arXiv subjects

Bekir Sait Ciftler

Publications and source records attributed to Bekir Sait Ciftler.

10 recordsLinked to original sources

arXiv2026

The Gate Always Closes: On Injecting Auxiliary Signals into Frozen Vision-Language Models

Auxiliary signal pathways in VLMs are routinely fitted with learnable gates so the optimiser can decide how much of the signal to admit. We find that the optimiser almost always decides on zero: across five injection designs, every gated pathway becomes behaviourally closed, with accuracy invariant to ablating the pathway at inference even when the gate parameter would nominally pass 30-45% of the signal. We attribute this suppression phenomenon to two regimes, a dead-gradient regime formalised through the caption-invariance of image-derived signals, and a negative-utility regime in which the auxiliary signal actively hurts the loss. Rather than fight suppression, we exploit it: we regularise LoRA fine-tuning with geometric auxiliary losses from hyperbolic visual relational graphs (IoA-driven entailment cones and angular repulsion on the Lorentz manifold), coupled only through the forward pass at training time and dropped at inference. Disaggregating GQA by question type exposes a clean dissociation. Three configurations without geometric losses at inference lose 2.85-3.39pp on relational questions while gaining ~1pp on attribute questions; a fourth that trains with the losses but infers through a soft prompt loses 5.14pp on rel for only +0.23pp on attr, so training-time regularisation alone does not protect relational accuracy without a geometric inference pathway. Configurations that keep the geometric pathway at inference preserve vanilla-level relational accuracy and match the attribute gain. Out of distribution on VSR, the RMS-prefix recipe preserves the spatial signal; stripping the geometric losses (G2) collapses VSR by 4.6pp, isolating them as the OOD source. A secondary result: embedding-norm alignment is necessary for generation-safe prefix injection, and learnable gates should be replaced with fixed, non-optional injection at matched scales.

cs.CV↗

arXiv2022

Hierarchical Multi-Agent DRL-Based Framework for Joint Multi-RAT Assignment and Dynamic Resource Allocation in Next-Generation HetNets

This paper considers the problem of cost-aware downlink sum-rate maximization via joint optimal radio access technologies (RATs) assignment and power allocation in next-generation heterogeneous wireless networks (HetNets). We consider a future HetNet comprised of multi-RATs and serving multi-connectivity edge devices (EDs), and we formulate the problem as mixed-integer non-linear programming (MINP) problem. Due to the high complexity and combinatorial nature of this problem and the difficulty to solve it using conventional methods, we propose a hierarchical multi-agent deep reinforcement learning (DRL)-based framework, called DeepRAT, to solve it efficiently and learn system dynamics. In particular, the DeepRAT framework decomposes the problem into two main stages; the RATs-EDs assignment stage, which implements a single-agent Deep Q Network (DQN) algorithm, and the power allocation stage, which utilizes a multi-agent Deep Deterministic Policy Gradient (DDPG) algorithm. Using simulations, we demonstrate how the various DRL agents efficiently interact to learn system dynamics and derive the global optimal policy. Furthermore, our simulation results show that the proposed DeepRAT algorithm outperforms existing state-of-the-art heuristic approaches in terms of network utility. Finally, we quantitatively show the ability of the DeepRAT model to quickly and dynamically adapt to abrupt changes in network dynamics, such as EDs mobility.

eess.SP↗

arXiv2021

Deep Reinforcement Learning for Radio Resource Allocation and Management in Next Generation Heterogeneous Wireless Networks: A Survey

