer, whilst the control group and the movement plus weight group did not drift. It was suggested that the group differences were caused by fatigue.

Interpolated rests of 1 min between groups of 5 trials during the KB condition shortened the first post-rest trial whence performance was restored by the remaining 4 trials. During NoKB the first post-rest trial was again shorter, but since the remaining 4 were overestimated it was more accurate.

In view of these results, apparently redundant movements might aid in timing tasks requiring temporal and spatial organization.     

Integrating discrete events and continuous head movements for video-based interaction techniques

Human head gestures can potentially trigger different commands from the list of available options in graphical user interfaces or in virtual and smart environments. However, continuous tracking techniques are limited in generating discrete events which could be used to execute a predefined set of commands. In this article, we discuss a possibility to encode a set of discrete events by integrating continuous head movements and crossing-based interaction paradigm. A set of commands can be encoded through specific sequences of crossing points when a head-mouse cursor such as a scaled pointer interacts with a graphical object. The goal of the present experiment was testing the perceptual-motor performance of novices in target acquisition tasks using a subset of round head gestures and symbolic icons designating eight types of directional head movements. We have demonstrated that the novices can equally well execute round head gestures in clockwise and counter-clockwise directions by making two crossings for about 2 s or three crossings for about 3 s. None of the participants reported neck strain or other problems after 360 trials performed during a 40-min test in each of 5 days.     

Adaptability of training in simulated process control: knowledge- versus rule-based guidance under task changes and environmental stress

OBJECTIVE: The study examined the adaptability of different types of process control training across changes in task and environmental stress. BACKGROUND: The literature on training leads us to expect greater flexibility for system-based training, as opposed to procedure-based training. However, the stress literature suggests that knowledge-based strategies (making use of executive control) may be more vulnerable under stress conditions. METHOD: Two groups were given 6 hr of training on the Cabin Air Management System (CAMS), a complex, multilevel, PC-based process control task, emphasizing either system knowledge or use of procedures. They were then required to carry out the task for 3 hr (with noise during the middle 1 hr) across a range of both familiar and unfamiliar fault scenarios. RESULTS: For the primary control task, the system-trained group performed better, especially for less familiar and complex faults. However, for lower priority tasks requiring executive control, procedure-trained operators performed better and were less impaired by noise. CONCLUSION: System training was more effective for managing unexpected task events, whereas procedural training was better under noise. The results are interpreted in terms of the rationale for instructing operators in the range of strategies required for effective process skills in complex work environments. APPLICATION: Training methodologies for safety critical applications should aim to develop skill in the use of both procedural and system knowledge strategies. Operators should be trained in the most effective deployment of these strategies during unfamiliar task events and environmental stress and given stress exposure training.     

The comparisons of the influences of prior knowledge on two game-based learning systems

Game-based learning provides many benefits, such as enhancing the interaction with students and stimulating their learning motivation. Thus, it is popular to learners who have diverse characteristics. To this end, individual differences play an essential role. Among various individual differences, previous studies demonstrated that prior knowledge has great effects on game-based learning. However, such studies mainly considered a single game-based learning system. To address this issue, this study examined how prior knowledge affects students' reactions to two different types of game based learning systems, i.e., the Machinarium and the CSI: Web Adventures. The former delivers procedural knowledge while the latter provides declarative knowledge. The results from this study indicate that prior knowledge has positive impacts in the CSI group while it has negative impacts in the Machinarium group. These findings imply that prior knowledge is useful for the context of declarative knowledge while it may not be helpful for the context of procedural knowledge.     

