<?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 Traubad on Medium]]></title>
        <description><![CDATA[Stories by Traubad on Medium]]></description>
        <link>https://medium.com/@traub?source=rss-bb5f94c9756c------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*KNAqv1sFY4Kvux_uCv6Q5w.jpeg</url>
            <title>Stories by Traubad on Medium</title>
            <link>https://medium.com/@traub?source=rss-bb5f94c9756c------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 14:37:38 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@traub/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[Understanding Federated Learning]]></title>
            <link>https://medium.com/@traub/understanding-federated-learning-21b360578c59?source=rss-bb5f94c9756c------2</link>
            <guid isPermaLink="false">https://medium.com/p/21b360578c59</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[distributed-systems]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[federated-learning]]></category>
            <category><![CDATA[computer-science]]></category>
            <dc:creator><![CDATA[Traubad]]></dc:creator>
            <pubDate>Mon, 09 Jan 2023 17:09:12 GMT</pubDate>
            <atom:updated>2023-01-09T23:47:39.098Z</atom:updated>
            <content:encoded><![CDATA[<h4>Exploring the Benefits and Concerns of Federated Learning</h4><h4>A Machine Learning Technique for Decentralized Data</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VdG8AYhcYr0kNdpcro6RBw.png" /><figcaption>source: <a href="https://federated.withgoogle.com/">https://federated.withgoogle.com/</a></figcaption></figure><p>Federated learning is a machine learning technique that allows models to be trained on decentralized data, such as data on users’ cell phones, tablets, or laptops. Generally, federated learning works by having a central server coordinate the training and testing process by sending model updates to participating devices. Each chosen device trains the model on its local data and sends the updated model back to the server. The server then aggregates the updates from all devices and produces a new global model. It’s important to note that the private data used for training and testing isn’t sent to the server, instead only the results of training and testing.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*c7KLuilgqAJy8qOjut00dA.png" /><figcaption>source: <a href="https://federated.withgoogle.com/">https://federated.withgoogle.com/</a></figcaption></figure><p>One key advantage to federated learning is that it allows models to be trained on potentially enormous datasets that can be highly diverse because it can take data from a large number of devices. Traditional machine learning approaches typically rely on a single centralized dataset. With federated learning, because the dataset will likely be far more diverse, it can lead to better model performance and more accurate predictions.</p><p>A common concern for those who are just starting to learn about federated learning is the potential for privacy violations. On the surface, it certainly sounds like a privacy nightmare to take data from user’s individual devices and use it to train an AI model. However, because the personal information is not sent to the server, and instead only the model is sent back and forth, it affirms a user’s privacy rather than violating it. This heightened level of privacy can be especially important in cases where the data contains sensitive information, such as medical records or financial transactions. On top of the privacy measures implicit in federated learning, there are additional steps that can be taken to improve privacy, such as by encrypting the data being sent to and from the server and also by adding noise to promote differential privacy. Differential privacy is a rich topic in its own right, leave a comment if you’d like an article that goes more in depth on it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ddOE-i-4gEoG2CkgiB-jBw.png" /><figcaption>source: <a href="https://federated.withgoogle.com/">https://federated.withgoogle.com/</a></figcaption></figure><p>One interesting and often overlooked benefit of federated learning is the distributed nature of its processing. In traditional machine learning, there is a centralized machine responsible for storing and processing the entire dataset. In federated learning, the centralized server is not used for preprocessing the data, training the model, or testing the model. The more traditional machine learning models will require more significant hardware resources such as memory, storage, and processing power (frequently in the form of GPUs). In contrast, in federated learning the central server only needs to coordinate the training process and receive model updates from the participating devices. There is no requirement to store an enormous dataset and much of the processing is handled by the individual devices where the data is generated.