Search for: All records

The capability to generate responses with diversity and faithfulness using factual knowledge is paramount for creating a human-like, trustworthy dialogue system. Common strategies either adopt a two-step paradigm, which optimizes knowledge selection and response generation separately and may overlook the inherent correlation between these two tasks, or leverage conditional variational method to jointly optimize knowledge selection and response generation by employing an inference network. In this paper, we present an end-to-end learning framework, termed Sequential Posterior Inference (SPI), capable of se- lecting knowledge and generating dialogues by approximately sampling from the posterior distribution. Unlike other methods, SPI does not require the inference network or assume a simple geometry of the posterior distribution. This straightforward and intuitive inference procedure of SPI directly queries the response generation model, allowing for accurate knowledge selection and generation of faithful responses. In addition to modeling contributions, our experimental results on two common dialogue datasets (Wizard of Wikipedia and Holl-E) demonstrate that SPI outperforms previous strong baselines according to both automatic and human evaluation metrics. 

The activity of the grid cell population in the medial entorhinal cortex (MEC) of the mammalian brain forms a vector representation of the self-position of the animal. Recurrent neural networks have been proposed to explain the properties of the grid cells by updating the neural activity vector based on the velocity input of the animal. In doing so, the grid cell system effectively performs path integration. In this paper, we investigate the algebraic, geometric, and topological properties of grid cells using recurrent network models. Algebraically, we study the Lie group and Lie algebra of the recurrent transformation as a representation of self-motion. Geometrically, we study the conformal isometry of the Lie group representation where the local displacement of the activity vector in the neural space is proportional to the local displacement of the agent in the 2D physical space. Topologically, the compact abelian Lie group representation automatically leads to the torus topology commonly assumed and observed in neuroscience. We then focus on a simple non-linear recurrent model that underlies the continuous attractor neural networks of grid cells. Our numerical experiments show that conformal isometry leads to hexagon periodic patterns in the grid cell responses and our model is capable of accurate path integration. 

With the rapid growth of online learning at community colleges and the low course completion and performance associated with it, there has been increasing need to identify effective ways to address the challenges in online teaching and learning at this particular setting. Based on open-ended survey responses from 105 instructors and 365 students from multiple community colleges in a state, this study examined instructors’ and students’ perceptions of effective and ineffective instructional practices and changes needed in online coursework. By combining structural topic modelling techniques with human coding, we identified instructional practices that were perceived by both instructors and students as effective in supporting online learning as well as ineffective and needing improvement. Moreover, we identified a handful of misalignments between instructors and students in their perceptions of online teaching, including course workload and effective ways to communicate. 

An extensive theoretical and empirical literature stresses the challenges of online learning, especially among students enrolled in open-access institutions who often struggle more due to job and family commitments and a lack of self-regulated learning skills. As online expansion continues in higher education, understanding the specific challenges students encounter in online coursework, and learning strategies that can help them cope with these challenges, can provide valuable insights to be widely shared. Using open-ended survey data collected from 365 students at a state community college system, this study examined students’ perceptions of challenges of online learning that may lead to undesirable learning outcomes and specific strategies they found effective in addressing these challenges. We combined structural topic modeling and human coding in analyzing student responses. Three sets of challenges—including insufficient time management skills, greater tendencies of multitasking and being distracted in an online learning environment, and ineffective interaction and frustrations with help-seeking—emerged from student responses. In response to these challenges, students reflected on ways to improve online learning experiences and outcomes, including improving time management skills, maintaining an organized and distraction-free study environment, proactively seeking help, and using study strategies to improve learning effectiveness. 

Developmental education is the most widespread strategies used by colleges to provide academically weak students with additional training in key subject areas. To reduce costs and also to address the large volume of enrollment in these courses, many institutions have replaced traditional face-to-face instruction with online instruction in developmental coursework. This paper examines the impact of fully online instruction, compared with traditional face-to-face instruction, on both concurrent developmental course outcomes, and on downstream outcomes, using a unique administrative dataset from a state community college system that includes longitudinal student-unit record data from more than 40,000 students enrolled in developmental education courses. Results from a two-way fixed effects model that controls for selection both at the course- and student-level indicate that taking one’s first developmental course through the online format reduces developmental course completion rate by 13 percentage points and subsequent enrollment in the gatekeeper course by 7 percentage points.