Virtual Public Seminar Series

Introducing the biweekly CREATE DAV Virtual Public Seminar Series.
The series will begin on November 4, 2020 and run live on Zoom from10:00 to 11:00 am, covering a range of topics in the field of Data Analytics and Visualization. Postgraduate and postdoctoral program trainees from York University and OCAD University will present their research in the following areas.

  • Machine Learning
  • Data Mining
  • Signal Processing
  • Computer Vision
  • Geomatics Engineering
  • Computer Graphics
  • Virtual Human Modeling
  • Serious Games
  • Natural Language Processing
  • Human Perception & Cognition
  • Visualization & Design

The seminars are designed for undergraduate students interested in pursuing industry-relevant research in big-data science and/or visualization.

A Certificate of Attendance will be awarded for each presentation.

Please register on Eventbrite for the Zoom link.

2020 SCHEDULE
November 4

Speaker: Dr. Nicola Bragazzi, Postdoctoral Fellow, York University

Topic: Artificial intelligence and Big Data analytics of COVID-19

Abstract: SARS-CoV-2 is a novel coronavirus, responsible for the still ongoing COVID-19 pandemic. Thanks to the latest advancements in the field of molecular and computational techniques and information and communication technologies, artificial intelligence (AI) and Big Data analytics can help in handling the huge, unprecedented amount of data derived from public health surveillance, real-time epidemic outbreaks monitoring, trend now-casting/forecasting, regular situation briefing and updating from governmental institutions and organisms, and health facility utilization information. The present review is aimed at overviewing the potential applications of AI and Big Data in the global effort to manage the pandemic, with a particular focus on the use of radiological images (chest radiograph and chest computed tomography). Several AI algorithms have been implemented, including Deep Learning (DL) based architecture models in order to quickly and automatically diagnose COVID-19, identifying the disease-associated patterns (such as ground glass opacity and lung consolidation). The objective of our presentation is to provide an updated, comprehensive synthesis of AI-based applications that could potentially help health personnel in the diagnosis, treatment and control of the spread of the disease.

November 18

Speaker: Ms. Sara Mozafari, MFA candidate, OCAD U

Topic: How to visualize the traffic light impact on the trains’ timing on Leslie Station to Longview Station of Eglinton Crosstown Line LRT using ArcGIS?

Abstract: The initial goal of the project was to visualize the train path of Leslie Station to Longview Station of Eglinton Crosstown Line LRT utilizing ArcGIS Pro to show various traffic light situations that could impact the timing of the train. However, COVID-19 forced us to work remotely, which created a critical impact on our plans. Not having access to proper computers of OCAD University or SNC-Lavalin, we could not run the ArcGIS Pro on our laptops. To deliver the tasks in a valuable way for the team, we researched alternative resolutions and came across the ArcGIS Maps for Adobe Creative Cloud solution. As a result, we defined a new workflow based on our limitations and what we had access to. In this presentation, I will take you through all the steps we took to deliver our tasks considering all the conditions. I will also present what we have learned through our internship at SNC-Lavalin that created new opportunities for us.

December 2

Speaker: Dr. Mufleh Al-Shatnawi, Postdoctoral Fellow, York University

Topic: Distributed deep learning for video processing

Multiple Pedestrian Tracking Based on Modified Mask R-CNN and Enhanced Particle Filter using an Adaptive Information Driven Motion Model

Abstract: In the recent years, multiple pedestrian tracking (MPT) has been one of the most important components in a wide range of applications in computer vision, such as video surveillance, traffic monitoring, and sports analysis, to name a few. In these applications, the scene is in continuous motion hence typical tracking systems that are using background modeling and handcrafted features fail to detect pedestrians efficiently. Furthermore, the scene in these applications shifts between random and continuous pedestrian motion. Most of the existing MPT algorithms based on particle filters assume that the motion of pedestrians is mostly or piecewise linear and predictable. Hence, these tracking algorithms adopt a linear constant velocity motion model for pedestrian tracking. However, the motion of some pedestrians is highly dynamic, as they are often stopping, moving backward, or turning around in real-world surveillance video. To overcome these problems, we propose an approach for multiple pedestrian tracking that can be divided into two main components: detection and tracking. For the detection component, we combine novel post-processing steps with the Mask Region Convolutional Neural Network (Mask R-CNN) to identify multiple pedestrians in a given video frame. For the tracking component, we propose a robust MPT algorithm using enhanced particle filtering with an adaptive information driven motion model and resampling scheme. The proposed tracking algorithm is suitable for online and real-time applications. Since data association is a key issue in tracking-by-detection schemes, we propose a combination between an efficient adaptive information driven motion model and a new resampling scheme that retains information pertaining to the weighted particles during the particle propagation and resampling steps. Experimental results show the benefits of using the proposed post-processing steps and the adaptive information driven motion model for detecting and tracking pedestrians with unpredictable movements. Moreover, the tracking accuracy and precision are significantly improved, and the number of tracker identification (ID) switches is reduced simultaneously.

