Filters

2021 - 20222
Reset filters

Settings

Subject: reinforcement learning

Search Results

Now showing 1 - 2 of 2
  • Thumbnail Image
    Article
    Citation - WoS: 11
    Citation - Scopus: 20
    Reinforcement Learning Using Fully Connected, Attention, and Transformer Models in Knapsack Problem Solving
    (Wiley, 2022) Yildiz, Beytullah; Yıldız, Beytullah; Yıldız, Beytullah
    Knapsack is a combinatorial optimization problem that involves a variety of resource allocation challenges. It is defined as non-deterministic polynomial time (NP) hard and has a wide range of applications. Knapsack problem (KP) has been studied in applied mathematics and computer science for decades. Many algorithms that can be classified as exact or approximate solutions have been proposed. Under the category of exact solutions, algorithms such as branch-and-bound and dynamic programming and the approaches obtained by combining these algorithms can be classified. Due to the fact that exact solutions require a long processing time, many approximate methods have been introduced for knapsack solution. In this research, deep Q-learning using models containing fully connected layers, attention, and transformer as function estimators were used to provide the solution for KP. We observed that deep Q-networks, which continued their training by observing the reward signals provided by the knapsack environment we developed, optimized the total reward gained over time. The results showed that our approaches give near-optimum solutions and work about 40 times faster than an exact algorithm using dynamic programming.
  • Thumbnail Image
    Conference Object
    Citation - WoS: 4
    Citation - Scopus: 7
    A Reinforcement Learning Algorithm for Data Collection in Uav-Aided Iot Networks With Uncertain Time Windows
    (Ieee, 2021) Cicek, Cihan Tugrul
    Unmanned aerial vehicles (UAVs) have been considered as an efficient solution to collect data from ground sensor nodes in Internet-of-Things (IoT) networks due to their several advantages such as flexibility, quick deployment and maneuverability. Studies on this subject have been mainly focused on problems where limited UAV battery is introduced as a tight constraint that shortens the mission time in the models, which significantly undervalues the UAV potential. Moreover, the sensors in the network are typically assumed to have deterministic working times during which the data is uploaded. In this study, we revisit the UAV trajectory planning problem with a different approach and revise the battery constraint by allowing UAVs to swap their batteries at fixed stations and continue their data collection task, hence, the planning horizon can be extended. In particular, we develop a discrete time Markov process (DTMP) in which the UAV trajectory and battery swapping times are jointly determined to minimize the total data loss in the network, where the sensors have uncertain time windows for uploading. Due to the so-called curse-of-dimensionality, we propose a reinforcement learning (RL) algorithm in which the UAV is trained as an agent to explore the network. The computational study shows that our proposed algorithm outperforms two benchmark approaches and achieves significant reduction in data loss.