When simulated environments make the difference: the effectiveness of different types of training of car service procedures

An empirical analysis was performed to compare the effectiveness of different approaches to training a set of procedural skills to a sample of novice trainees. Sixty-five participants were randomly assigned to one of the following three training groups: (1) learning-by-doing in a 3D desktop virtual environment, (2) learning-by-observing a video (show-and-tell) explanation of the procedures, and (3) trial-and-error. In each group, participants were trained on two car service procedures. Participants were recalled to perform a procedure either 2 or 4 weeks after the training. The results showed that: (1) participants trained through the virtual approach of learning-by-doing performed both procedures significantly better (i.e. p < .05 in terms of errors and time) than people of non-virtual groups, (2) the virtual training group, after a period of non-use, were more effective than non-virtual training (i.e. p < .05) in their ability to recover their skills, (3) after a (simulated) long period from the training—i.e. up to 12 weeks—people who experienced 3D environments consistently performed better than people who received other kinds of training. The results also suggested that independently from the training group, trainees’ visuospatial abilities were a predictor of performance, at least for the complex service procedure, adj R ² = .460, and that post-training performances of people trained through virtual learning-by-doing are not affected by learning styles. Finally, a strong relationship (p < .001, R ² = .441) was identified between usability and trust in the use of the virtual training tool—i.e. the more the system was perceived as usable, the more it was perceived as trustable to acquire the competences.     

Open-shop dense schedules: properties and worst-case performance ratio

Dense schedules are easy to construct and can be used as heuristic solutions for open-shop problems. It is conjectured that in the worst-case a dense schedule will result in a makespan no more than (2-\frac1m)(2-\frac{1}{m}) times of the makespan of the optimal schedule, where m is the number of machines. Different approaches have been used in proofs for different number of machines. The conjecture has been proved for the number of machines m≤6, and for some special types of dense schedules. In this paper, we first summarize some useful properties of dense schedules, and then provide some special types of dense schedules that make the conjecture hold. Finally, we give complete proofs of the conjecture for m=7 and 8.     

Measuring mental workload: ocular astigmatism aberration as a novel objective index

This study assessed the effect of two perceptually matched mental tasks with different levels of mental demand on ocular aberrations in a group of young adults. We measured ocular aberration with a wavefront sensor, and total, internal and corneal RMS (root mean square) aberrations were calculated from Zernike coefficients, considering natural and scaled pupils (5, 4.5, and 4 mm). We found that total, internal and corneal astigmatism RMS showed significant differences between mental tasks with natural pupils (p < .05), and this effect was maintained with 5 mm scaled pupils (total RMS astigmatism, p < .05). Consistently, pupil size, intraocular pressure, perceived mental load and cognitive performance were influenced by the level of mental complexity (p < .05 for all). The findings suggest that ocular astigmatism aberration, mediated by intraocular pressure, could be an objective, valid reliable index to evaluate the impact of cognitive processing in conjunction with others physiological markers in real world contexts.

Practitioner Summary: The search continues for a valid, reliable, convenient method of measuring mental workload. In this study we found ocular astigmatism aberration is sensitive to the cumulative effect of mental effort. It shows promise of being a novel ocular index which may help to assess mental workload in real situations.     

INCREMENTAL LEARNING OF PROCEDURAL PLANNING KNOWLEDGE IN CHALLENGING ENVIRONMENTS

Autonomous agents that learn about their environment can be divided into two broad classes. One class of existing learners, reinforcement learners, typically employ weak learning methods to directly modify an agent's execution knowledge. These systems are robust in dynamic and complex environments but generally do not support planning or the pursuit of multiple goals. In contrast, symbolic theory revision systems learn declarative planning knowledge that allows them to pursue multiple goals in large state spaces, but these approaches are generally only applicable to fully sensed, deterministic environments with no exogenous events. This research investigates the hypothesis that by limiting an agent to procedural access to symbolic planning knowledge, the agent can combine the powerful, knowledge-intensive learning performance of the theory revision systems with the robust performance in complex environments of the reinforcement learners. The system, IMPROV, uses an expressive knowledge representation so that it can learn complex actions that produce conditional or sequential effects over time. By developing learning methods that only require limited procedural access to the agent's knowledge, IMPROV's learning remains tractable as the agent's knowledge is scaled to large problems. IMPROV learns to correct operator precondition and effect knowledge in complex environments that include such properties as noise, multiple agents and time-critical tasks, and demonstrates a general learning method that can be easily strengthened through the addition of many different kinds of knowledge.