I’m almost done with my neural network article, but I didn't quite manage to finish it today. It’s a bit longer than I expected it to be, but this is a complex topic, and I think the article's length adds to the article's clarity. It's taking longer than I expected to figure out how to explain the more technical aspects of neural networks in a way that is both accessible and thorough. My goal is for this article to be suitable for somebody who doesn't know anything about neural networks, but I also want to do my best to make sure an expert wouldn't take issue with the way I simplify things. Simplifying things while simultaneously preserving nuance is a challenge, but I feel that I'm learning a lot about the process by doing it. I'll explore the mechanics of that process in greater detail in my final essay, which will be a tutorial on making a complex topic simple for others to understand.
One thing I’m certainly learning from this independent study is that it can be challenging to predict exactly how long writing a detailed explanation of a complex topic will take! I'm not concerned that I won't get everything done by the end of the intensive period, but I should remember to expect a bit more variability in the future. The next article I’ll be writing is about how credit risk modeling works. Credit risk modeling is a branch of a field called actuarial science, which is all about risk modeling for insurance companies, banks, and other companies. I chose to write about credit risk modeling after neural networks because I’m going to build a simple neural network that performs this modeling process, just like a real world actuary would. However, this article likely won’t take me as long to write, because the research I’ve done so far about neural networks will be highly applicable to credit risk modeling, and I have experience building similar risk-evaluation models.
0 Comments
I created my own neural network today! I found this a very valuable experience for solidifying my understanding of the fundamentals of neural networks. I also started writing my article about neural networks. I haven't finished it yet, but it will be done by tomorrow.
In addition, I've been in communication with my sponsor, Mr. Cambisios, about edits I can make to my cryptocurrency article. In addition to pointing out minor errors in my post, he asked a variety of interesting questions, which I've listed below in his words. These are questions whose answers I'm eager to highlight in my post, and they are questions that I would expect many readers to have. After each question, I wrote my response. I'll use these responses as a starting point for editing my article. Q: When you talk about decentralization early on, you say, "To make this more democratic, let's have every user notify everyone else whenever..." Would I be able to delete all the transactions I would then be given access to? A: I should clarify that each user’s computer downloads the entire transaction history. A user could delete the records from their machine, but it wouldn’t affect anyone else’s records. It’s the electronic equivalent receiving a new paper copy of the ledger every time somebody makes a transaction. Q: Early you talk about Jim's crucial role. But since Jim set up the code or structure, is he a sort of systems administrator with more power than others in the system? In a banking system (which you briefly mention), we have the Securities and Exchange Commission overseeing banks. Is there a central authority in cryptocurrency ("SuperJims"?) A: That’s an interesting question. The short answer is that there are no centralized authorities in Bitcoin. My understanding is that Bitcoin is maintained by a large community of volunteer developers, rather than a single organization. Every line of code that runs the bitcoin network is fully accessible to the public. This means that anybody can analyze new changes to the network and ensure that the Bitcoin system remains fair. Transaction fraud is quite effectively mitigated by distributing the ledger to all users and requiring new blocks to be validated through mining. However, there’s nothing like the SEC to “protect investors and maintain fair, orderly and efficient markets.” This lack of a governing body is one reason that Bitcoin is an attractive form of currency for criminals, political dissidents, and other users who desire autonomy. Q: When you start talking about creating a block with 17 zeroes at the beginning, why would you want a block with such a prefix? A: The number of zeros that a block’s hash needs is continuously adjusted to keep average block validation at a rate of 1 block/10 minutes. I chose 17 zeros arbitrarily for my example, but the real number of zeros constantly fluctuates. Q: You say a motivation for mining is for the successful miner to receive users' transaction fees. Why would or should users pay transaction fees, and to whom are they paid? A: Users are required to set aside a sum of money as a transaction fee every time they pay someone. If they don’t set aside a transaction fee, their payment won’t be added to the blockchain. The transaction fees ultimately go to the person who mines the block. Thus, the purpose of these fees is to incentivize mining. The number of bitcoins a person receives for mining a block will continue to decrease over time, so transaction fees will become increasingly important for ensuring that mining is a profitable activity as time goes on. Keeping mining profitable is very important, because the reliability of bitcoin depends on a community of miners. I like to think of transaction fees as comparable to credit card processing fees; they are both paid to people whose job is to ensure the integrity of the payment system. I think I’ll put almost this exact text in my article, shortly after I introduce transaction fees. Q: Why are bitcoins limited to 21 billion ..where did that number come from? The number of bitcoins that will ever exist is typically viewed as an arbitrary cap. The important thing is to have a cap on the number of bitcoins in the first place, because the cap ensures that bitcoins will remain scarce. Scarcity is essential for a successful currency. A successful currency must be harder to generate than to receive if people are to pay each other using it. If it were too easy to generate, people wouldn’t bother selling things; they would just focus on generating more currency. If everybody just focused on generating new currency instead of selling, the currency would quickly become useless, because nobody would use it to pay each other at all. However, I did a bit of research, and it appears that the creator of Bitcoin, Satoshi Nakamoto, wanted the price of Bitcoin to grow to resemble the price of ordinary currencies as Bitcoin matured. Here’s what the creator had to say when asked this very question: "I wanted to pick something that would make prices similar to existing currencies, but without knowing the future, that’s very hard. I ended up picking something in the middle.” Nakamoto came to the number 21 million by doing some calculations based on expected inflation rates and growth of bitcoin users over time, and he decided that 21 million total bitcoins would maximize the ease of long-term bitcoin/euro/dollar conversion. A very interesting effect that I researched but decided not to dive into in my article is that Bitcoin will be the world’s first deliberately deflationary currency. If the number of bitcoins in circulation remains constant while the number of bitcoin users grows, the value of each bitcoin will grow as well. Furthermore, when somebody loses their private key, all the bitcoins in their account are lost forever. Over time, more and more people will lose their private keys, which will actually reduce the number of bitcoins in circulation. Most economics seem to agree this won’t have major negative effects on the currency, but there are a variety of predictions, and it will be intriguing to see what ends up happening. Especially when writing about such a fascinating subject, one of the most difficult parts of writing is deciding what to leave out! Today I did more research on neural networks in preparation to write my article on neural networks. In addition, I'll create my own neural network as I investigate credit risk modeling for my third article. The term "neural network" is often tossed around in technological conversations, but like Bitcoin, it can be hard to untangle what a neural network actually is. Based on my research, I'd describe a neural network as a complex mathematical process. I'll explore the structure of neural networks more deeply, but they're based on a series of mathematical functions whose output is the next function's input. It's a bit like those f(g(x)) problems you may remember from high school math class. Instead of just one function inside another, however, neural networks can consist of hundreds of functions inside of hundreds of others. This model, however, is an oversimplification, because there are several other crucial aspects of neural networks, including that different functions are given different "weights" by the network, and functions often take many different functions' results as input. As I explain my thoughts now, I realize that this whole thing sounds very confusing. I'll be sure to organize my post effectively so that a person who's never heard of a neural network before can make sense of it! One of the coolest aspects of neural networks is how effectively they can be applied to solve a wide variety of problems. For instance, an interesting application of neural networks is computer vision, or programming computers to recognize images in the same way that a human would. This process begins with "training" the neural network by showing it hundreds of images labelled by humans. The neural network then uses these labels to adjust the parameters of its functions so that it can identify images as accurately as possible. This technology is becoming increasingly important in a wide variety of fields, from facial recognition to satellite-based farm management, and understanding neural networks is a wonderful way to better understand the modern world.
Tomorrow, I'll continue researching neural networks and start writing! I finished a draft of my cryptocurrency article! Here is a link to it: https://medium.com/@a.jeremymah/how-cryptocurrency-works-a-deep-explanation-of-the-fundamentals-ddc5f8aa53e3 I decided to compliment the article with simple stick figure drawings. Although these might look slightly childish at first glance, I think they're actually quite effective in making potentially daunting concepts more accessible. I believe that one secret to explaining complex ideas effectively is to start at a place that readers feel like they are 100% comfortable with, and work up from there. Stick figure drawings are a great way to get readers of all ages and ability levels comfortable, while also giving a potentially dry technology article a bit of character. It’s worth noting that the scope of this article is slightly different than I originally planned it to be. My top priority was explaining the mechanics of cryptocurrency, but I also planned on extensively discussing other aspects of cryptocurrency, such as potential issues with it, controversy around its use, and how it may impact society. However, my detailed explanation of the fundamentals of cryptocurrency ended up being over 5,000 words long. Although the research I did would be sufficient to write much more than this, generating this explanation was my top priority, which I achieved. I need to make time to write my other pieces as well, so I decided to end the post without extensively discussing some issues I planned to explore.