December 16

Speaker: Mr. Amin Omidvar, PhD candidate, York University

Topic: How to design a neural network for different supervised machine learning tasks

Abstract: If we don’t know how to properly design the last layer of a neural network model, our neural network may not work, and we won’t be able to solve our supervised machine learning problem properly using Neural networks. If our data has labels, we are probably going to solve one of the supervised machine learning tasks. These supervised tasks are binary classification, multi-class classification, multi-label classification, regression, and Ordinal Regression. It is important to know which one of these categories is related to our machine learning task since the activation function and the number of neurons in the last layer of the neural network may be different based on what kind of machine learning problems we are going to solve. In this presentation, I will introduce each one of these supervised machine learning tasks briefly and I will show you how we should design a feedforward neural network model to solve each of them.

2021 SCHEDULE
January 6

Speaker: Ms. Mahta Shafieesabet, MSc candidate, York University

Topic: ML with Graphs: Graph Neural Networks

Abstract: Researchers in network science have traditionally relied on user-defined heuristics to extract features from complex networks (e.g., degree statistics or kernel functions). However, recent years have seen a surge in approaches that automatically learn to encode network structure into low-dimensional embeddings, using techniques based on deep learning and nonlinear dimensionality reduction. These network representation learning (NRL) approaches to remove the need for painstaking feature engineering and have led to state-of-the-art results in network-based tasks, such as node classification, node clustering, and link prediction.

In this tutorial, we will cover key advancements in NRL over the last decade, with an emphasis on fundamental advancements made in the last two years. We will discuss very recent advancements in graph neural networks. Techniques for deep learning on network/graph-structured data (e.g., graph convolutional networks and GraphSAGE).

January 20

Speaker: Ms. Zahra Arjmandi, PhD candidate, York University

Topic: Towards Autonomous Visual Inspection of Large-Scale Infrastructures Using Aerial Robots

Abstract: The goal of this project is developing an autonomous, unmanned aerial photogrammetry system to inspect hard-to-reach vertical infrastructures. Firstly, we need to develop a precise navigation and positioning system that can work even in GPS denied environments. We used the ultra-wideband (UWB) sensor to build the radio frequency-based positioning system. For localization, we implemented multilateration (MLAT) technique which is a positioning technique based on range. For a more robust solution in the ML algorithm, we used Levenberg–Marquardt non-linear minimization algorithm. We tested offline localization in various environments including outdoor, indoor and under bridges by mounting UWB and IMU sensors on the drone. The accuracy of the estimated location was validated with data collected by the total station. The method is also capable to perform online positioning.

February 17

Speaker: Mr. Zachary McCarthy, PhD candidate, York University

Topic: Infectious disease transmission: insights from mathematical modelling and data synthesis

Abstract: Infectious diseases pose threats to both public health and the economy. Insights into disease transmission can be used to devise control strategies and inform decision-making to mitigate the burden. In this regard, data analysis and visualization techniques are used to leverage a spectrum of disease surveillance and clinical data. Also, mathematical models of disease transmission often integrate both data sources and provide a critical aid to response efforts. Models offer a unique utility in that they can provide key metrics, such as reproduction numbers, predict disease trajectories and weigh the impacts of different control measures. We will provide several instances which utilize data analysis and visualization techniques, with a focus on model-based analyses, in the context of the COVID-19 pandemic. Each of these techniques have played an instrumental role in informing the response to COVID-19 through informing public health policy.

March 3

Speaker: Mr. Muhammad Kamran, PhD candidate, York University

Topic: Boundary regularized building footprint extraction from satellite images using deep neural networks

Abstract: In recent years, an ever-increasing number of remote satellites are orbiting the Earth which streams vast amount of visual data to support a wide range of civil, public and military applications. One of the key information obtained from satellite imagery is to produce and update spatial maps of built environment due to its wide coverage with high resolution data. However, reconstructing spatial maps from satellite imagery is not a trivial vision task as it requires reconstructing a scene or object with high-level representation such as primitives. For the last decade, significant advancement in object detection and representation using visual data has been achieved, but the primitive-based object representation still remains as a challenging vision task. Thus, a high-quality spatial map is mainly produced through complex labour-intensive processes. In this paper, we propose a novel deep neural network, which enables to jointly detect building instance and regularize noisy building boundary shapes from a single satellite imagery. The proposed deep learning method consists of a two-stage object detection network to produce region of interest (RoI) features and a building boundary extraction network using graph models to learn geometric information of the polygon shapes. Extensive experiments show that our model can accomplish multi-tasks of object localization, recognition, semantic labelling and geometric shape extraction simultaneously. In terms of building extraction accuracy, computation efficiency and boundary regularization performance, our model outperforms the state-of-the-art baseline models.