</p><p>Yet another benefit of federated learning is that it allows for flexible training and testing of models. Given a large set of suitable devices, a subset of them will be marked for training the model and then a completely different subset will be designated to test the model. This is particularly useful when it’s not possible or practical to use all the data for both training and testing. By following this approach, it’s possible to get a more realistic evaluation of the model’s performance and ensure that it is generalizable to a wider range of devices and data.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Riu8q2vso7zrkibd5seRRA.png" /><figcaption>source: <a href="https://federated.withgoogle.com/">https://federated.withgoogle.com/</a></figcaption></figure><p>While federated learning is an exciting approach to machine learning, it is not without its concerns and potential drawbacks. While centralized datasets often go through extensive preprocessing, datasets with federated learning often go through less preprocessing because its performed locally on the device. Thus, the quality of the global model may be limited by the quality of the data on each individual device. If some devices have poor quality data, it could reduce the quality of the model as a whole. To address this, it may be necessary to implement measures to ensure that the data on each device is of high quality.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_bRr_icKjqzV4EVU3sa5IA.png" /><figcaption>source: <a href="https://federated.withgoogle.com/">https://federated.withgoogle.com/</a></figcaption></figure><p>Another issue is that federated learning by definition requires the participation of a large number of devices. For many organizations, it’s simply not feasible or realistic to expect access to the number of devices required to made federated learning possible. Further, even if there are enough devices where federated learning is possible, it’s not guaranteed that the devices will be representative of the population being studied. This can be mitigated by implementing incentives to encourage participation.</p><p>Despite these concerns, federated learning has the potential to enable more privacy-preserving and decentralized AI systems. Federated learning is already being used for a variety of tasks including natural language processing, computer vision, and recommendation systems (e.g. Google). The sky is the limit for the applications of federated learning, it’s a very hot topic in the field of AI and as the technology continues to evolve, it will be very interesting to see how and where it is used in the future.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=21b360578c59" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Solving Modulo Arithmetic on multiplied exponents Easily]]></title>
            <link>https://medium.com/@traub/solving-modulo-arithmetic-on-multiplied-exponents-easily-d9859f206151?source=rss-bb5f94c9756c------2</link>
            <guid isPermaLink="false">https://medium.com/p/d9859f206151</guid>
            <category><![CDATA[módulo]]></category>
            <category><![CDATA[computer-science]]></category>
            <category><![CDATA[exponent]]></category>
            <category><![CDATA[math]]></category>
            <dc:creator><![CDATA[Traubad]]></dc:creator>
            <pubDate>Sun, 13 Sep 2020 21:08:20 GMT</pubDate>
            <atom:updated>2020-09-14T11:01:38.357Z</atom:updated>
            <content:encoded><![CDATA[<h3>Solving Modulo Arithmetic on Multiplied Exponents Easily</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*nFWKrmX6ITRhjaDiu97X8A.jpeg" /></figure><p>It was a Friday evening on my vacation when I received a call from a good buddy of mine. He said, “Hey, do you know how to do modulo arithmetic?” To which I replied, “Yeah, I can literally teach it to you in under a minute. Remember when we were kids and were first taught division? We learned that 11 divided by 4 is 2 remainder 3. Modulo just asks for the remainder after division, so 11 mod (%) 4 is 3.” He replied that he knew this, but was given two numbers, raised to exponents and then multiplied together and was tasked with taking the resulting product modded by 10. “Well, if you use a calculator — ” My buddy quickly interjected, “He said no calculators.” Annoyed I asked, “This is a programming class?” He replied it was. We shared a brief, annoyed sigh.