Next generation wireless networks are expected to be extremely complex due to their massive heterogeneity in terms of the types of network architectures they incorporate, the types and numbers of smart IoT devices they serve, and the types of emerging applications they support. In such large-scale and heterogeneous networks (HetNets), radio resource allocation and management (RRAM) becomes one of the major challenges encountered during system design and deployment. In this context, emerging Deep Reinforcement Learning (DRL) techniques are expected to be one of the main enabling technologies to address the RRAM in future wireless HetNets. In this paper, we conduct a systematic in-depth, and comprehensive survey of the applications of DRL techniques in RRAM for next generation wireless networks. Towards this, we first overview the existing traditional RRAM methods and identify their limitations that motivate the use of DRL techniques in RRAM. Then, we provide a comprehensive review of the most widely used DRL algorithms to address RRAM problems, including the value- and policy-based algorithms. The advantages, limitations, and use-cases for each algorithm are provided. We then conduct a comprehensive and in-depth literature review and classify existing related works based on both the radio resources they are addressing and the type of wireless networks they are investigating. To this end, we carefully identify the types of DRL algorithms utilized in each related work, the elements of these algorithms, and the main findings of each related work. Finally, we highlight important open challenges and provide insights into several future research directions in the context of DRL-based RRAM. This survey is intentionally designed to guide and stimulate more research endeavors towards building efficient and fine-grained DRL-based RRAM schemes for future wireless networks.

eess.SP↗

arXiv2021

DQN-Based Multi-User Power Allocation for Hybrid RF/VLC Networks

In this paper, a Deep Q-Network (DQN) based multi-agent multi-user power allocation algorithm is proposed for hybrid networks composed of radio frequency (RF) and visible light communication (VLC) access points (APs). The users are capable of multihoming, which can bridge RF and VLC links for accommodating their bandwidth requirements. By leveraging a non-cooperative multi-agent DQN algorithm, where each AP is an agent, an online power allocation strategy is developed to optimize the transmit power for providing users' required data rate. Our simulation results demonstrate that DQN's median convergence time training is 90% shorter than the Q-Learning (QL) based algorithm. The DQN-based algorithm converges to the desired user rate in half duration on average while converging with the rate of 96.1% compared to the QL-based algorithm's convergence rate of 72.3% Additionally, thanks to its continuous state-space definition, the DQN-based power allocation algorithm provides average user data rates closer to the target rates than the QL-based algorithm when it converges.

cs.NI↗

arXiv2020

Federated Mimic Learning for Privacy Preserving Intrusion Detection

Internet of things (IoT) devices are prone to attacks due to the limitation of their privacy and security components. These attacks vary from exploiting backdoors to disrupting the communication network of the devices. Intrusion Detection Systems (IDS) play an essential role in ensuring information privacy and security of IoT devices against these attacks. Recently, deep learning-based IDS techniques are becoming more prominent due to their high classification accuracy. However, conventional deep learning techniques jeopardize user privacy due to the transfer of user data to a centralized server. Federated learning (FL) is a popular privacy-preserving decentralized learning method. FL enables training models locally at the edge devices and transferring local models to a centralized server instead of transferring sensitive data. Nevertheless, FL can suffer from reverse engineering ML attacks that can learn information about the user's data from model. To overcome the problem of reverse engineering, mimic learning is another way to preserve the privacy of ML-based IDS. In mimic learning, a student model is trained with the public dataset, which is labeled with the teacher model that is trained by sensitive user data. In this work, we propose a novel approach that combines the advantages of FL and mimic learning, namely federated mimic learning to create a distributed IDS while minimizing the risk of jeopardizing users' privacy, and benchmark its performance compared to other ML-based IDS techniques using NSL-KDD dataset. Our results show that we can achieve 98.11% detection accuracy with federated mimic learning.

cs.NI↗

arXiv2020

Exploiting Unlabeled Data in Smart Cities using Federated Learning

Privacy concerns are considered one of the main challenges in smart cities as sharing sensitive data brings threatening problems to people's lives. Federated learning has emerged as an effective technique to avoid privacy infringement as well as increase the utilization of the data. However, there is a scarcity in the amount of labeled data and an abundance of unlabeled data collected in smart cities, hence there is a need to use semi-supervised learning. We propose a semi-supervised federated learning method called FedSem that exploits unlabeled data. The algorithm is divided into two phases where the first phase trains a global model based on the labeled data. In the second phase, we use semi-supervised learning based on the pseudo labeling technique to improve the model. We conducted several experiments using traffic signs dataset to show that FedSem can improve accuracy up to 8% by utilizing the unlabeled data in the learning process.

