One approach to the problem of formulating domain models for planning is to learn the models from example action sequences. The LOCM system demonstrated the feasibility of learning domain models from example action sequences only, with no observation of states before, during or after the plans. LOCM uses an object-centred representation, in which each object is represented by a single parameterised state machine. This makes it powerful for learning domains which fit within that representation, but there are some well-known domains which do not. This paper introduces LOCM2, a novel algorithm in which the domain representation of LOCM is generalised to allow multiple parameterised state machines to represent a single object. This extends the coverage of domains for which an adequate domain model can be learned. The LOCM2 algorithm is described and evaluated by testing domain learning from example plans from published results of past International Planning Competitions.

Most planning formalisms allow instances with shortest plans of exponential length. While such instances are problematic, they are usually unavoidable and can occur in practice. There are several known cases of restricted planning problems where plans can be exponential but always have a compact (ie. polynomial) representation, often using recursive macros. Such compact representations are important since exponential plans are difficult both to use and to understand. We show that these results do not extend to the general case, by proving a number of bounds for compact representations of plans under various criteria, like efficient sequential or random access of actions. Further, we show that it is unlikely to get around this by reformulating planning into some other problem. The results are discussed in the context of abstraction, macros and plan explanation.

We address a scheduling problem in the context of military aircraft maintenance where the goal is to meet the aircraft requirements for a number of missions in the presence of breakdowns. The assignment of aircraft to a mission must consider the requirements for the mission, the probability of aircraft failure, and capacity of the repair shop that maintains the aircraft. Therefore, a solution both assigns aircraft to missions and schedules the repair shop to meet the assignments. We propose a dispatching heuristic algorithm; three complete approaches based on mixed integer programming, constraint programming, and logic-based Benders decomposition; and a hybrid heuristic-complete approach. Experiments demonstrate that the logic-based Benders variation combining mixed integer programming and constraint programming outperforms the other approaches, that the dispatching heuristic can feasibly schedule the repair shop in a very short time, and that using the dispatching solution as a bound marginally improves the complete approaches.

Bayesian Network Classifiers are popular approaches for classification problems where instances have to be assigned to one of several classes. However, in many domains, it is necessary to assign instances to multiple classes at the same time. This task has been normally addressed either by (i) transforming the problem into a single-class scenario by defining a new class variable with all of the possible combinations of classes or, (ii) by building an independent classifier for each class variable. Either way, the resulting models do not capture all the relations and dependencies between classes and features resulting into unprecise multidimensional classifiers. In this paper, we introduce a two-step method for learning Multidimensional Bayesian Network Classifiers (MBC) from data based on mutual information measures. The first step of the method learns an initial MBC structure which then, in the second step, is refined. Our approach is simple and keeps all the interactions and dependencies among classes and features. The method was tested on three benchmark multidimensional data-sets. Preliminary experimental results show how our method outperforms state-of-the-art methods used in multidimensional classification.

Teaching an upper-level undergraduate robotics course at Historically Black Colleges and Universities (HBCUs) is challenging. The lack of suitable teaching materials is one of the biggest challenges, although there are many great masterpieces in developing robotics course materials, which are, however, generally developed for teaching students at elite Research I universities. This paper presents ideas and details in adopting and revising these course materials and creating upper-level undergraduate robotics course materials that are suitable for underrepresented students.

The mission of the ARTSI (Advancing Robotics Technology for Societal Impact) Alliance, a consortium of 19 Historically Black Colleges and Universities (HBCUs) and 9 major research universities (R1s), is to enlarge the nation’s engineering and science talent pool by increasing the number of students from underrepresented groups who pursue advanced training in computer science. ARTSI is one of several alliances funded by the National Science Foundation’s Broadening Participation in Computing Program. ARTSI focuses specifically on institutions serving African Americans and uses robotics education to attract and engage students. In this paper we describe the activities comprising ARTSI, our vision of a robotics curriculum for CS undergraduates, and ways to integrate robotics modules into existing CS courses.

This allows for supporting more adequate and accurate representations of learners, their learning goals, learning material and contexts of its use, as well as more efficient access and navigation through learning resources. The goal is to advance intelligent educational systems, so as to achieve improved e-learning efficiency, flexibility and adaptation for single users and communities of users (learners, instructors, courseware authors, and others). The special track follows the workshop series Ontologies and Semantic Web for e-Learning, which was conducted successfully from 2002-2009 at a number of different conferences. The goals of this track are to discuss the current state-of-the-art in using ontologies and semantic web technologies in e-learning applications; and to attract the interest of the related research communities to the problems in the educational social semantic web and serve as an international platform for knowledge exchange and cooperation between researchers. This special track will be of interest to researchers interested in using ontologies, semantic web and social semantic web technologies in web-based educational systems, distributed hypermedia and open hypermedia systems, as well as in web intelligence and semantic web and social semantic web engineering.

In this work we hypothesize that giving students a reflective reading after spoken dialog tutoring in qualitative physics will improve learning. The reading is designed to help students compare similar aspects of previously tutored problems, and to abstract their commonalities. We also hypothesize that student motivation will affect how well the text is processed, and so influence learning. We find that the beneficial effects of the reflective text significantly interact with motivation, such that moderately motivated students learn significantly more from the reflective text than from a non-reflective control text. More poorly or highly motivated students did not benefit from reflective text. These results demonstrate that implicit reflection can improve learning after dialog tutoring with a qualitative physics tutor. They further demonstrate that this result can be obtained with a reflective/abstractive text without recourse to dialog, and that the effectiveness of the text is sensitive to the motivation level of the student.

An important aspect of Intelligent Tutoring Systems is providing assistance to students as well as assessing them. The standard state-of-the-art algorithms (Knowledge Tracing and Performance Factor Analysis) for tracking student knowledge, however, only look at the correctness of student first response and ignore the amount of assistance students needed to eventually answer the question correctly. In this paper, we propose the Assistance Model (AM) for predicting student performance using information about the number of hints and attempts a student needed to answer the previous question. We built ensemble models that combine the state-of-the-art algorithms and the Assistance Model together to see if the Assistance Model brings improvements. We used an ASSISTments dataset of 200 students answering a total of 4,142 questions generated from 207 question templates. Our results showed that the Assistance Model did in fact reliably increase predictive accuracy when combined with the state-of-the-art algorithms.

Research on automated essay scoring (AES) indicates that computer-generated essay ratings are comparable to human ratings. However, despite investigations into the accuracy and reliability of AES scores, less attention has been paid to the feedback delivered to the students. This paper presents a method developers can use to quickly evaluate the usability of an automated feedback system prior to testing with students. Using this method, researchers evaluated the feedback provided by the Writing-Pal, an intelligent tutor for writing strategies. Lessons learned and potential for future research are discussed.