Adaptive learning computer systems
In a one-size-fits-all conventional teaching context, student disparities in the background of pre-knowledge, skills, or comprehension lead to the alienation of the struggling students while boring those who are more experienced. At issue is the appropriateness of the one-size-fits-all pedagogical model. An alternative is adaptive learning. Dated back to 1912, Edward L. Thorndike proposed the idea of a mechanical miracle that intends to conduct personal instructions through special print. This idea has inspired century-old efforts to automate education by creating such a kind of teaching machine that adaptively features automation, timely feedback, and self-pacing to student mastery in classrooms. With the technology evolving, a great range of state-of-the-art works that provide digital “teaching machines” have emerged, allowing adaptive learning. However, most of them focus on math or elementary reading and writing skills, and few are on programming-based courses. The few available are costly, heavy-weighted in adaptability, or limited in mapping competencies to an entire formal assessment, with questions that are not naturally differentiated in an assessment and thus, lacking flexibility. In this study, we would like to design and develop a knowledge graph-based chatbot that gives students context-aware interventions and guidance that meet student needs in knowledge skills and learning style. We are going to use a core gateway course (ITSC 2214) as an example.
Side-project: Toolkit for educational knowledge graphs
An educational knowledge graph (EKG) is a type of knowledge graph that represents various educational entities, participants, resources, and their semantic relationships in a visual and interactive format. In AI-driven education, EKGs play a critical role in supporting teaching and learning by providing an integrative, dynamic, interconnected view of educational multifaceted information for better action-based decision-making in classrooms and beyond. However, the dynamic nature of the education knowledge graph as well as the fact of its growing larger poses issues in its scalability and reliability for reasoning, especially during its evolution. In this undergraduate research project, we are going to measure the efficacy of different approaches in storing and inquiring about educational knowledge graphs and work toward designing a framework and developing a toolkit to construct and fuse knowledge graphs.
Side-project: Model and optimize knowledge structure transformation to facilitate adaptive learning
One of the most significant challenges in education at scale is how to adjust individual student learning experiences and prepare students to succeed in the ever-changing world. Today's learning systems often assume a one-size-fits-all approach, disregarding student variation, which often frustrates struggling students while boring advanced students. An alternative is adaptive learning, which aims at addressing student variation from different perspectives by using the technology to tailor individual student learning based on their abilities and needs.
In this study, we are going to 1) model student knowledge structures as cognitive graphs by synthesizing different forms of data and integrate advanced technologies such as graph optimization algorithms, large-language-models, machine learning, data mining, and artificial intelligence to identify knowledge gaps or misconceptions with actionable recommendation and visualization for individual students.
Side-project: Edge AI-driven sentinel to support connected teaching and learning
Flipped and adaptive learning is becoming widespread as an innovative pedagogy, where adaptability in time is the key to handling student individuality and constant change. To maximize the student learning experience, a demanding task is to identify academically or emotionally struggling students on the fly for targeted timely interventions. In this study, we would like to develop a framework to help instructors monitor classrooms through NVidia Jetson TX 2s or student cell phones, collect and intelligently process streaming data, conduct face and emotion recognition, detect student behavior pattern changes, and identify academically or emotionally struggling students as the in-class activity data is streaming.
Qualified students are skillful at Linux/Unix utilities and Python programming.
Students would be able to
- Demonstrate a web dashboard
- Conduct high-performance real-time object detection by using Jetson TX 2
- Conduct emotion recognition by using Jetson TX 2
- Use TensorFlow and map-reduce
- Demonstrate an understanding of dynamic knowledge graph extraction and integration
- Conduct knowledge graph-based online machine learning
Side-project: Edge AI-driven sentinel and adaptability to support collaborative learning in a computer systems course
Cloud computing is driving the green revolution of digital industries in the age of big data. To master cloud computing techniques to solve problems of massive datasets in groups, students need sufficient hands-on practice as well as a deep understanding of distributed computer systems. However, students entering large lab-intensive undergraduate classes with heterogeneous personal computing environments often possess weak background knowledge in computer systems, programming, and data analytics and prediction, in need of individual guidance. Collaborative learning (namely students learn from each other in a group) could be a part of a solution to reduce the workload in the instructional team caused by the individual needs of students. However, it could be a challenge in a collaborative learning environment to handle student individuality and constant change in a timely manner.
To tackle these issues and respond to the urgent need, we propose to use NVidia hardware and AI products for students who need to program clusters to solve data-intensive problems within groups.
In this study, we would like to quantify the emotional intensities of students in groups and develop a framework to measure individual student learning gains over a semester, significantly identify struggling students and their problem patterns on the fly, and give constructive recommendations to reinforce their learning.
