Today, speech technology is only available for a small fraction of the thousands of languages spoken around the world because traditional systems need to be trained on large amounts of annotated speech audio with transcriptions. Obtaining that kind of data for every human language and dialect is almost impossible.
Wav2vec works around this limitation by requiring little to no transcribed data. The model uses self-supervision to push the boundaries by learning from unlabeled training data. This enables speech recognition systems for many more languages and dialects, such as Kyrgyz and Swahili, which don’t have a lot of transcribed speech audio. Self-supervision is the key to leveraging unannotated data and building better systems.
In the Developmental Intelligence Laboratory, we are interested in understanding fundamental cognitive mechanisms of human intelligence, human learning, and human interaction and communication in everyday activities. To do so, we collect and analyze micro-level multimodal behavioral data using state-of-the-art sensing and computational techniques. One of our primary research aims is to understand human learning and early development. How do young children acquire fundamental knowledge of the world? How do they select and process the information around them and learn from scratch? How do they learn to move their bodies and to communicate and interact with others? Learning this kind of knowledge and skills is the core of human intelligence. To understand how human learners achieve the learning goal, the primary approach in our research is to attach GoPro-like cameras on the head of young children to record egocentric video from their point of view. Using this innovative approach, we've been collecting video data of children’s everyday activities, such as playing with their parents and their peers, reading books with parents and caregivers, and playing outside. We've been using state-of-the-art machine learning and data mining approaches to analyze high-density behavioral data. This research line will ultimately solve the mystery on why human children are such efficient learners. Moreover, the findings from our research will be used to help improve learning of children with developmental deficits. A complimentary research line is to explore how human learning can teach us about how machines can learn. Can we model and simulate how a human child learns and develops? To this end, our research aims at bridging and connecting developmental science in psychology and machine learning and computer vision in computer science.
This paper provides a summary account of Activity-Centred Analysis and Design (ACAD). ACAD offers a practical approach to analysing complex learning situations, in a way that can generate knowledge that is reusable in subsequent (re)design work. ACAD has been developed over the last two decades. It has been tested and refined through collaborative analyses of a large number of complex learning situations and through research studies involving experienced and inexperienced design teams. The paper offers a definition and high level description of ACAD and goes on to explain the underlying motivation. The paper also provides an overview of two current areas of development in ACAD: the creation of explicit design rationales and the ACAD toolkit for collaborative design meetings. As well as providing some ideas that can help teachers, design teams and others discuss and agree on their working methods, ACAD has implications for some broader issues in educational technology research and development. It questions some deep assumptions about the framing of research and design thinking, in the hope that fresh ideas may be useful to people involved in leadership and advocacy roles in the field.
Task analysis is the systematic study of how users complete tasks to achieve their goals. This knowledge ensures products and services are designed to efficiently and appropriately support those goals.
J. Kim, P. Guo, D. Seaton, P. Mitros, K. Gajos, and R. Miller. Proceedings of the First ACM Conference on Learning @ Scale Conference, page 31–40. New York, NY, USA, Association for Computing Machinery, (2014)
P. Adamopoulos. ICIS, Association for Information Systems, (2013)The findings of our analysis illustrate that Professor(s) is the most important factor in online course retention and has the largest positive effect on the probability of a student to successfully complete a course. The sentiment of students for Assignments and Course Material also has positive effects on the successful completeness of a course whereas the Discussion Forum has a positive effect on the probability to partially complete a course. Furthermore, self-paced courses have a negative effect, compared to courses that follow a specific timetable. In addition, the difficulty, the workload, and the duration of a course have a negative effect. On the other hand, for the more difficult courses, self-paced timetable, longer duration in weeks, and more workload have a positive effect on the probability to successfully complete a course. Besides, final exams and projects, open textbooks, and peer assessment have also positive effects. Moreover, whether a certificate is awarded upon the successful completion of a course also affects retention. Additionally, the better a university is considered (i.e. higher ranking), the more likely that a student will successfully complete a course. Further, our results illustrate that the courses which belong to the academic disciplines of Business and Management, Computer Science, and Science have a positive significant effect in contrast to courses in other disciplines (i.e. Engineering, Humanities, and Mathematics). Finally, attrition was not found to be related with student characteristics (i.e. gender, formal education)..