Name of the Lab: Deep Learning Lab
Principal Investigator:
Dr. Faisal Shafait
Department Of Computing (DOC), School of Electrical Engineering & Computer Science (SEECS),
National University of Sciences and Technology (NUST)
Main Campus, Sector H-12, Islamabad.
Phone: +92-51-90852064, Email: faisal.shafait@seecs.nust.edu.pk
CO-Principal Investigators:
Muhammad Shahzad
Department Of Computing (DOC), School of Electrical Engineering & Computer Science (SEECS),
National University of Sciences and Technology (NUST)
Main Campus, Sector H-12, Islamabad.
Phone: +92-51-90852157, Email: muhammad.shehzad@seecs.nust.edu.pk
Muhammad Imran Malik
Department Of Computing (DOC), School of Electrical Engineering & Computer Science (SEECS),
National University of Sciences and Technology (NUST)
Main Campus, Sector H-12, Islamabad.
Phone: +92-51-90852357, Email: malik.imran@seecs.edu.pk
Muhammad Ali Tahir
Department Of Computing (DOC), School of Electrical Engineering & Computer Science (SEECS),
National University of Sciences and Technology (NUST)
Main Campus, Sector H-12, Islamabad.
Phone: +92-332-5405920 , Email: ali.tahir@seecs.edu.pk
Goal: Application of AI and deep learning to improve quality and safety of citizens
Scientific Objectives:
- Develop robust object detection algorithms using state-of-the-art deep learning frameworks
- Generalization from less data
Description of Lab:
The term Machine Learning is used to describe computer systems that model human cognitive and perceptual behavior. Machine learning is used to recognize objects from imagery, transcribe human speech into text, group similar news and opinions according to people’s interests, and learn customer behavior from large amounts of data. In the last decade, a new class of algorithms called Deep Learning has become so successful at performing the above-mentioned tasks that the term deep learning is increasingly becoming synonymous with artificial intelligence and machine learning.
The performance of previously popular machine learning techniques was limited because of their inability to handle real world abstractions. Thus, constructing input features from raw data required careful engineering and problem-domain expertise. These high-level features would then be given to a learning classifier, which could identify useful patterns and clusters in the input.
Basically, deep learning allows computational models to have multiple levels of abstraction to learn representations of data. As previously mentioned, these methods have dramatically improved the recognition performance of automatic speech recognition, computer vision, drug discovery, business analytics and genomics. If provided large amounts of data (even if in the raw form), deep learning discovers intricate structure in large data sets. The basic computation workhorse is the back-propagation algorithm, which indicates how a neural network should change its internal parameters in each layer from the representation in the previous layer. The research is deep neural networks is very dynamic nowadays, and new variants are emerging everyday e.g. feed-forward, convolutional, highway, recurrent neural networks, long short-term memory networks to name a few. Deep Convolutional Nets have revolutionized the processing of images, video and audio, whereas Long Short-Term Memory Networks have been used to tackle sequential data such as text and speech.
The Deep Learning Laboratory at the National Center for Artificial Intelligence will serve as the Brain of the center. It will not only conduct cutting edge research in deep learning, but also support all other labs of the center by providing them access to deep learning expertise to solve their problems. This lab will be equipped with the state-of-the-art hardware and software to do both theoretical and applied deep learning research and become a national leader in the field.
Applications Domains:
-
- Vehicle detection for road traffic analysis
- Person identification/re-identification for monitoring and activity analysis
- Natural language processing
- Urban mapping and land cover classification
- Forest monitoring
- Audiovisual saliency detection