Projects

Heterogeneous Graph Representation Learning

This project tends to develop new neural-based models for heterogeneous graph representation learning. For instance, proposed neural architectures are able to perform downstream tasks on dynamic heterogeneous graphs, or to perform subgraph mining in heterogeneous graphs. The findings from this research project are published in top-tier conferences, e.g. CIKM’22, PVLDB’23.
Keywords: Graph Representation Learning, Graph Search, Team Discovery, Dynamic Heterogenous Graph

Team Formation

The focus of this project is on the composition of teams of experts from a collaboration network that would collectively cover a given set of skills. The team Formation problem as a use case scenario of the Graph Search problem faces many challenges. In addition to covering required skills, a group of experts should show good productivity, strong communication abilities among each other and other qualities. This makes Team Formation a challenging example of keyword search in a graph where in addition to keywords, there are multiple criteria to meet to find the target subgraph. Results and findings from this project were published in form of research papers in top-tier conferences e.g. CIKM’20, SIGIR’21, CIKM’21, EDBT’22.
Keywords: Team Formation, Graph Search, Heterogenous Graph, Subgraph Mining

PyTFL Project

Supervising the PyTFL project, an open-source end-to-end framework for graph search using neural networks. This framework uses the Team Formation problem as a case study to search the graph for a potential subgraph that addresses a given query. Different functionalities including data preparation, neural network models, training/testing pipelines, various benchmarks and fairness analysis are part of this toolkit. Along with the development, my role included supervising an undergrad student during the implementation.
Labs for Systems, Software and Semantics, Ryerson University
Keywords: Team Formation, Graph Search, NLP, Toolkit

Data Stream Forecasting

In this project we explored new techniques to develop an effective and accurate forecast system that operates on live data stream. This project led to a journal paper that can be found here.
Big Data Analytics Lab. Amirkabir University of Technology
Keywords: Concept Drift, Decision Trees, Data Stream, Random Forest

Indoor-Localization

This project intends to implement an AI system that can be an alternative for GPS, based on RSSI signals and IMU sensors for indoor places. Project succeeded in winning multiple awards by Presidential Deputy for Science and Technology.
M.S.P. research Lab. Amirkabir University of Technology
Keywords: Signal Processing, Wide-area Positioning, Indoor Positioning, Fingerprinting

Action Recognition

This project proposes a novel action recognition algorithm that utilizes the HMM model to recognize the position based on the spatial data obtained by the KINECT sensor. The result is published as a paper and can be found here.
Artificial Creatures Lab. Sharif University of Technology
Keywords: Human Action Recognition, Skeleton Data, Linear Discriminant Analysis (LDA), Hidden Markov Model (HMM), Fisherpose