With the emergence of 5G networks and the anticipation of 6G networks, the communication performance requirements are becoming increasingly stringent, and the networks are approaching Shannon's theoretical capacity limits. Therefore, a new communication paradigm has emerged, namely, semantic communication. Although the term itself is not new, research in this field has increased recently due to the need of sending huge amounts of data with limited communication resources and due to the emergence of Artificial Intelligence and the increasing computational capabilities of the communicating devices.

In the past few years, research in the field of semantic communication systems has tackled the different approaches of semantic communication, different types of data and how they can be transmitted semantically, and also the challenges and limitations of semantic communication systems were highlighted in many works. However, semantic communication research is still in its early stages and it is still not clear whether we should adopt this paradigm when designing a new communication system. Additionally, the reliability of semantic communication systems is still not thoroughly investigated and compared with traditional communication systems. Finally, most of the currently proposed semantic communication systems are based on machine learning models that are trained on specific datasets; this does not guarantee their generalization capability.

Therefore, the main objective of this project is to investigate techniques to enhance the reliability and generalization of semantic communication techniques and to compare the performance of networks with and without the use of semantic communication systems.

Figure: Reliable semantic communication framework for 6G and beyond networks.

The transition toward 6G wireless networks demands new approaches for efficient data transmission. As current systems approach theoretical limits for raw data compression, transmitting actionable knowledge instead of traditional bit level data becomes essential. Legacy communication frameworks based on Shannon separation principles face severe challenges with high dimensional machine observations. Increasing compression discards critical semantic features and pushes error correcting codes to their limits, causing communication to collapse in what is known as the digital cliff effect.

This project addresses this bandwidth limitation by shifting the objective from pixel level reconstruction to task relevant knowledge preservation. By leveraging Deep Joint Source Channel Coding (Deep JSCC), the research demonstrates how autoencoders can jointly optimize compression and channel robustness in a single end to end process. This approach leads to resilient knowledge transmission that enables autonomous agents to maintain high performance even as channel conditions degrade compared to traditional techniques.

The project extends the evaluation of Deep JSCC by characterizing bottleneck configurations under 6G Rayleigh fading channels, specifically benchmarking convolutional neural networks (CNN) and Vision Transformer (ViT) architectures against traditional JPEG source and Low-Density Parity Check (LDPC) channel encoders. By measuring reward preservation for remote reinforcement learning agents across these varied conditions, this study identifies the most effective architectures for ensuring reliable goal-oriented communication.

Figure: Goal-oriented semantic communication with Deep JSCC.

The rapid emergence of extended reality (XR) applications within the metaverse imposes stringent requirements on future wireless networks, including ultra-low latency, high data rates, and reliable quality of experience (QoE). At the same time, ensuring secure transmission of immersive XR data introduces additional overhead that can negatively impact network performance.

This project proposes a cross-layer optimization framework for secure XR data transmission in 6G-oriented metaverse environments, jointly considering communication efficiency, QoE, and security requirements. The proposed approach integrates adaptive bitrate control, dynamic compression strategies, and selective encryption mechanisms into a unified decision framework.

A reinforcement learning (RL)-based agent is developed to intelligently adjust transmission parameters based on real-time network conditions and user motion dynamics. The system aims to balance competing objectives, including latency, packet loss, visual quality, and security overhead, by learning optimal policies for resource allocation and protection levels. The framework is evaluated using realistic XR traffic datasets and simulated network conditions.

Figure: Secure XR pipeline with reinforcement learning for 6G metaverse.

The rapidly evolving landscape of wireless communication, particularly with the advent of 5G and future 6G networks, demands innovative approaches for efficient resource management. As networks become increasingly complex, the ability to optimize resource allocation in real-time while maintaining Quality of Service (QoS) across diverse service types becomes paramount.

This project focuses on addressing the issue of convergence acceleration in RL for wireless resource allocation. By leveraging expert knowledge transfer, we demonstrate how policies learned in specialized environments (such as eMBB, URLLC, and mMTC) can be reused by a learner agent to accelerate its convergence. This process leads to more efficient decision-making, reducing the time required to achieve optimal resource allocations.

Building on this foundation, the project extends the application of multi-agent reinforcement learning (MARL) by introducing a competency-based coordination model. This model dynamically adjusts the roles of agents based on their competencies, ensuring that each agent contributes in the most effective manner for its level of expertise.

Figure: Competency-based multi-agent RL for radio resource management.

