Wednesday Research Seminar Series - Bridging the Gap: Combining End-to-End Learning with Closed-Form Solutions in Deep Learning

We are pleased to invite you to our Wednesday Research Seminar. It will be held online on 30th October from 4pm via Microsoft Teams platform. Wednesday Research Seminar Series was launched in 2008 and has featured more than 350 presentations to date. The seminars provide a forum for researchers to share their work. Presenters include faculty from Middlesex University Dubai and other universities in the United Arab Emirates, as well as researchers from other global institutions. Murtaza will deliver seminar on:

“Bridging the Gap: Combining End-to-End Learning with Closed-Form Solutions in Deep Learning”

Murtaza Taj

Abstract

In recent years, deep learning has largely shifted toward end-to-end learning frameworks, where models are trained to directly map inputs to outputs, often bypassing the need for traditional analytical methods. This approach has revolutionized fields like computer vision, natural language processing, and autonomous systems. However, before the rise of neural networks, much of the research in machine learning and related disciplines focused on deriving closed-form solutions — elegant, interpretable mathematical formulations that could solve problems directly. This talk will explore how these two paradigms, often viewed as competing approaches, can complement each other to create more robust, interpretable, and efficient models. We will discuss how incorporating domain knowledge and closed-form solutions into modern deep learning frameworks can enhance performance, and lead to models that are both accurate and interpretable. Key topics will include integrating camera projection and 3D reconstruction equations into DNNs for camera calibration, embedding stereo vision formulations for accurate building height estimation, and applying urban planning constraints to refine house boundary extraction.

Presenter Bios

Dr. Murtaza Taj is an Associate Professor at the Syed Babar Ali School of Science and Engineering at Lahore University of Management Sciences (LUMS), Pakistan, where he leads the Computer Vision and Graphics Lab. He also serves as Co-PI at the National Agriculture and Robotics Lab (NARL) and has been Director of Technology for the People Initiative (TPI). He holds a Ph.D. and M.Sc. in Electronic Engineering and Computer Science from Queen Mary University of London (QMUL), UK. Dr. Taj’s research centers on computer vision, machine learning, and digital heritage, with a focus on optimizing neural networks for real-time applications in agriculture, forestry, and wildlife. His work has advanced early warning systems and digital preservation techniques, leveraging cutting-edge imaging and AI technologies. He co-founded Ingrain.io and Groopic Inc., where he led R&D efforts in image processing and pattern recognition.

With over 60 publications in top journals and conferences, Dr. Taj has received several prestigious awards, including the CyArk Summit Award in Education, the Times Higher Education Award for Excellence and Innovation in the Arts, and the 2024 Asia-Pacific Triple-E Runner-up Award for Impactful Research Team of the Year.