<?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 Janet Zhong on Medium]]></title>
        <description><![CDATA[Stories by Janet Zhong on Medium]]></description>
        <link>https://medium.com/@janetzh?source=rss-253cdf5852a2------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*P3H-1NJvLacLjapeKSdcPg.jpeg</url>
            <title>Stories by Janet Zhong on Medium</title>
            <link>https://medium.com/@janetzh?source=rss-253cdf5852a2------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 18:15:43 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@janetzh/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[My Relationship with Physics (4 years into the PhD)]]></title>
            <link>https://medium.com/@janetzh/my-relationship-with-physics-4-years-into-the-phd-386861bec031?source=rss-253cdf5852a2------2</link>
            <guid isPermaLink="false">https://medium.com/p/386861bec031</guid>
            <category><![CDATA[grad-school]]></category>
            <category><![CDATA[reflections]]></category>
            <category><![CDATA[physics]]></category>
            <dc:creator><![CDATA[Janet Zhong]]></dc:creator>
            <pubDate>Thu, 18 Jul 2024 07:10:20 GMT</pubDate>
            <atom:updated>2025-11-15T01:38:21.515Z</atom:updated>
            <content:encoded><![CDATA[<p>How I feel about physics has changed a lot over the years. It brings me a deep amount of joy, but at some point I felt very existential doing physics. How did it all change? I could not have cared less about physics for the first 17 years of my life. I did not grow up wanting to be a physicist. In my spare time, I would think about people, not physics. At some point, physics became very colorful and creative and it also became a deep anchor to me. I didn’t start daydreaming about physics until last year — the third year of my PhD. I’m now in my fourth year, the end of my PhD is soon, so I thought it pertinent to pause and think about how I feel now while I am in the thick of it.</p><p><strong>Phase 1: Could not care less (first 17 years of my life)</strong></p><p>I was not a particularly science-y kid. I liked all the other subjects similarly, and my hobbies were much more art based. I drew a lot.</p><p><strong>Phase 2: Slight curiosity (high school physics competition)</strong></p><p>I did well in school but I don’t think I stood out that much. After one random physics test, my high school physics teacher made me sit a physics competition which sent me to a science summer camp. The camp was overwhelming, I struggled in it, but I also met people that actually did physics for fun, which blew my mind. I didn’t know physics was supposed to be fun — it was so different to how I felt about physics at school and it was the first hint that physics was supposed to feel creative.</p><p><strong>Phase 3: Existential black hole (undergraduate)</strong></p><p>Encouraged by these high school physics episodes, I continued physics but also tried all the sciences. I did better in physics, but I didn’t love it yet. In fact, I often felt quite useless in a practical sense. I didn’t feel like I was doing anything of impact. To me at the time, impact was through people and physics felt devoid of people. In college, people had debates about the Real World at the dining table, and I spent my day in some dungeon twiddling laser knobs only for it to be slightly more off centre than when I started.</p><p>I was solving some astrophysics problem about how a star was blinking (because something was orbiting around it) and I just couldn’t overcome a feeling of “what is the point??” Friends called this my `existential black hole,’ era. At this stage, me and physics were not looking good. (No offense to astro friends, I’m sure there’s very interesting problems there, and I might feel differently now).</p><p><strong>Phase 4: I love pretty plots (undergraduate)</strong></p><p>The existential period slowly ended when I had to write a class project report on chaos. I came across this beautiful rainbow plot in a textbook and really wanted to reproduce it for my report. It was the first time I got lost in a project. I could not code at this point, so the project was rough. I remember having to ask for help from someone in my dorm because I was completely helpless with Mathematica. With a bit of aid, I started to get some plots, and I wanted more and more colorful ones.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/408/1*I9xBkdL4BIbgsL_UifuKXQ.png" /><figcaption>It was some plot about Duffing oscillators that I tried to replicate from Strogatz’ textbook, I think.</figcaption></figure><p>I learnt that when I got immersed doing physics, I was actually having fun, and that was meaningful despite it having no impact on people in the moment. That was enough, for now. (I used to tell friends that ‘the meaning of life is immersion,’ after this when they asked how my ‘existential black hole’ was going).</p><p><strong>Phase 5: Exploring other things (graduate school)</strong></p><p>I came to graduate school in physics because it opened the most doors for me (I moved to America and Silicon Valley, the land of abundance, over-abundance and under-abundance). But I didn’t go just for physics — I wanted a window into the world. I explored all sorts of things: design-y classes, business-y classes, app-building classes etc. I like a lot of things outside of physics (I still love art). Even though I was grateful for physics — it had already given me so much in life — I still wasn’t sure I’d end up sticking with it. Physics felt like it satiated one spike of my brain, but I wanted something that encapsulated all the other parts of me too. I sometimes get a glimpse of a feeling of an alternate life and the feeling that those other paths also feel right.