• Developed and implemented HTTP-Based Adaptive Streaming for Mobile Clients using Markov Decision Process.

  • Investigation, development, and empirical experience of our model and achieving significant results in decreasing the bandwidth consumption of video streaming applications while increasing their perceptual quality.

Hypertext transfer protocol (HTTP) is the fundamental mechanic supporting web browsing on the Internet. An HTTP server stores large volumes of content and delivers specific pieces to the clients when requested. HTTP promises seamless integration of video delivery to existing HTTP-based server platforms. This is achieved by segmenting the video into many small chunks and storing these chunks as separate files on the server. For adaptive streaming, the server stores different quality versions of the same chunk in different files to allow real-time quality adaptation of the video due to network bandwidth variation experienced by a client. For each chunk of the video, which quality version to download, therefore, becomes a major decision-making challenge for the streaming client, especially in a vehicular environment with significant uncertainty in mobile bandwidth. In this project, we proved that for such decision-making, Reinforcement Learning (RL) techniques are superior to previously proposed non-RL solutions. We demonstrated how our model can achieve up to a 15x higher perceptual quality compared to well-known non-ML solutions when the RL has prior knowledge of the bandwidth model. This also led me to work on another project in parallel to find the correlations of network throughput and geo-locations which resulted in building a Geo-Adaptive video player. The figure below shows how different video quality components (i.e., average quality (AQ), the rate of quality changes (QC), and deadline miss (DM)) could be used for customizing a streaming experience in our model [15]:

The nonuniform sampling in the human visual system (HVS) is used in a video compression technique called foveation in which the region of interest (ROI) is given a higher bitrate. This technique can significantly reduce network traffic or provide higher quality with a similar bitrate. ROI or fovea region can be predicted using offline algorithms with a considerable prediction error. In a real-time video streaming scenario, although the fovea region can be detected precisely using an eye-tracker device, accessing this data is not possible on a real-time basis due to network latency. In this project, we developed a prediction model which used streaming clients’ gaze locations on a set of frames to predict the fovea region on future frames. As illustrated in the figure below, with this method, we achieved 10x higher prediction accuracy compared to the offline model [17]:

However, interacting functions such as high-quality zooming for online video streaming from cloud servers remained a challenge due to the intertwined relationships among video chunk lengths, the viewer’s fast-changing Region of Interest (ROI), and network latency. It is possible to utilize the Tiled Video technique and store picture tiles in separate files with their unique URLs on the media server with smaller chunk sizes. However, it introduces a significant burden on the network core due to increased total video length contributed by combined non-video bits from too many smaller chunks. To overcome this, we defined a new research project and proposed the use of edge computing to achieve high quality zooming function for video streaming. We introduced a system architecture using Tiled-DASH (T-DASH) video encoding on edge servers, and a novel ROI prediction method combining online and offline models on the client side. Our evaluations showed that a high level of ROI prediction accuracy is achieved by our approach, fulfilling a core condition for making the zooming function a reality.

Working on this project, motivated us to utilize UAVs to assist our considered networks. However, these devices come with numerous challenges and limitations such as limited onboard processing and power resources which opened our UAV studies pathway.

• Using Machine Learning and Image Processing models to increase the QoS of UAVs for different applications such as livestock monitoring and Smart Agriculture.

• Developed models to decrease the energy consumption of UAVs.

• Investigated different possibilities and challenges for remote energy transfer to the flying UAVs.

Despite the increasing popularity of commercial usage of UAVs or drone-delivered services, their dependence on the limited-capacity onboard batteries hinders their flight time and mission continuity. As such, developing in-situ power transfer solutions for topping-up UAV batteries have the potential to extend their mission duration. We studied a scenario where UAVs are deployed as base stations (UAV-BS) providing wireless Hotspot services to the ground nodes while harvesting wireless energy from flying energy sources. These energy sources are specialized UAVs (Charger or transmitter UAVs, tUAVs), equipped with wireless power transmitting devices such as RF antennae. tUAVs have the flexibility to adjust their flight path to maximize energy transfer. With the increasing number of UAV-BSs and environmental complexity, it was necessary to develop an intelligent trajectory selection procedure for tUAVs to optimize the energy transfer gain. In this project, we modeled the trajectory optimization of tUAVs as a Markov Decision Process (MDP) problem and solved it using the Q-Learning algorithm. As illustrated in the figure below, our simulation results confirmed that the Q-Learning-based optimized trajectory of the tUAVs outperformed two benchmark strategies, namely random path planning and static hovering of the tUAVs [7]:

My current biomedical research aims to bridge the gap in bioinformatics by developing computational models for large-scale data analysis of cancer and HIV genomes. Specifically, I have been working on a comprehensive toolkit to detect motif changes in viral and cancer genomes, including HIV and SARS-CoV-2, using machine learning methods. This toolkit helps researchers identify mutational signatures that are crucial for understanding disease progression, therapeutic resistance, and potential therapeutic targets. 

In my work with the Texas Biomedical Research Institute, I have been applying signal processing and machine learning to identify mutational signatures in cancer genomes. This research contributes significantly to the field by offering a computational approach that allows for more precise identification of mutagenic processes, which can improve personalized treatment strategies for cancer patients.

A major goal of my research is to build bioinformatics tools that can be used by scientists to continue their studies in greater depth. These tools will enable the identification of novel patterns within large genomic datasets, shedding light on the underlying mechanisms of diseases and potentially guiding the development of more effective treatments.

• Leading an interdisciplinary research team at Zeelamo Academy.

There are significant milestones in modern human civilization in which mankind stepped into a different level of life with a new spectrum of possibilities and comfort. From fire-lighting technology and wheeled wagons to writing, electricity, and the Internet, each one changed our lives dramatically. In this project, we took a deep look into the invasive Brain Machine Interface (BMI), an ambitious and cutting-edge technology that has the potential to be another important milestone in human civilization.  Not only beneficial for patients with severe medical conditions, but the invasive BMI technology can also significantly impact different technologies and almost every aspect of human life. We reviewed the biological and engineering concepts that underpin the implementation of BMI applications. There are various essential techniques that are necessary for making invasive BMI applications a reality. We reviewed these by providing an analysis of (i) possible applications of invasive BMI technology, (ii) the methods and devices for detecting and decoding brain signals, as well as (iii) possible options for stimulating signals into the human brain. At this beginning step of this project, we focus on the challenges and opportunities of invasive BMI and investigate how AI can play a more important role in our lives.