I am slowly migrating to a new website at sachingrover211.github.io
Peer Reviewed Workshops, Conference and Journals
Grover, S.; Arora, K.; Mitra, S.K., " Text Extraction from Document Images Using Edge Information", Proceedings of India Conference of IEEE INDICON'09, published by IEEE. [abstract and pdf link]
VanLehn, K., Chung, G., Grover, S., Madni, A. & Wetzel, J. 2016. "Learning science by constructing models: Can Dragoon increase learning without increasing the time required?" International Journal of Artificial Intelligence in Education, pp. 1-36. [abstract and pdf link]
VanLehn, K., Wetzel, J, Grover, S. & van de Sande, B. 2016. "Learning how to construct models of dynamic systems: An initial evaluation of the Dragoon intelligent tutoring system." IEEE Transactions on Educational Technology. [abstract and pdf link]
Wetzel, J., VanLehn, K., Chaudhari, P., Desai, A., Feng, J., Grover, S., Joiner, R., Kong-Silvert, M., Patade, V., Samala, R., Tiwari, M. & van de Sande, B. 2016. "The design and development of the Dragoon intelligent tutoring system for model construction: Lessons learned." Interactive Learning Environments, pp.1-21.
Grover, S., Wetzel, J. and VanLehn, K., 2018, June. How Should Knowledge Composed of Schemas be Represented in Order to Optimize Student Model Accuracy?. In International Conference on Artificial Intelligence in Education (pp. 127-139). Springer, Cham.
Grover, S., Chakraborti, T. and Kambhampati, S., 2018. What can Automated Planning do for Intelligent Tutoring Systems?. ICAPS SPARK.
Chakraborti, T., Sreedharan, S., Grover, S. and Kambhampati, S., 2018. Plan Explanations as Model Reconciliation--An Empirical Study. arXiv preprint arXiv:1802.01013. HRI, 2019.
Plan Explicability in Stochastic Environment
Supervisor: Prof. Subbarao Khambampati, ASU, Tempe (Ongoing)
Performing explicable actions are inherent to humans. Here explicable means the actions that any observer can understand what an automated agent is trying to do. An example can be, an agent fetchin coffee for a user and making sure that actions will help the human understand what he is trying to do. It is easy to understand this from motion planning perspective, where humans have a naive model of moving in a straight line towards the goal. But task or action planning is more general, as this naive human model is not readily available. I am working on this problem in a stochastic environment, to understand the effect of stochasticity on the problem. As a subpart of the original problem, I am currently looking at learning preferences of a human observer using Skip Gram technique over plan traces generated from optimal policies.
System of equation solver
Supervisor: Prof. Kurt VanLehn, ASU, Tempe
Implemented matrix operations like addition, multiplication and calculating inverse in JavaScript. Using the base class implemented Newton Raphson method to solve a system of equations. This is being used to solve dynamic system models, and create graphs in TopoMath, an Intelligent Tutoring System for teaching, modeling to high school and college students.
Making PeopleBot a People's Bot
Supervisor: Prof. Yu Zhang, ASU, Tempe. Group size: 4 (including)
The agenda for the project was to make PeopleBot a member of a normal meeting. A complete pipeline was created to do an end to end analysis by a robot. First phase involved recognizing the members using HAAR wavelets to recognize faces, recognizing the speaker using LSTMs and sound recognition techniques and speech to text conversion using Google API. The complete work for speaker recognition was taken from Github. The second phase involved coherence recognition from the spoken discussion by each speaker. Spoken dialogues were tagger using NLP tagger and then these were provided to an HMM to decide coherence among the group members. The complete code can be checked at Github.
Assignment of Blame Problem
Supervisor: Prof. Kurt VanLehn, ASU, Tempe
In a tutoring system, if a solution step requires the knowledge of multiple KCs and student is not able to answer the step correctly, then it is unclear which knowledge component to blame. This project involves implementating different ways to assign blame to knowlege components. The results are calculated on Dragoon data using AUC values.
Online Assessment of Student Learning - Dragoon
Supervisor: Prof. Kurt VanLehn, ASU, Tempe
Dragoon [homepage] is a tutoring software which teaches students to build models of dynamic systems. Working on implementing Bayesian Knowledge Tracing (BKT), to assess student's work in Dragoon problems. Implementing algorithms to calibrate BKT and use problem difficulty, to improve accuracy in student modeling. The code for the software is available at Github.
Dragoon Dashboard
Supervisor: Prof. Kurt VanLehn & Brett Van de Sande, ASU, Tempe
Designed and Developed the log structure to track student actions. Created a platform which can be used by teachers in class to know how the class is performing and also findout if any student is stuck or needs help. Dashboard was used in a class in Spring 2014 with a running class. The data values given by Dashboard are also being used to analyse the differences between normal and control case for the software. The complete implementation ran over two different versions of the software. Both version can be reached from the github.
Web Ontology Language (OWL) to Answer Set Programming
Supervisor: Prof. Joohyung Lee, ASU Tempe. Group Size: 2 (including)
Proposed a way to convert OWL DL in RDF/XML format to First order Logic Programs. The first order logic language used was F2LP. So an in depth understanding of the usage was studied. The conversion could be used to find the exhaustive stable models with specific properties. Something that is done a little differently using Protege. We discussed and worked on the method for conversion and then implemented the parser. The code for converter was written in Python and was able to convert huge OWL ontologies quickly.
Stepwise Incremental Convergence PSO: A novel PSO based algorithm
Supervisor: Prof. Banshidhar Majhi, N.I.T. Rourkela, Members: 2 (including)
A novel approach to Particle swarm optimization is proposed which overcomes the limitations of the basic PSO algorithm, namely of becoming stuck in a local optimum. The new PSO based algorithm, called SIPSO, shows a better convergence to the global optimum solution and a greater reliability of the solution for combinatorial optimization problems.
Text Extraction from an Image using Edge Information
Supervisor : Prof. Suman Mitra, DAIICT, Gandhinagar. Members : 2 (including)
A novel method of marking the text areas in an image was tested. The proposed method was based on collecting the edge information first using sobel operators and then harnessing the property of sharp edges for the text and there by marking the areas as text or non text regions.
Learning 3-D Scene Structuring from a Single Monocular Image
Supervisor : Prof. Suman Mitra, DAIICT, Gandhinagar. Members :2 (including)
In this project the task of depth estimation from a single monocular image was considered. A supervised learning approach was taken to predict the depth map as a function to this image. The monocular cues like texture, texture gradient, haze and occlusion were used to define the feature vector which were trained to predict the depth based on learning. The earlier approach used in make3D [Homepage] was understood and a new method was worked upon.
Image Restoration, NASRIF method using Swarm Intelligence Techniques
Supervisor : Prof. Banshidhar Majhi, N.I.T., Rourkela. Members :2(including)
Image Restoration is currently classified into 6 categories, NASRIF being the recent most method blind Image Restoration and we intended to improve it further using Ant Colony Optimization and compare it with the original results.