Software:
Context-bound AI Tutoring (link for students) (links for an instructor to set up prompt contexts)
Adaptive Learning Management Platform (link, requiring a Canvas authentication)
Educational data analysis and dashboards for instructors
Intelligent tutoring systems
pSTAP: a tool of computational identification of diverse mechanisms underlying transcription factor-DNA occupancy
The transcription-DNA occupancy is a temporal-spatial dynamic concept, which plays a critical role in the regulation of gene expression. The dynamics related to the nuclear concentrations and the intrinsic properties of driver transcription factors to what happens on the cis-regulatory DNA. Essentially it depends on the concentrations of the factors that bind to these sites, the energetics of their interactions with the specific DNA sequence, and the energetics of any interactions with adjacent specifically bound factors. The technique of chromatin immunoprecipitation and high-throughput sequencing characterize the DNA affinity of specific TF and precisely measure their genome-wide distribution in vivo. It provides a chance for us to understand the underlying mechanisms that direct TF-DNA binding. Considering that there are quite limited information of the sequence-specific TF-TF interactions, in the studies towards the ultimate goal of computationally predicting a TF’s occupancy profile in any cellular condition, we have applied the thermodynamics-based model to
- discriminate the direct and indirect sequence-specific TF-TF interaction (PLoS Genetics in revision)
- discriminate the direct and indirect TF-DNA binding (Manuscript in preparation)
- facilitate the identification of zine-finger motifs (Genome Research 2013, accepted)
Condition-dependent dynamics prediction of protein-protein interaction networks (BIBM2011 and ACM-BCB 2011 )
Recent technological advances have produced large data sets of protein-protein interactions (PPI), which construct a static map. However, it fails to represent dynamic aspects of interactions. The dynamics inference of PPI networks is critical for supporting the comprehensibility of the biological systems.
Proposed a graph-based conflict-free scheduling algorithm to build a training model, which is a critical component in multi-label classification for predicting dynamics of PPI networks.
Explored statistical learning theory and several data mining techniques and tools
Anticipated the writing of an extended NSF proposal based on the project.
Supplementary materials available
AlignNets : Network alignment and motif finding (Bioinformatics 2009, CIKM demo, BIBM 2010,ACM-BCB 2010,....)
Metabolism is a vital cellular process whose understanding is critical to human disease studies and drug discovery. The accumulation of high-throughput genomic, proteomic and metabolic data allows for increasingly accurate modeling and reconstruction of metabolic networks. Alignment of the reconstructed networks can catch model inconsistencies and infer missing elements. Existing alignment tools are mostly based on isomorphic and homeomorphic embedding effectively solving a problem that is NP-complete even when searching a match for a tree in acyclic networks. In this project, I
Designed the first polynomial-time algorithm for efficiently finding optimal homo-homeomorphic embedding from multi-source trees into arbitrary networks which allow for enzyme deletions and insertions
Extended the algorithm to arbitrary networks even with cycles
Proposed a framework of detecting and filling pathways by embedding sequence alignment tools and doing database search for missing enzymes and proteins with the matching prosites and the resulted high sequence similarity
Implemented a web service tool MetNetAligner which can be used for predicting unknown pathways, comparing and finding conserved patterns, and resolving ambiguous identification of enzymes. The tool supports two alignment algorithms.
Published papers: Recomb Satellite'07 paper, BIBE'07 paper , BIBE'07 slides , ISBRA'07 , CSBS 2007
Tool Website available.
iC2mpi : A Platform for Parallel Execution of Graph-Structured Iterative Computations (Paper in PDSEC'07 ) (CSc6310 course project)
The project aims to study the parallelization of sequential programs. A unique proof-of-concept prototype platform (iC2mpi) has been proposed for parallelization of a class of applications that have similar computational structure, namely graph-structured iterative applications such as the time-stepped simulation of battlefield management. I
Studied static and dynamic load balancing in distributed computing;
Implemented Battlefield simulation based on iC2mpi;
MANET : Self Management Routing of Mobile Sensor Network (Paper in SpringSim'06 Slides in SpringSim'06 Paper in Studies in Computational Intelligence'06 ).
Most existing routing protocols in mobile ad hoc networks mainly focus on addressing the problems of dense and sparse yet connected networks. However, the crisis-driven and geography-driven applications, such as those in ecology forests and modern battlefield, typically challenge current routing protocols in disjoint mobile sensor networks (DMSN) where network partitions can occur and last for a significant period.
In this project, I
Implemented an agent-based simulation for studying the problems of efficient route discovery in the disjoint networks
Proposed a novel autonomous messenger-based route discovery and routing protocol for disjoint clusters-based topology in DMSN
Proposed another two routing discovery and maintenance protocols which applied genetic fuzzy system and genetic algorithm and improved the performance
Software is available HERE.
CrayfishSim : A behavior-model simulation for natural ecology which is based on DEVSJAVA (Poster in SECABC05 )
In sea ecology, there exists the mechanism which decides the movement of prey and predators with the dynamics of the environment. The mechanism can be described as a network of mutually inhibiting centers for specific behaviors. This research aims to study behavioral choice models which account for more of the complexities of animal behavior, including changes in behavioral state, and formation of social dominance hierarchies.
In this project, I
Provided a heap-based implementation of discrete event simulation(DEVS)
Studied the behavior-model simulation and its application in sea ecology
Designing and implementing a GUI-based controller for the behavior-model simulation which also shows real time behavior analysis data in dynamic charts
Software is available HERE.
Optimization Algorithm (course project)
Code is available HERE
Report is available HERE
Protein Sequence Analysis (course project)
Obesity, as an increasingly serious public healthy disease, has been attracting more and more researchers. INSIG2, as an attractive candidate protein associated with obesity, has been studied recently. The structure prediction and verification have been significantly expected. So our project aim is to identify its structure by employing some popular bioinformatics tools such as PDB, BLAST, ClusterW, RASMOL and AMMP.
- Slides are available HERE
- Report is available HERE
Task Allocation in Real Team and Multi Robots (CSc8350 course project) ( .doc)
BAO Ontology Alignment
Resources (OWLapi OWLapi_doc org_class.zip)
..