Areas of Interest
We believe assessments are really important to scale learning and give individuals the socio-economic benefit of learning in form of employment. We undertake research that makes this happen and impacts lives of millions of people.
Some key areas which interest us and are active research areas for us include:
- IRT based adaptive testing instruments
- Scalable personality and soft skill assessment
- Modeling techniques to build interpretable models that predict job success or employment outcome based on assessment scores and other parameters.
- Grading of computer programs
- Grading of spoken language
- Semi automated ways of grading
Machine Learning is the glue
Research in assessments is highly multidisciplinary that involves areas of education psychology, computer science, statistics and others. For us, a machine learning based framework glues these together to help develop a highly accurate and automated constructed response assessment. We presented such a framework in our 2013 NIPS DDE paper. The right feature engineering based on an assessment rubric is the key to address assessment problems. We have demonstrated this concept successfully for automated evaluation of computer programs.
Adding human intelligence to machine learning
Secondly, we are excited about combining human intelligence with machine learning. We use crowdsourcing specifically for the feature derivation step of machine learning. We identify human intelligence tasks that could aid accurate feature derivation and use the crowd to complete them. This technique makes machine learning super powerful. We demonstrated the mettle of the technique to automatically grade spoken English, a hitherto unsolved problem.