</p><p><strong>Phase 6: Physics gets deep (graduate school)</strong></p><p>I didn’t think about physics in my free time like my high school physics camp friends. It made me feel like I would never be like them, which is actually fine (and I still think this way). But at some point, life was painful in graduate school for non-academic reasons. I started to have anxiety or panic attacks, where I wouldn’t be able to drive and I couldn’t talk to people. I struggled to rest my mind enough to sleep, and it was in this time one of the only things that calmed me was watching physics videos. It was deeply soothing. It was like my brain needed to latch on to something, and physics was so cleanly devoid of people that it was calming. The feeling of physics in this time was like going on a very satisfying run. It also felt like every new piece of physics was just for me. As I learnt more, the more I could do, so learning about physics weirdly felt like self-care.</p><p><strong>Phase 7: Daydreamin’ (now)</strong></p><p>I’d experienced things like the last paragraph in smaller doses before so I tried to be careful to not be too obsessive in a negative way. I’m no longer in that phase of life so the relationship is different again, and lighter. I think it’s less intense, but nowadays I do daydream about physics in a fun way (sometimes). Maybe it is because I am in my fourth year, where I am in the tail end of my PhD.</p><p>There was a GIF I saw about programmers realizing the solution to bugs in their code in really random settings — on a walk, on the toilet, mid-talking to someone and having a huge “ohhh” moment out of nowhere, and this is how physics has felt to me in this time. Sometimes something would click really randomly, weeks later after building more and more intuition, and it would be a little burst of dopamine. Sometimes I’d solve a bunch of things successively, and it would be like lots of little joys.</p><p>I still get stuck on problems many times, and certainly go into dead ends too, but that’s fine. I feel genuinely excited hearing about some topics, it almost feels like hearing some of my favourite artists. I sometimes feel like I don’t have control over what topics I like.</p><p>I guess I’ve been shielded by a lot of the more unpleasant parts of academic life, because I had great people and environments to work in, so I have a privileged perspective.</p><p><strong>Conclusion</strong></p><p>So that’s some slice of how I feel about physics. It’s a very rich tapestry of many feelings to me, so it is hard to convey it all. But I do love it a lot, and it’s also given me a lot in life. But I didn’t always love it, and I can still picture my life going different ways. Interestingly, the reason I fell out of love for physics (because it was devoid of people) also became the reason why it felt so deeply soothing for me later. I also now see physics in so many aspects of useful things to society. I suspect this see-saw of feeling too far from people doing physics (high in my ivory tower) or feeling like physics can bring me closer to people (through technology and inventions, or just pure science) will continue to oscillate.</p><p>But I’ve had a very meaningful journey thus far, so thank you physics.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=386861bec031" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to design a mobile app in ten weeks (using interviews, Figma and React Native)]]></title>
            <link>https://medium.com/@janetzh/how-to-design-a-mobile-app-in-ten-weeks-using-interviews-figma-and-react-native-349c7ce6a49f?source=rss-253cdf5852a2------2</link>
            <guid isPermaLink="false">https://medium.com/p/349c7ce6a49f</guid>
            <category><![CDATA[ui]]></category>
            <category><![CDATA[figma]]></category>
            <category><![CDATA[ux]]></category>
            <category><![CDATA[react]]></category>
            <category><![CDATA[mobile-app-development]]></category>
            <dc:creator><![CDATA[Janet Zhong]]></dc:creator>
            <pubDate>Fri, 23 Dec 2022 02:35:53 GMT</pubDate>
            <atom:updated>2022-12-23T02:35:53.074Z</atom:updated>
            <content:encoded><![CDATA[<p>This article summarizes the process of going from nothing to designing an iOS or android app in ten weeks. It covers four-steps:</p><ol><li><strong>Need finding interviews:</strong> interviewing people to determine a problem to solve (three weeks)</li><li><strong>Low-fi prototyping:</strong> brainstorming/ validating a solution to the problem (two weeks)</li><li><strong>User interface (UI)/ user experience (UX) design: </strong>med-fi/ high-fi prototyping with Figma (two weeks)</li><li><strong>Coding implementation:</strong> Mostly front-end engineering and a bit of back-end engineering with React Native (JavaScript code that works for iOS and android), Expo, GitHub, Microsoft Visual Studio, API’s etc. (three weeks)</li></ol><p>This article loosely follows my <a href="https://hci.stanford.edu/courses/cs147/2022/au/">introduction to human computer interaction (HCI) course</a> at Stanford. HCI is the interaction between humans and machines. It is a class mostly focused on UI/ UX design and less on business feasibility. I learnt that great design is a combination of deep people perceptiveness, aesthetics, inventiveness, pragmatism, good coding/ engineering and lots of user feedback.