Muhammad Kamran Janjua

Art is Long; Time is Fleeting!

I am a Machine Learning Researcher @ Huawei, Canada working in Dr. Chao Gao's team, previously in Dr. Mohammad Salameh's team, at Edmonton Research Center. I work on online video, and motion understanding for various low-level vision tasks such as video restoration, and prediction.

Previously, I was a graduate student with Dr. Martha White at University of Alberta, where I worked on continual learning. Particularly, I worked on making online, and continual anticipatory predictions in real-time on non-stationary systems. My work is possibly one of the first to show the feasibility of true online TD prediction agent on a real-world industrial plant.

Before that, I worked as a Research Associate at Qatar Computing Research Institute (QCRI) working on interpretability in neural language models. Specifically, I worked on understanding how neural language models build (if at all) grammatical structure of language internally. I was supervised by Dr. Hassan Sajjad.

Before that I did my undergraduate from National University of Sciences & Technology (NUST), Islamabad working in TUKL-NUST R&D Lab working with Dr. Faisal Shafait. I have had the pleasure of visiting AaltoVision and AaltoML lab at Aalto University, Finland to work on estimating depth from stereo setups advised by Dr. Arno Solin. I have also worked as a research intern at ARTE Lab, University of Insubria back in 2018 where I worked on designing multi-modal neural networks, and was advised by Dr. Ignazio Gallo.

Fun Fact: My Erdős Number is 4.

Email  /  CV  /  Google Scholar  /  Github

I am interested in online, and continual learning from continuous streams of data (text/tabular/videos). Some (representative) papers are highlighted.

We propose Turtle to learn the truncated causal history model for online video processing. The causal design in Turtle enables recurrence in inference through state-memorized historical features while allowing parallel training by sampling truncated video clips. We report new state-of-the-art results on a multitude of video restoration benchmark tasks.

We present CascadedGaze Network (CGNet), an encoder-decoder architecture that employs Global Context Extractor (GCE), a novel and efficient way to learn global information for image restoration.

We show the importance of learning in deployment, by comparing a TD agent trained purely offline with no online updating to a TD agent that learns online. This final result is one of the first to motivate the importance of adapting predictions in real-time, for non-stationary high-volume systems in the real world.

We propose a method for fusing stereo disparity estimation with movement-induced prior information.

We propose a new measure SemanticMap to evaluate the performance of cross modal systems. Our proposed measure evaluates the semantic similarity between the image and text representations in the latent embedding space.

We propose a novel deep training algorithm for joint representation of audio and visual information which consists of a single stream network (SSNet) coupled with a novel loss function to learn a shared deep latent space representation of multimodal information.

In order to further enhance the discriminative capability of deep features, we introduce a joint supervision signal, Git loss, which leverages on softmax and center loss functions. The aim of our loss function is to minimize the intra-class variations as well as maximize the inter-class distances.

This super cool design (at least I find it to be cool) is taken from ✩