March 17

Speaker: Mr. Tilemachos Pechlivanoglou, PhD candidate, York University

Topic: Machine learning with embeddings: word2vec and node2vec

Abstract: Word2vec is a technique for natural language processing. The word2vec algorithm uses a neural network model to learn word associations from a large corpus of text. Once trained, such a model can detect synonymous words or suggest additional words for a partial sentence. As the name implies, word2vec represents each distinct word with a particular list of numbers called a vector. The vectors are chosen carefully such that a simple mathematical function (the cosine similarity between the vectors) indicates the level of semantic similarity between the words represented by those vectors.
Node2vec is a similar technique based on word2vec, producing vector representations of objects in a graph.  Prediction tasks over nodes and edges in networks require careful effort in engineering features used by learning algorithms.

The node2vec framework learns low-dimensional representations for nodes in a graph by optimizing a neighborhood preserving objective. The objective is flexible, and the algorithm accomodates for various definitions of network neighborhoods by simulating biased random walks. Specifically, it provides a way of balancing the exploration-exploitation tradeoff that in turn leads to representations obeying a spectrum of equivalences from homophily to structural equivalence.

Learning useful representations from highly structured objects such as text or networks is useful for a variety of machine learning applications.  Besides reducing the engineering effort, these vector representations can lead to greater predictive power.

March 31

Speaker: Mr. Chenxing Zheng, MSc candidate, York University

Topic: How machine understands language – an introduction to word vectors

Abstract: Deep learning techniques have made significant progress in computer vision and natural language processing (NLP). Nowadays machines can outperform human beings in a number of tasks, such as image recognition, sentiment analysis and even reading comprehension. How could this be possible? We know that everything that a computer can handle is made up of numbers. For example, most pictures are stored as 3-dimensional matrices (Height, Width, Channel), and object recognition in pictures is to look for patterns in those numbers behind these pictures. What about languages? How is a piece of text converted into its digital representation?

In this presentation, I would like to take you on an exploration of one of the fundamental techniques in NLP – Word Embeddings, which turn a basic unit of language – a word – into the machine-favored vector form. The learned vectors can explicitly encode interesting linguistic regularities and patterns. Many of these patterns can be represented as linear translations, perhaps the most famous example is the equation vec(‘king’) – vec(‘man’) = vec(‘queen’) – vec(‘woman’). I’ll first cover some of the classical models, such as CBOW and Skip-Gram, then all the way to the contextual word embedding used in the advanced transformer models.

April 14

Speaker: Ms. Melissa Kremer, PhD candidate, York University

Topic: Modelling Distracted Agents In Crowd Simulations

Abstract: Multi-agent simulations are a useful tool in studying the movement of pedestrians in arbitrary environments for the purpose of architectural design and analysis. Traditional methods have largely employed homogeneous agents in terms of physical and sensory abilities as well as behaviours. However, the fidelity of simulations can benefit from agents exhibiting a wide array of behaviours, sensory abilities, and locomotion. The inclusion of such agents can aid in the validity of simulation analyses, and better reflect crowds in the real world. This presentation will describe a method for modelling agents using cellphones in crowds in a commonly used crowd simulation steering algorithm, including the modelling of navigation error arising from distracted behaviour as well as visual attention. Finally, the impact of the different aspects of this model on crowd flow statistics and virtual crowd creation will be discussed.

April 28

Speaker: Shima Khoshraftar, PhD candidate, York University

Topic: Dynamic Graph Embedding via LSTM History Tracking

Abstract: Many real world networks are very large and constantly change over time. These dynamic networks exist in various domains such as social networks, traffic networks and biological interactions. To handle large dynamic networks in downstream applications such as link prediction and anomaly detection, it is essential for such networks to be transferred into a low dimensional space. Recently, network embedding, a technique that converts a large graph into a low-dimensional representation, has become increasingly popular due to its strength in preserving the structure of a network. Efficient dynamic network embedding, however, has not yet been fully explored. In this work, we present a dynamic network embedding method that integrates the history of nodes over time into the current state of nodes. The key contribution of our work is 1) generating dynamic network embedding by combining both dynamic and static node information 2) tracking history of neighbors of nodes using LSTM 3) significantly decreasing the time and memory by training an autoencoder LSTM model using temporal walks rather than adjacency matrices of graphs which are the common practice. We evaluate our method in multiple applications such as anomaly detection, link prediction and node classification in datasets from various domains.

May12

Event has been canceled.