</p><h3><strong>Dishcovery (the app we built in ten weeks)</strong></h3><p>In the class, my group (four people) picked the theme ‘keeping culture alive’. We made an app called Dishcovery which is an image recognition app that scans unfamiliar ingredients, gives you additional context about the ingredient (cultural information, nutritional information, flavour profile etc.) and allows you to explore recipes with it. The primary use case is <strong>reducing intimidation to explore cooking in unfamiliar cuisines.</strong></p><p>The whole process of making Dishcovery took about ten weeks. Here is a demo video of Dishcovery:</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FDagL9r_Kp08%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DDagL9r_Kp08&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FDagL9r_Kp08%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/41fd3e5c5b14f4c70040c3d9cb1d88ab/href">https://medium.com/media/41fd3e5c5b14f4c70040c3d9cb1d88ab/href</a></iframe><p>We won best overall project in the class which was a nice bonus!</p><h3>Need finding interviews (in person, zoom etc.)</h3><ul><li>Time taken: three weeks</li></ul><p>Our class emphasized that we should <strong>interview people with a strong emphasis that we don’t have any potential app idea or solution in mind.</strong> This is called <a href="https://hci.stanford.edu/courses/dsummer/handouts/NeedFinding.pdf">“need finding”</a> and it is a big thing in Stanford d.school (the design institute of Stanford). <strong>You are meant to talk to people and let them paint their world. From that, you figure out pain points or problems in their lives. </strong>Often they turn out to be pretty different to what we might imagine.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/325/0*xxa3zEkvcKuIhdzF.jpg" /><figcaption>99 Ranch, an Asian grocery store</figcaption></figure><p>Since we wanted to make something in food and culture, we interviewed a bunch of people at locations or jobs relating to this. One of them was a manager at 99 Ranch (an Asian grocery in America) where we found that 80–90% of shoppers (in that branch) were Asian. Foreigners were unlikely to shop at 99 Ranch due to a mixture of cultural and language barriers. They often didn’t know what ethnic ingredients were and we found that there is a lot of intimidation surrounding cooking (especially in new cuisines). If foreigners brought in recipes from the internet, they were often not very authentic recipes.</p><p>In practice, this part of the design process for an actual start-up might also involves some greater emphasis on market / industry research or more quantitative/ large scale data on users rather than just a few interviews. <strong>That said, I really liked the emphasis on getting out into the world and learning from the ground what people need and want.</strong> It’s actually quite fun but also daunting to talk to random strangers. I think it’s also okay to have preconceived ideas too though. However, we’re probably far more likely to be biased by our own ideas and to view things from a limited vantage point so having such a strong emphasis on a process such as this need finding one is a better bias than coming in too solution-oriented. Apps made with this need finding method certainly felt meatier or more purposeful but I think landing on a problem to solve is a very grey process. Probably the best thing to do is to collide ideas with reality as soon as possible and very rapidly adjust feedback/ the market. This whole process is probably quite subjective.</p><h3>Low-fi prototyping</h3><ul><li>Time taken: two weeks</li></ul><p>Once you have a problem to solve it is now time to generate some solutions. A fun idea-generating exercise is called <strong>“</strong><a href="https://www.youtube.com/watch?v=yz4g87XapQ0"><strong>crazy 8&#39;s</strong></a><strong>”</strong> where you divide a piece of paper into eight and come up with eight designs in eight minutes for the problem identified in need finding. We had different topics for each of the eight panels including AR/VR, voice, speech based, dark horse (any crazy idea ignoring feasibility), and more but you could probably reinterpret this exercise with whatever prompts you like.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/774/1*m07KucObrR0syShf_8jIjQ.png" /><figcaption>Examples of some low-fi design ideas from our crazy 8’s exercise</figcaption></figure><p>Then we sketched out low-fi designs and tested it on a bunch of random strangers we approached on the street. We used the <a href="https://marvelapp.com/">Marvel Pop app</a> to switch between screens of different sketches. You could also achieve a similar thing using <a href="https://bootcamp.uxdesign.cc/creating-wireframes-and-low-fidelity-prototypes-in-figma-for-the-google-ux-course-326a7e11fa60">Figma</a> and its transitions. The sketches are made using a tablet or paper. A key thing is that it actually is <strong>more effective if it looks very low fidelity. </strong>If designs look too close to a final solution, or the <strong>design is too polished, people will probably give less high-level feedback</strong> and focus on more minute design elements<strong>.</strong> I thought that was interesting as low-fi prototypes are likely things that people often skip but this process might reshape your entire project.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7cjEgwPgx7eUck-DBT-oow.png" /><figcaption>Our low-fi prototype in Marvel Pop app, from very quick iPad sketches</figcaption></figure><h3>User interface/ user experience design (Figma)</h3><ul><li>Time taken: two weeks</li></ul><p>Finally we were moving on to some more design. <strong>Figma is a design software that is really amazing and by far the superior platform for making design prototypes. </strong>There are different frames for different screens, “parent components” to have consistent design frameworks or templates, third-party plugins to make life easier and you can even create animations with Figma. Most notably, you can create <strong>transitions between different screens on Figma so that you can have a very realistic mobile or web prototype without coding anything.