<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Elif Cansu Saglam on Medium]]></title>
        <description><![CDATA[Stories by Elif Cansu Saglam on Medium]]></description>
        <link>https://medium.com/@saglamelifcansu?source=rss-d6cce53345f6------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*M101mZ4EfUvk1w6V</url>
            <title>Stories by Elif Cansu Saglam on Medium</title>
            <link>https://medium.com/@saglamelifcansu?source=rss-d6cce53345f6------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 09:56:43 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@saglamelifcansu/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Reinforcement Learning]]></title>
            <link>https://medium.com/@saglamelifcansu/reinforcement-learning-1efbe37b3647?source=rss-d6cce53345f6------2</link>
            <guid isPermaLink="false">https://medium.com/p/1efbe37b3647</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[q-learning]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[robot-operating-system]]></category>
            <category><![CDATA[reinforcement-learning]]></category>
            <dc:creator><![CDATA[Elif Cansu Saglam]]></dc:creator>
            <pubDate>Sat, 16 Sep 2023 14:27:37 GMT</pubDate>
            <atom:updated>2023-09-16T14:57:50.392Z</atom:updated>
            <content:encoded><![CDATA[<p>Reinforcement learning is a method of ML where the agent learns over time through trial and error iteratively with the environment. It is also known as the semi- supervised learning model. It focuses more on learning from interaction than other ML approaches. The purpose of the algorithm is for an agent to learn to perform certain actions in the environment that lead to maximum reward. The RL representative always seeks to learn with an optimal decision-making strategy.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/892/1*ekmVTjKNt2T_uhLzLiL-pQ.jpeg" /><figcaption>The interaction of the agent with the environment</figcaption></figure><p>Reinforcement Learning, the agent chooses an action when it is in a certain state. Then the agent receives positive or negative rewards according to the chosen action. And with the information sent back from the environment, the agent switches to its new state.</p><p>This process takes place in a cycle and sequence that represents the sequence of states, actions, and rewards.</p><p>The goal of an RL agent is to maximize the total reward or cumulative amount of reward received by acting in certain situations.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/856/1*CJZkIrKZAojyiisHt-2wVg.png" /><figcaption>RL usage areas</figcaption></figure><p>Reinforcement learning has many uses, and is becoming more popular day by day due to its wide applicability in solving real-world problems. One of the reasons for this is that it was learned that the RL algorithm was used in autonomous cars and rockets produced by Elon Musk’s SpaceX company. Moreover, its popularity increased even more when the best go player in the world was defeated by the RL agent.</p><p>Q-Learning, one of the most popular RL algorithms.</p><p><strong>Q-LEARNING</strong></p><p>Q-Learning is a reward system-based reinforcement learning algorithm used to select the most appropriate action using a Q function. The purpose of the algorithm is to maximize the value function Q.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/912/1*bLBH3sAyraegRYtfMszBOQ.jpeg" /><figcaption>Q-Learning Equation</figcaption></figure><p>The basic structure of this formula is Bellman Equation. Each new situation depends on many parameters. On the way to the reward, the agent uses the experience gained in each iteration to choose where to go. It keeps these experiences in a table called the Q-Table.</p><p>The Q-Table is a chart that helps us find the best action for each situation. Since the Q-Table is initially empty, the agent moves randomly in the first stage. These random movements then continue until the agent finds the first prize. As soon as the agent finds the prize, it updates the Q-Table so it can be kept in memory. Each time the agent anticipates the next step according to this algorithm and tries to reach the maximum reward.</p><p>After reaching the reward, the agent starts acting randomly again and tries to find the reward again. As this process continues, the agent learns the environment thoroughly and figure out where to go in which situation.