Abdelrahman Fahim Email & Phone Number

Riverside, Ca, Us

Riverside County, California, United States

* AcTrak: Controlling a Steerable Surveillance Camera using Reinforcement Learning (TCPS 23)– An RL agent dynamically balances the frequency of zooming out and schedules zoom-ins on existing targets.– Designing a simulator that mimics the camera, target movements, and other dynamics to enable the RL training.– Carefully designing and fine-tuning compact models to enable real-time camera operations without latency.– The trained agent is deployed on a real camera, and it acquires twice as many zoomed-in images as the best baseline.– Tags: Motion planning, Emulation, Reinforcement learning, surveillance cameras, embedded systems, robotics.◦ NET-SYN: Generating synthetic network traffic constrained by the topology (On going)– An emulation tool that generates synthetic network traces resembling realistic traces for a given network topology.– Since the networks community lacks simultaneous network traces and topology datasets, NET-SYN can bridge thisgap, allowing researchers to access this data.– Tags: Synthetic data generation, emulation, GenAI, GANs, conditional GANs, Networks◦ BigEye: Event Detection & Summarization (Textual and Visual) from distributed data dources (IoT-J 22)– BigEye is theoretically proven to achieve the same accuracy as the centralized case with far less fetched data– With SDN testbed deployments and crawled Twitter datasets, BigEye only needs 1% of raw at the central engine.– Tags: Distributed Systems, SDN Deployment, Network emulation, Data mining◦ FoB: Fair distributed training over a bandwidth constrained network (Publication under submission)– A novel federated learning formulation and approach to enable training in scenarios where fairness is a must.– FoB maximizes the minimum local accuracy (tested on a private validation dataset) to ensure fairness.– FoB intelligently allocates more bandwidth to struggling clients to achieve the fairness objective.– Tags: Federated Learning, personalization, Networks

(NLP + Generative AI + Network Security) DOLOS: Smart DNS exfiltration attacks. (EuroS&P2024) - Using GANs to encode exfiltrated data to deceive detectors while ensuring decodability. - With real cloud deployment, DOLOS achieves exfiltration rates 15x faster compared with best baselines.- Prompt engineering: Fine-tuning OpenAI GPT models to generate a dataset of benign DNS queries.

Redmond, Washington, United States

- Visualization (PowerBI): Multiple dashboards for Azure networking incidents (e.g., detection/ times).

Redmond, Washington, United States

(Recommendation systems + NLP) Auto diagnostics for Azure networks incidents- Designed and deployed a model (utilizing SVD +KNN) that provides recommendations for diagnostics.- In 78% of incidents, frontline engineers successfully find all relevant diagnostics to the given incident.

Al Jizah, Egypt

◦ MRTI-BT: Robust RFID tag identification for multi-RFID reader systems (IEEE Access Journal)– Utilizing passive sensing to expedite the parallel identification of tags and accelerate the identification process.– MRTI-BT leverages previous identification sessions to identify and correct bits that are received in error due to noise.– On average, MRTI-BT achieves 113% faster identification time in comparison with existing baselines.– Tags: RFID, Collisions, MAC protocols

Doctor Of Philosophy - Phd, Computer Science

Master'S Degree, Electrical, Electronics And Communications Engineering

Conducting research in tag identification in multi-reader RFID systems tackling the problem of reader-reader collision and tags collision.

Bachelor'S Degree, Electrical, Electronics And Communications Engineering