Tomorrow, I'll delve into researching neural networks and prepare to write my article on neural networks Friday and over the weekend! I unfortunately wasn't able to finish my Bitcoin article today, as I had originally planned. However, I've begun writing the main part of the article, and I have my complete outline with all my ideas finished. I should be able to finish it by tomorrow. I also did a bit more research on the potential economic and societal implications of Bitcoin, and I discovered some of the most interesting aspects of bitcoin I've found yet! One fascinating thing about Bitcoin is that it's designed to be a deflationary currency, as opposed to an inflationary currency. Almost all currency in the world, such as the US dollar, is inflationary, because a central authority is constantly printing new units of the currency and will continue doing so for the indefinite future. This means that the value of one unit of such currency will decrease over time, leading to inflation. However, the rate at which new bitcoins are being generated is consistently decreasing, and over 80% of the bitcoins that will ever be in circulation already exist. If the Bitcoin user base continues increasing, then there will be fewer bitcoins per person as time goes on. This means that the value of a single bitcoin will increase over time, leading to deflation! This is perhaps the first time a currency has ever been developed with deflation specifically in mind. Economists disagree over the potential effects of mass adoption of a deflationary currency, but some express concern that mass adoption of the deflationary Bitcoin would slow the economy. Because each bitcoin that a person keeps out of the economy will be worth more in the near future, people have an incentive to avoid spending money. This decreases aggregate demand, which could cause firms to find ways to cut costs and reduce supply, which would increase unemployment and further decrease aggregate demand. This could lead to a downward spiral quite quickly. Another fun, but more far-fetched idea, is that cryptocurrency could be used by a semi-sentient form of artificial intelligence to incentivize humans to improve it. The AI system could pay its developers in Bitcoin, utilizing the privacy features of bitcoin to make sure its developers remained anonymous. Even if a governmental authority tried to ban development of the AI system, the AI system could continue paying anonymous developers. Over time, the AI system could continue improving and eventually spiral out of control. This situation certainly isn't upon us yet, and likely won't be for some time, but it is nevertheless fascinating to consider.
Today, I began writing my cryptocurrency article, did a bit more cryptocurrency research, and began researching neural networks. First, I decided on the rough structure of my article. My article will begin with a discussion of the basics of the mechanics of cryptocurrency, such as how it's traded and created. I won't start my article with broad generalizations about cryptocurrency; I think that will confuse the reader if they don't have a strong foundation. I don't intend to talk about cryptocurrency as an abstract entity until I've made it so concrete in my readers' minds that they could explain it to another person. After I explain what Bitcoin is under the hood, I'll start getting into the implications of cryptocurrency for the economy, including semi-private transactions and potential for crime. Finally, I'll mention some common misconceptions about cryptocurrency and correct them. By the time a reader has finished my article, they should be at roughly "level 5" understanding of cryptocurrency. One way I continued my research today was by typing in biased google search queries with the intent to uncover biased sources. Things I typed in included, "Bitcoin is a scam," "Bitcoin is for criminals," "Bitcoin is the future," and "why invest in Bitcoin." I learned this technique from Tim Urban of waitbutwhy.com. He says that he often does this as he writes articles to get a good sense of the division between what everyone in a field agrees on and what even experts disagree on. It seems that most experts in the realm of cryptocurrency research agree that bitcoin can be a reasonably secure method to make semi-anonymous transactions, but they disagree on whether or not Bitcoin will stay stable in the future or is a good investment to make today. Though it may seem early to be researching neural networks, my self-imposed deadline for my neural network article is Friday, so I started learning about the math behind neural networks today as well. I looked at the basic algorithm that inspired early neural networks: linear regression. I learned two ways of creating a linear regression algorithm. One is called gradient descent, which uses multivariable calculus, and one is called least squares regression, which uses linear algebra and matrix calculus. I won't be directly discussing linear regression in my neural network article, but neural networks and linear regression algorithms are fundamentally related because they are both predictive algorithms that take ideas from one another. Understanding the simpler mathematical system of linear regression helps tremendously with understanding the more complex system of a neural network. I don't intend to discuss the math that I learned about today in my article, but learning the math gives me a deeper perspective on the function of the algorithm. With this perspective, it will be easier for me to avoid misrepresenting aspects of neural networks when I simplify them in my article.