</p><p>I reminded him that 2³⁰ % 10 could be rewritten as ((2¹⁰ % 10) * (2¹⁰ % 10)* (2¹⁰ % 10)) % 10, so it reduces down to 64 % 10 resulting in 4. For this trivial example, that works fine but for the more complex problems, this doesn’t work so well. A question like (8³⁵ * 7²⁴) % 10 is going to be painfully difficult with this method. So… what do we do?</p><p>This was when I had what would later be called, “A Beautiful Mind” moment. It came from a simple thought: the ‘ones’ digit of values raised exponentially will almost certainly begin repeating in some pattern. Granted, I was on vacation and didn’t have much to work with, but I decided to test it.</p><p>8¹ = 8<br>8² = 64</p><p>At this point, I’ll admit that while on vacation, i had no desire to multiply 64 * 8 in my head, but I didn’t have to because we only care about the ‘ones’ digit.</p><p>8³ must end in 2, because the 4 from the one’s digit in 8² * 8 yields 32<br>8⁴ must end in 6, because the 2 from the one’s digit in 8³ * 8 yields 16<br>8⁵ must end in 8, because the 6 from the one’s digit in 8⁴ * 8 yields 48</p><p>Now we have a repeat. Our pattern is [8, 4, 2, 6] repeating. Now, we just need to figure out what value is 35th in our repeating pattern. For that, we can simply use modulo 35 % 4 = 3, so it’s the third number in our pattern: 2. Therefore don’t need to compute 8³⁵ to know that one’s digit will be a 2. We now know that 8³⁵ % 10 = 2.</p><p>To save a little time, I won’t bore you with all the math, but I observed the same sort of pattern for 7’s: [7, 9, 3, 1]. Thus 24 % 4 = 0, which means the one’s digit in 7²⁴ is 1. We now know that 7²⁴ % 10 = 1. Using this information, we can then say (2*1) % 10 = 2. Therefore (8³⁵ * 7²⁴) % 10 = 2.</p><p>This is a cool little trick. I don’t know precisely what properties I’m unknowingly employing to make this discovery, but I enjoyed discovering it.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d9859f206151" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Cracking your first crackme with Ghidra and writing a keygen with Python]]></title>
            <link>https://medium.com/@traub/cracking-your-first-crackme-w-gand-writing-a-key-generator-w-python-91c345f0dfa2?source=rss-bb5f94c9756c------2</link>
            <guid isPermaLink="false">https://medium.com/p/91c345f0dfa2</guid>
            <category><![CDATA[crackme]]></category>
            <category><![CDATA[keygen]]></category>
            <category><![CDATA[ghidra]]></category>
            <category><![CDATA[cracking]]></category>
            <category><![CDATA[python]]></category>
            <dc:creator><![CDATA[Traubad]]></dc:creator>
            <pubDate>Thu, 14 May 2020 22:03:30 GMT</pubDate>
            <atom:updated>2020-05-14T23:57:05.275Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>Ghidra is awesome. </strong>While getting started with Reverse Engineering can be daunting, Ghidra certainly makes it far easier to get rolling. As a general rule, before you get started, you’ll need to know how to write code and it helps if you have a strong background in a lower level language like C. You’ll also want to get up to speed on Assembly as you’ll need it for more difficult problems. I’ll be the first to admit that my skill in Assembly is not where I want it yet. For this article, I’ll be downloading a basic CrackMe from crackmes.one and then walking through it. This is only my second crackme so don’t expect my initial approach to be particularly professional. The good news is that if you’re like me and aren’t confident in your assembly skills yet, it won’t be a big problem for this particular crackme.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/600/1*O7PtiN8wbXj9gbw6of538A.png" /><figcaption>Created by the NSA and oh so beautiful</figcaption></figure><p>To get started, download the <a href="https://crackmes.one/crackme/5ccecc7e33c5d4419da559b3">crackme crackmepal</a>, unzip it using the password <strong>crackmes.one</strong> and import it into a Ghidra project. This particular crackme requires a “serial key” to run. Once we determine the format of this key, we can program our own key generator and test it. We can test is with some random password and see what happens.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/666/1*iEGIYknmSnu6AjERBlF3fQ.png" /><figcaption>We will try again, pal.