Steven Ritter's research while affiliated with Carnegie Learning and other places

... The model might exploit such information during training -e.g., to achieve representation invariance -or during inference, in order to perform group-speci c calibration or thresholding. Such a model inherits the well-documented problems that arise from sorting humans into categories based predominantly on sociocultural constructs [4,30], as opposed to physiological di erences relevant to the prediction task at hand. As an alternative, consider now the case of a model that has learned an implicit representation of group membership. ...

... Scientific findings can then generally contribute to the theory used to inform subsequent educational software designs. Further extensions of this work can be found at upgradeplatform.org, which supports UpGrade, an open-source platform for A/B testing in educational software (Ritter et al, 2020). ...

... In some cases, these two types of dashboard are being integrated. For example MATHia LiveLab predicts if a student will fail to reach mastery and provides these predictions to teachers (Fancsali et al., 2020). ...

... Online English teaching assistance system, using decision tree Frontiers in Education 03 frontiersin.org algorithms and neural network models, was also implemented, which improved the efficiency of teaching (Fancsali et al., 2018;Zheng et al., 2019;Sun et al., 2021). However, the hidden layer(s) likes a black box, as the common factors inside cannot be directly observed. ...

... This could consequently have an impact on their performance and graduation prospects. The same goes for students who graduated from high school with a low Grade-Point Average (GPA), who may be more likely than those with high GPAs to perform poorly in college [5]. Earlier LA models depended on preset parameters to deliver a single set of predictions within a given time frame. ...

... The Shipborne Over-and Underwater integrated Mobile Mapping System (SiOUMMS) described in this paper integrates a multi-beam echo sounder, laser scanner, GNSS, and Inertial Measurement Unit (IMU) device onboard and fixedly connects the sensors to ensure that their spatial relationship remains unchanged, which realizes their physical integration and provides the basis for seamless integration of over-and under-water point clouds. To meet the practical demands of various application scenarios and support a diversity of sensors, plug-in (Grapenthin et al. 2014;Ritter et al. 2016) and multi-process network technologies are used to integrate data acquisition and cooperative control with multi-sensors and to ensure high cohesion and low coupling in the system. This enhances the performability, testability, maintainability, and load balancing of the system, as is highlighted in this paper and can provide a good reference framework for the integration of similar systems. ...

... Correct attempts after an initial hint are now credited (i.e., skill mastery estimates increase) like immediate correct attempts, and correct attempts after mid-level hints now leave the skill mastery estimate unchanged. In addition, MA-THia's BKT parameter estimates for each skill (used to determine the models' responsiveness to correct and incorrect answers) are now frequently set based on data-driven estimation techniques [32,39] as opposed to mostly being set according to expert judgment in earlier Cognitive Tutor versions. ...

... Quality math education, particularly at a young age, is of crucial importance for students growing into an increasingly technology-driven world. Intelligent tutoring systems (ITSs) have demonstrated their effectiveness in improving math learning outcomes [1,10,24]. Many ITSs have a component that provides automated feedback for students while they solve math questions. ...

... Later, with the rise of classic machine learning methods such as logistic regression models [117], another direction followed by KT is to use parametric factor analysis approaches which trace a student's knowledge states and perform the answer prediction based on modeling a variety of factors [16,17,86], including: (1) aspects about students such as prior knowledge, learning capacity, or learning rate; (2) aspects about learning materials such as familiarity, number of previous practices, or difficulty; (3) aspects about a learning environment itself such as the nature of the learning channel (paper-or computer-based) and the temporal context of the practice time (within an examination period or a regular study period). In addition to these, psychological studies about the learning behavior [75] and forgetting behavior [52] of students have also suggested additional factors, such as the time lapse between a student's different interactions and the number of times on practicing learning materials, to be considered when tracing knowledge states. It is worth noting that this direction of KT is still active [35,111] and is considered as an alternative to the recent state-of-the-art approaches based on deep learning. ...

... Log data from math tutors have also been used to predict student scores on end-of-year state accountability exams, resulting in better prediction than paper-pencil benchmark tests and standardized tests [4,18,21]. These models become better still when supplemented with data on student strategies [36]. ...