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/916/1*eiXWmKCLJSQ6ZDkO74TqiA.png" /><figcaption>Iteration of Q-table</figcaption></figure><p><strong>Q-Learning Terminology</strong></p><p>There are some important terms related to Q-Learning Algoritm.</p><p><strong>1.Agent</strong></p><p>Agent is an entity trying to learn the best way to perform a certain task in an unknown environment to win a reward.</p><p><strong>2.Action</strong></p><p>All possible actions the agent can take.</p><p><strong>3.State</strong></p><p>Describes the current status of the agent. After performing an action, the agent can switch to different states. The state information is given by the environment.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/972/1*z8uwhvM05IsmDXssXNX0Hw.png" /><figcaption>States</figcaption></figure><p>The agent chooses an action when moving from one state to another. And while changing a situation, the agent chooses the action that will bring the highest reward as agent learns.</p><p><strong>4.Reward</strong></p><p>A type of feedback given to the agent based on the agent’s action. This information is given by the environment to evaluate the last action of the agent. If the agent’s action is good, the agent is given a positive reward, otherwise a negative reward. In this way, the agent is encouraged to learn and choose actions by receiving positive rewards.</p><p><strong>5.Environment</strong></p><p>The environment represents the physical world in which an agent operates.</p><p><strong>Exploration and Exploitation</strong></p><p>Exploration finds more information about the environment, while exploitation uses known information to maximize reward. In the case of exploration and exploitation, the agent can use known states to get high rewards or explore more of the maze. A balance must be struck between these two strategies in order to find effective ways to optimal values and to get the most reward in the shortest time.</p><p>This balance is achieved by using the Epsilon-Greedy function. When the agent starts learning, must choose random actions to discover more paths in the maze. In the Epsilon-Greedy function, an epsilon value, ε, is determined.</p><p><strong>Github</strong></p><p>You can access my project that I created using the Q-learning algorithm here : <a href="https://github.com/elifcansusaglam/Maze-Solving-Robot">https://github.com/elifcansusaglam/Maze-Solving-Robot</a></p><p><strong>I would especially like to thank </strong><a href="https://medium.com/@mhmdnsbs"><strong>@mhmdnsbs</strong></a><strong>, who made many contributions and efforts to this story and this project.</strong></p><p><strong>REFERENCES</strong></p><p>Sutton R.S., Reinforcement Learning: An Introduction, 2nd Edition, pp. 2–5, pp.157- 160, A Bradford Book, London-England, 2015.</p><p>Zheng G., Zhang F., Zheng Z., Xiang Y., Yuan N.J., Xie X., and DRN Z.L.: A Deep Reinforcement Learning Framework for News Recommendation, 2018.</p><p>Hu, J.; Niu, H.; Carrasco, J.; Lennox, B.; Arvin, F., Voronoi-Based Multi-Robot Autonomous Exploration in Unknown Environments via Deep Reinforcement Learning. pp. 14413–14423, IEEE Transactions on Vehicular Technology, 2020.</p><p>Anonymous, Reinforcement learning in Introduction, <a href="https://en.wikipedia.org/wiki/Reinforcement_learning">https://en.wikipedia.org/wiki/Reinforcement_learning</a> , Wikipedia , last edited on June 2021.</p><p>Gunady M.K. , Gomaa W. , Reinforcement Learning Generalization Using State Aggregation with a Maze-Solving Problem, pp. 157–159 , 2012.</p><p>Sarker I.H., Machine Learning: Algorithms, Real-World Applications and Research Directions, pp. 13–15, 2021.</p><p>Hang B., Kim M., Kim J., Harerimana G., Q-Learning Algorithms: A Comprehensive Classification and Applications, pp. 1–9, Seoul-South Korea, 2019.</p><p>Watkins C.J.C.H. , Dayan P., Machine Learning in Technical Note: Q-Learning , pp. 279–292 , 1992.</p><p>Violante A., Simple Reinforcement Learning: Q-learning, <a href="https://towardsdatascience.com/simple-reinforcement-learning-q-learning-">https://towardsdatascience.com/simple-reinforcement-learning-q-learning-</a> fcddc4b6fe56 , last edited in March 2019.</p><p>Shyalika C., A Beginners Guide to Q-Learning, <a href="https://towardsdatascience.com/a-">https://towardsdatascience.com/a-</a> beginners-guide-to-q-learning-c3e2a30a653c , last edited in November 2019.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1efbe37b3647" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>