My top priority tomorrow will be finishing a draft of my bitcoin article! I spent yesterday getting myself familiar with the basics of how cryptocurrency works. Today, my goal was to dive into the controversy surrounding cryptocurrency and evaluate various people's stances on the issue. First, it's important to note that there are over a thousand different cryptocurrencies in existence. However, according to Forbes, the cryptocurrency Bitcoin currently makes up about 79% of the cryptocurrency market. According to surveys, Bitcoin is seen as the most reliable cryptocurrency, and it was also the first cryptocurrency to utilize decentralized blockchain validation. (I won't explain complex terms like that in these small, daily blog posts, but I'll be sure to explain these terms in my completed articles.) Every type of cryptocurrency is a bit different, so it doesn't make sense to talk about "cryptocurrency" as a whole when discussing debates about the topic. Doing so would be a bit like complaining about the volatility of "money," rather than criticizing the Mexican Peso or the Canadian Dollar specifically. As a result of Bitcoin's dominance, I'm choosing to focus on controversy around Bitcoin in my article. I'll briefly mention other interesting cryptocurrencies (such as Ethereum, which is associated with a smart contract programming language!), but Bitcoin will be the focus. One of the questions I'll address in my article is whether Bitcoin is truly anonymous. Unsurprisingly, there's no simple answer to that question. There certainly are ways of using Bitcoin that make it one of the more private methods of exchanging money. However, a Bitcoin user who does not take deliberate steps to increase their privacy is not anonymous. All transactions with Bitcoin are recorded on a public ledger. As a result, it is often possible to trace the flow of cash between users' addresses. Although the addresses are not directly linked to the user through Bitcoin, it's often possible to find an unsuspecting Bitcoin user's identity given only their address. For example, people often post their addresses online so that people can pay them, and these posted addresses are often linked to identity-revealing accounts. The flow of Bitcoins can be complex and difficult to trace, which I'll discuss in my full article, but it is often possible to trace the flow of Bitcoins using advanced analytics. In my article, I'll also discuss several interesting methods I found that people can use to increase their privacy when using Bitcoin.
I also discovered a fascinating fact today! But understanding this fact requires knowing the definition of an important word in the world of cryptocurrency: mining. Put briefly, mining is the process of solving a difficult math problem in the cryptocurrency system, and the owner of the computer that successfully mined gets paid in cryptocurrency. I'll explain mining in greater depth in my full article, but the important thing here is that mining requires a powerful computer (or a network of computers) and lots of electricity. Here's the fascinating fact: an estimated 90% of all remote hacking involves hackers attempting to use a victim's computer to mine cryptocurrency! The potential for privacy when using cryptocurrency unfortunately makes it an ideal way for criminals to deal with money. I'm really enjoying learning about cryptocurrency, so I'll probably read about it for fun over the weekend. On Monday, I'll start writing a first draft of my article, and I'll finish the draft on Tuesday. Today was the first day of my independent study! I spent the first half of the day making a detailed plan for the next three weeks, and the second half of the day researching the basics of cryptocurrency. In this independent study, I’ll write simple, yet nuanced explanations of five different complex topics. Ultimately, I’ll post these explanations on Medium.com. My target audience consists of interested people with a high school or college student reading level, and they are not expected to have any background knowledge of any of these topics. To reach this audience, I’ll avoid jargon (clearly explaining any I must use), include stories or jokes wherever possible, and focus on building up the reader's background knowledge adequately before diving into the actual topic. Below are the five topics I’m going to write about. For each topic, I’ve given myself a deadline by which I will have written a draft. Once I complete each draft, I’ll send it to my sponsor, Mr. Cambisios, and he will give me feedback. After revising, I'll post the final draft on medium.com, and it will show up here: (https://medium.com/@a.jeremymah) The topics, along with the date I will complete my draft of my blog post on each topic, are listed below: 1. Cryptocurrency. Draft submission date: Tuesday, August 25. 2. How computational neural networks work (explained using as little math as possible). Draft submission date: Friday, August 28. 3. Credit risk modeling. Draft submission date: Wednesday, September 2. 4. Adderall: neuroscience and national usage patterns. Draft submission date: Monday, September 7. 5. Analyzing the process of breaking down complex topics. Draft submission date: Wednesday, September 9. When thinking about my target audience, I find it helpful to rate understanding of a subject that a person can have on a scale from 1-10. This chart summarizes blogger Tim Urban's description of this scale, and it aligns closely with how I view the scale as well: As the numbers increase from 1 to 10, understanding increases by orders of magnitude. My target audience has a 2 or 3 level understanding, and my blog posts are designed to get people up to a 5 or 6.
The first step, of course, is to get myself from a 2 or 3 level up to a 5 or a 6, at which point I can begin writing. At the beginning of the day, I barely knew what cryptocurrency was. I could recite a few simple facts about it, but nothing coherent. Now, after about 5 hours of research, I would say I've elevated my understanding to roughly 4 level. Tomorrow, I'll focus on investigating controversy, false claims, and predictions around cryptocurrency to elevate myself to 5 or low-6 level understanding. |
Jeremy MahoneyImpromptu teacher, lover of learning, and admirer of the universe ArchivesCategories |