This study provides a controlled evaluation of modern RL algorithms for adaptive compression in latency-sensitive telesurgical communication systems. Telesurgery systems require stable, low-latency video transmission that remains synchronized with latency-critical haptic and control feedback. However, wireless bandwidth in realistic environments is inherently time-varying, making static compression strategies inadequate.

When video bitrate approaches or exceeds available channel capacity, delay can disrupt real-time synchronization. This work investigates reinforcement learning (RL) for adaptive video compression under stochastic wireless capacity conditions. A simulation framework models regime-based bandwidth variations, with delay computed as a nonlinear function of bitrate and instantaneous capacity.

Video quality is represented using a PSNR-based proxy that distinguishes compression-induced distortion from channel-related degradation. Two continuous-control reinforcement learning algorithms (Proximal Policy Optimization, Soft Actor-Critic) and one discrete-action algorithm (Deep Q-Network) are evaluated and compared in terms of convergence, delay–quality tradeoff, and stability under bandwidth fluctuations.

Figure: RL-based adaptive compression for telesurgery under stochastic bandwidth.

This project aims to develop an advanced AI-driven peptide engineering and validation platform to target anti-apoptotic BCL-2 family proteins in colorectal cancer (CRC). Colorectal cancer remains a major global health challenge, largely due to resistance to therapy caused by dysregulated apoptosis. Overexpression of BCL-2 proteins allows cancer cells to evade cell death, highlighting the need for innovative therapeutic strategies.

To address this, the proposed research integrates artificial intelligence, computational modeling, and experimental validation to design novel pro-apoptotic peptides capable of restoring programmed cell death in cancer cells. The platform will utilize cutting-edge AI approaches, including generative modeling and inverse protein folding, followed by molecular docking, molecular dynamics simulations, and binding free energy analysis to identify high-affinity peptide candidates.

Selected peptides will be further optimized and experimentally validated using colorectal cancer cell models through assays such as MTT, flow cytometry, Western blotting, and biophysical techniques. Beyond developing novel therapeutic candidates, the project aims to establish a scalable and reproducible AI-driven drug discovery framework in Qatar, contributing to national biopharmaceutical innovation.

Figure: AI-powered generative design & validation pipeline (Phase 1–4) for pro-apoptotic peptides.

This project aims to evaluate Torosanin as a potential multi-target inhibitor for colorectal cancer (CRC) using an integrated computational approach. Given the complexity of CRC and the involvement of multiple dysregulated pathways, a multi-target strategy is essential for effective therapy.

The study will employ network pharmacology and bioinformatics analyses to identify Torosanin-associated targets and key CRC-related genes, followed by protein-protein interaction (PPI) network construction and GO/KEGG pathway enrichment to uncover underlying mechanisms. Subsequently, molecular docking will be performed to assess binding affinity with selected targets, and molecular dynamics (MD) simulations will evaluate the stability and dynamics of protein-ligand complexes.

Further analyses, including RMSD, RMSF, Rg, SASA, hydrogen bonding, and MM/GBSA or MM/PBSA binding free energy calculations, will be conducted to validate interactions. Additionally, physicochemical and ADMET predictions will assess drug-likeness and safety. Overall, this study will provide mechanistic insights into the multi-target anti-cancer potential of Torosanin and establish a computational foundation for future experimental validation and drug development.

Q-VISION is a three-year research program developing a real-time AI framework to predict and mitigate post-operative complications in robot-assisted nephrectomy for kidney cancer. The project is funded by the Qatar Research, Development and Innovation (QRDI) Council under grant ARG01-0522-230266, with Hamad Medical Corporation (HMC) as the lead institution, in collaboration with Qatar University (QU) and Hamad Bin Khalifa University (HBKU).

Our research integrates scene segmentation, surgical vision-language reasoning, and temporal workflow analysis into a unified clinician-facing platform, with two translational assets currently in active development: ‘SurgXpert’, a browser-based inference platform, and ‘SurgPixel Studio’, a boundary-aware annotation tool powered by our CVPR 2026 main-track work.

Outputs so far include multiple peer-reviewed publications spanning CVPR, MIDL, ECAI, CBMS, and CASE (including Qatar-based first-authorships at flagship international venues), and an ongoing active clinical validation study on private HMC nephrectomy data. The overall aim is development of a complication-prediction prototype aligned to HMC’s clinical deployment pathways.

Figure: Real-time surgical AI framework for kidney cancer surgery.