</figcaption></figure><p>Once the key is loaded into Ghidra, click on the picture of the dragon to open the code browser. When asked to analyze the file, do so, and make sure to click the box that says Decompiler Parameter ID — they say it sometimes makes things easier. At this point, let’s try looking into the symbol tree for the main function and view the decompiled code. If you know C at all, you know what the method signature of the main method looks like, so let’s replace the current signature with the standard one by right clicking on the function and choosing Edit Function Signatures. Normally you’d use <strong>int main(int argc, char *argv[])</strong> but Ghidra has trouble with the square brackets, so instead we can use a reference <strong>int main(int argc, char** argv)</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/639/1*uzEJF5Iy1oXfjD-04ciVdw.png" /><figcaption>Our Decompiled program with an updated main method</figcaption></figure><p>Already things will be a little more clear. We see on line 12 that it is expecting argc to be equal to 2. Thus, it expects one argument to be passed in, and looking at the else on line 46 that corresponds to our condition, we see that the second argument is obviously a password. On line 13, we see that the second argument is assigned to some variable, but that name isn’t very clear. Let’s rename that variable to password by highlighting it and pressing L. On line 14, the password’s length is also assigned to a variable, so to make our lives easier, let’s rename that variable to password_length.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/729/1*GtVwB5jIhCVLMvZ0PqN2CA.png" /><figcaption>updated variable names — the code is already much more readable</figcaption></figure><p>The code is already a lot more clear, we see on line 15 that they’re expecting the password length to be 9 characters and on line 16, we see that the 5th character is a dash. We see on line 17 that a new variable is created and assigned the value of 0 and we subsequently enter a while loop. This variable gets compared to the length of password, it’s used in a condition to break the loop, and it gets incremented at the end of our loop — it’s a counter! With the above evidence, we know this is probably a for loop: <strong>for(int i=0; i&lt;strlen(password); i++)</strong>On line 21, we see that they’re checking if an ‘@’ is equal to the i’th letter of our loop, and if so we break out of our while loop and the text “good job! now keygen me!\n” will be displayed.</p><p>Thus, from what we’ve gathered, a serial key for this file must conform to 3 rules:</p><ol><li>Our serial key must be 9 characters long</li><li>Our serial key must have a dash as the 5th character</li><li>Our serial key must contain a ‘@’ symbol</li></ol><p>we can run the file and confirm this with the following key “@123–4567” which should get you the desired output.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/720/1*rCmQ4GxyjSYdAUHhr7WB1w.png" /><figcaption>You’ve been cracked, pal!</figcaption></figure><p>At this point, we can write a simple Python function that can be used to generate a key. There are tons of ways you can write this, but here’s a simple one</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/838/1*oRBu5TTBEXmRfqGohE0oLQ.png" /></figure><p>This key generator sets the possible keys as any ascii letter or number. We pick 7 characters randomly. We then insert a ‘@’ into some random position and then put a ‘-’ in the middle. The key is initially in a list format, so we turn it into a string with the join method.</p><p>Now, it’s time to test it! To make sure it works, I wrote a nice little test function to go along with it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LXabtv5ppSk-y-DkjCylIw.png" /><figcaption>50 tests? Why Not?!</figcaption></figure><p>This test works by generating a key and then running the crack me with the key a certain number of times. If the return code is 0, we know that the test was successful, if the return code is anything aside from 0, then we know that our key has failed. If we go back and look at the decompiled code of our crackme, we see that it only returns 0 when a good key is passed in and will return 1 in any other situation. If we run the test 10 times, we get the following output.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/624/1*o71dZ-s6t9t20MXnI0jzaA.png" /><figcaption>Beautiful results!</figcaption></figure><p>That’s it! We’ve now cracked our first crack me, built a key generator, and even wrote a small script to test it properly. I’ll probably tweak things a bit, but here’s a <a href="https://github.com/traubad/Keygen-for-crackmepal">github link</a> for the keygen (I’ve included the crackme). Lastly, here’s a youtube walkthrough of the whole process.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2Fwg2E69Ixnbg%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3Dwg2E69Ixnbg&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2Fwg2E69Ixnbg%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/03365f1d8231209d65c997d7852bb608/href">https://medium.com/media/03365f1d8231209d65c997d7852bb608/href</a></iframe><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=91c345f0dfa2" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>