cs.LG↗

arXiv2020

Federated Learning for Localization: A Privacy-Preserving Crowdsourcing Method

Received Signal Strength (RSS) fingerprint-based localization has attracted a lot of research effort and cultivated many commercial applications of location-based services due to its low cost and ease of implementation. Many studies are exploring the use of deep learning (DL) algorithms for localization. DL's ability to extract features and to classify autonomously makes it an attractive solution for fingerprint-based localization. These solutions require frequent retraining of DL models with vast amounts of measurements. Although crowdsourcing is an excellent way to gather immense amounts of data, it jeopardizes the privacy of participants, as it requires to collect labeled data at a centralized server. Recently, federated learning has emerged as a practical concept in solving the privacy preservation issue of crowdsourcing participants by performing model training at the edge devices in a decentralized manner; the participants do not expose their data anymore to a centralized server. This paper presents a novel method utilizing federated learning to improve the accuracy of RSS fingerprint-based localization while preserving the privacy of the crowdsourcing participants. Employing federated learning allows ensuring \emph{preserving the privacy of user data} while enabling an adequate localization performance with experimental data captured in real-world settings. The proposed method improved localization accuracy by 1.8 meters when used as a booster for centralized learning and achieved satisfactory localization accuracy when used standalone.

cs.NI↗

arXiv2017

Indoor UAV Navigation to a Rayleigh Fading Source Using Q-Learning

Unmanned aerial vehicles (UAVs) can be used to localize victims, deliver first-aid, and maintain wireless connectivity to victims and first responders during search/rescue and public safety scenarios. In this letter, we consider the problem of navigating a UAV to a Rayleigh fading wireless signal source, e.g. the Internet-of-Things (IoT) devices such as smart watches and other wearables owned by the victim in an indoor environment. The source is assumed to transmit RF signals, and a Q-learning algorithm is used to navigate the UAV to the vicinity of the source. Our results show that the time averaging window and the exploration rate for the Q-learning algorithm can be optimized for fastest navigation of the UAV to the IoT device. As a result, Q-learning achieves the best performance with smaller convergence time overall.

cs.NI↗

arXiv2017

Occupancy Counting with Burst and Intermittent Signals in Smart Buildings

Zone-level occupancy counting is a critical technology for smart buildings and can be used for several applications such as building energy management, surveillance, and public safety. Existing occupancy counting techniques typically require installation of large number of occupancy monitoring sensors inside a building as well as an established network. In this study, in order to achieve occupancy counting, we consider the use of WiFi probe requests that are continuously transmitted from WiFi enabled smart devices. To this end, WiFi Pineapple equipment are used for passively capturing ambient probe requests from WiFi devices such as smart phones and tablets, where no connectivity to a WiFi network is required. This information is then used to localize users within coarsely defined occupancy zones, and subsequently to obtain occupancy count within each zone at different time scales. An interacting multi-model Kalman filter technique is developed to improve occupancy counting accuracy. Our numerical results using WiFi data collected at a university building show that utilization of WiFi probe requests can be considered a viable solution tool for zone-level occupancy counting in smart buildings.

cs.NI↗

arXiv2017

IoT Localization for Bistatic Passive UHF RFID Systems with 3D Radiation Pattern

Passive Radio-Frequency IDentification (RFID) systems carry critical importance for Internet of Things (IoT) applications due to their energy harvesting capabilities. RFID based position estimation, in particular, is expected to facilitate a wide array of location based services for IoT applications with low-power requirements. In this paper, considering monostatic and bistatic configurations and 3D antenna radiation pattern, we investigate the accuracy of received signal strength based wireless localization using passive ultra high frequency (UHF) RFID systems. The Cramer-Rao Lower Bound (CRLB) for the localization accuracy is derived, and is compared with the accuracy of maximum likelihood estimators for various RFID antenna configurations. Numerical results show that due to RFID tag/antenna sensitivity, and the directional antenna pattern, the localization accuracy can degrade at blind locations that remain outside of the RFID reader antennas' main beam patterns. In such cases optimizing elevation angle of antennas are shown to improve localization coverage, while using bistatic configuration improves localization accuracy significantly.

cs.NI↗