About

Faruk is a 6th-year Ph.D. candidate in Machine Learning at Georgia Tech with expertise in signal processing, statistical detection-estimation, information theory, and private, verifiable AI for semantic communication. His journey began with an undergraduate internship where he developed an algorithm for the geo-localization of moving targets. This experience sparked an interest that extended into his undergraduate research, as he explored indoor, visible light-based statistical localization problems, eventually offering a solution through a database-assisted nonlinear least squares algorithm.

Faruk’s first Ph.D. project involved enhancing communication and privacy in Federated Learning using information-theoretic bounds as a tool. More recently his focus has shifted to semantic communication, which moves away from the traditional paradigm of faithful signal reconstruction and asks instead what a receiver needs to reason and act correctly, and why a given selection of information is the right one. The motivating problem is a gap between two traditions. Deep learning approaches to communication are bandwidth-efficient but operate as opaque encoders, which makes them ill-suited to safety-critical systems that must justify their decisions. Symbolic frameworks are transparent and verifiable but have lacked a scalable theory of communication.

Faruk’s work bridges the two by grounding semantic information in first-order logic. He has built theoretical basis for semantic content, entropy, and mutual information, together with the communication protocols that follow from them — point-to-point, distributed, and goal-oriented — with convergence guarantees and tractable algorithms for state spaces that are otherwise doubly exponential. The result is a neuro-symbolic foundation for efficient and trustworthy AI in autonomous systems, intelligent infrastructure, and 6G wireless networks. Currently, he is working on to develop a framework for systems to communicate logical meaning to enable AI agents to make effective, rational, and transparent decisions with less data.

Faruk is results-driven and willing to take calculated risks on difficult problems. Fellowship, mentoring, and teaching roles have sharpened his leadership and communication skills, particularly his ability to translate technical complexity into clarity and to navigate ambiguous, high-stakes situations.

Originally from Turkey, Faruk pursued his studies in Ankara and has extensively traveled across Europe, including Ukraine. Today, he is happily settled in the vibrant city of Atlanta. A firm believer in balance, Faruk embraces a lifestyle that includes regular workouts, outdoor adventures, and meaningful contributions to his community.

EDUCATION

Ph.D. in Machine Learning
2021 – present
Georgia Institute of Technology
Atlanta, GA, USA

M.S. in Electrical and Computer Eng. 2021 – 2023
Georgia Institute of Technology
Atlanta, GA, USA

B.S. in Electrical and Electronics Eng. 2016 – 2021
Bilkent University
Ankara, TURKEY

teaching

Graduate Teaching Assistant
2021 – 2023
Georgia Institute of Technology
Atlanta, GA, USA

“ECE3005: Professional and Technical Communications for ECE“

experience

Georgia Institute of Technology | Atlanta, GA | 2021 – Present
Graduate Research Assistant
Sensing, Learning, Communication and Decision Intelligence (SENTINEL) Lab (https://fekri.ece.gatech.edu)

– Designed an FOL-based semantic communication system for LLM-guided neuro-symbolic decision-making in collaborative autonomous driving, with budget-constrained uplink/downlink data selection and roadside deduction of safety and right-of-way conclusions over FOL-encoded California Driver Handbook rules. Introduced a random-support Dirichlet–Categorical model of inductive logical probability, a modern statistical reinterpretation of Carnap’s and Hintikka’s work on semantic information. Evaluated on the MDrive benchmark in CARLA. [INFOCOM ’26]

– Formulated goal-oriented semantic communication over goal states, the logical equivalence classes of world states induced by task rules, defining semantic content entropy, relative entropy, and mutual information over them and expressing the objective equivalently through semantic rate–distortion theory and the semantic information bottleneck. Devised a polynomial-time lexicographic sorting algorithm that bypasses the doubly-exponential state space. Evaluated in the LogiCity urban simulator. [GLOBECOM ’26]

– Developed a mathematical theory of semantic information and communication for logical inference, separating the physical entropy of transmitted symbols from the content they signify. Derived a content measure calibrated to language granularity, making content-informativeness comparable across systems built on different FOL ontologies, and made the world state space tractable through dynamically constructed evidential sub-languages. [WCNC ’26]

– Framed message selection for hypothesis deduction as posterior alignment between transmitter and receiver over the world state space, reducing it to a constrained resource allocation problem solved by truncated water-filling, with global optimality established through permutation invariance and convergence rates derived for the multi-round setting. [GLOBECOM ’25]

– Extended lossy semantic communication to distributed networks whose nodes hold partial, overlapping views of the world and exchange content-informative FOL sentences through a central server. Proved convergence of node posteriors toward the true distribution over the state space and derived a PAC bound on the sample complexity on the convergence. [ICMLCN ’25]

– Designed the point-to-point lossy semantic communication scheme for deducing the state of the world when the transmitter has no knowledge of the receiver’s downstream task. Made the exponentially large state space tractable by recasting inductive logical probability as Boolean model counting over a tree-decomposition SAT model counter. Validated on the FOLIO dataset. [WCNC ’25]

– Introduced attribute-based privacy to the federated learning literature under an NSF and Intel-funded project, formulating a lightweight framework that provides a more granular accuracy–privacy trade-off than differential privacy. [SPAWC ’23]

Intel Corporation | Atlanta, GA (Remote) | May 2022 – August 2022
Graduate Technical Intern

– Developed a Generalized Information Bottleneck privacy framework for NSF & Intel’s MLWiNS project, using generative adversarial minimax optimization to produce informative yet private transformations of input data in federated learning. Transferred the technology to Intel.

– Formulated attribute-based Differential Privacy (AB-DP), relaxing DP’s assumptions so that guarantees hold over a chosen subset of input attributes and shifting the privacy–accuracy trade-off toward accuracy, and established its compatibility with standard DP.

– Implemented in TensorFlow and PyTorch, integrated with the Intel OpenFL federated learning platform, and benchmarked against Meta Opacus on Colored MNIST.

Bilkent University | Ankara, Turkey | May 2021 – August 2021
Undergraduate Research Assistant

– Designed a novel visible light-based positioning algorithm for indoor non-line-of-sight environments (i.e., self-guided museum tours, indoor infant tracking) by fusing two existing frameworks in the literature, fingerprinting and non-linear least squares, to improve localization performance.

– Showcased algorithms superiority through MATLAB experimentation. Summarized findings in the form of a technical report and published on arxiv.

Meteksan Defence | Ankara, Turkey | July 2017 – August 2017
Summer Intern
_{Meteksan is a Turkish defense corp. that produces military radios and electronic warfare systems.} 

– Designed a printed log-periodic dipole antenna that can operate in the frequency range 1.5-2.5 GHz using CST software. Tested antenna with a network analyzer and presented the results to entire department.

– Received a return offer for a full-time position upon successful completion of internship.

TUBITAK Bilgem Iltaren | Ankara, Turkey | June 2016 – July 2016
Summer Intern
_{Tubitak is the Scientific and Technological Research Council of Turkey, a government research institution} 

– Created a simulation program in Matlab to geolocate a moving object using TDOA and AOA-based geometric localization techniques under various levels of noise and conducts error analysis to determine estimate reliability. Showcased the results to entire R&D department and intern cohort.

– Received a return offer for a full-time position upon successful completion of internship.

Honors & awards

Qualcomm Fellowship, 2022
– Awarded to students specializing in Wireless Comms. to support advancements in the field.

GT LEAD Leadership Fellowship, 2022 – 2023
– Endowed to exemplary students within GT community to bolster their leadership development.

Outstanding Success Scholarship, 2016 – 2021
– Granted by Turkish Education Association to students with distinct vision, character, and success to reinforce their college education.

Comprehensive Scholarship, 2016 – 2021
– Given to successful students. Covers tuition, housing, and includes a stipend.

298th in University Placement Exam, 2016
– In a cohort of 2+ million students.

High Honor Roll (Undergraduate), 2016 – 2021
– All semesters.

Trainings & certificates

QPR Gatekeeper Certificate, 2026
– Question, Persuade, Refer framework-based suicide prevention training.

Engaging with Different Personalities, 2019
– Turkish Education Association-led seminars to establish successful professional relationships with different personalities.

Ericsson Professional Foundation Program, 2019
– Series of educations on computer networks, 5G, AI, Machine Learning, Data Science, Cloud, Automation, Blockchain, IoT, Communications.

Rotary Foundation (Turkish Branch), 2018
– Attended Leadership Development camp.