</strong></p><p>Some screens from our first attempt at a Figma design looked like this:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/975/1*JbPD0E5OlvY1Msxyn_SKZw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/975/1*sZDUPEPD5QCjvBLiOSVWbA.png" /></figure><p>We then had our prototype evaluated according using the following <strong>12 evaluation heuristics</strong> (based off the 10 originally published by <a href="https://www.nngroup.com/articles/ten-usability-heuristics/">Nielsen</a>) which I pasted in the appendix at the end. From this, we redesigned our entire prototype. It is a good exercise to occasionally whip out this Evaluation Heuristic table. Once you start listing a few, it becomes easier to keep going and spot many others. <strong>This exercise totally changed how I view apps, and now I can’t unsee it when I look at good or bad UI/ UX.</strong></p><p>An example, is that I really love the <a href="https://apps.apple.com/us/app/calm/id571800810">Calm meditation app</a>, and find the UI/ UX very appealing. On the other hand, I also really like the <a href="https://apps.apple.com/us/app/waking-up-guided-meditation/id1307736395">Waking Up meditation app </a>but the UI/ UX is the most confusing thing I’ve ever seen and I can never find what I am looking for (what do the icons in the bottom tab mean?).</p><p>Some screens from version 2 of our design for Dishcovery looked like this:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/975/1*PKGCce02Tl1eB1NbRS7CsQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1021/1*0Mhs2gqOIKkfW6-Zu7h67Q.png" /></figure><p>Having an extremely clean and well thought out design on Figma to aspire to dramatically changed how it felt while coding. Being organized with the design is a very big process in itself called ‘design systems.’ Very small, well-designed UI elements brought me a lot of joy.</p><p>There is perhaps <a href="https://www.youtube.com/watch?v=xWiFFjT8R2g">some difference with UI and UX</a>. <strong>UI </strong>is more like the <strong>aesthetics and the visual design system</strong>. Whereas <strong>UX </strong>is more like the <strong>experience of the user flow and how easy or intuitive it is to use the app</strong>. UX requires a lot of empathy or perceptiveness of people to be well done and is more focused on the pragmatism of the app. UI is more of a visual design sense that is slowly honed over time. Arguably, it might be easier to train someone in UX, whereas a sense of good visual design is perhaps something that is harder to teach but honed over a long time. That said, good UI also sort of naturally implies a good UX so they are also coupled in ways.</p><h3>Coding implementation (React Native)</h3><ul><li>Time taken: three weeks</li></ul><p>Now that we had the design it was finally time to start coding. No one in my group had done any mobile app development before but I also took mobile development course <a href="https://web.stanford.edu/class/cs47/">CS47 </a>as this project was going. We also had various backgrounds in coding, so it wasn’t exactly going from nothing. Anyway, this section is a crash course on how to code a mobile app. I am pretty proud of what we did in three weeks from zero mobile development experience. If coding intimidates you, it may be helpful to know that you don’t have to write that much from scratch. Much of the code will end up being really similar to examples or documentation found on the internet, and you just piece it together.</p><p><strong>React Native:</strong><br>We used <strong>React Native</strong>, which is a coding framework that works for both iOS and android (developed by Facebook in 2015). Previous alternatives (like <strong>Swift</strong>) may only work for iOS / android. <strong>Flutter </strong>is another cross-platform solution. It is probably best to just pick React Native or Flutter and stick with it. I learnt React Native so that’s what this section will be about. When looking up coding tutorials for mobile apps, remember to search ‘React Native’, as ‘React JS’ is more about web apps. The code structure is similar but some of the syntax might look different. React is written in JavaScript, so learning things like what a JavaScript objects is was helpful.</p><p><strong>Expo:</strong><br>We used <strong>Expo </strong>to write our React Native code. This is helpful because Expo has things like the <strong>Expo Go </strong>app, where you can test your app as you are coding. You can also use android or iOS simulators on your laptop. In general, although it’s much better than previous alternatives, there were still lots of little hiccups to get through with using Expo, such as making sure your phone with the Expo Go app is on the same wi-fi as your laptop where you are running the code. The <a href="https://www.callstack.com/our-work/expo"><strong>Expo Snack</strong> </a>feature is a way to view code and its app simulation on the web. It is a very efficient tool for learning how to code, but for anything hefty you will probably have to store it in local files and open it in something like Microsoft VS Studio as you will probably have to download many packages (using package managers like npm and yarn). However, when learning new things about React Native, it was very useful to look at other people’s Expo Snacks.</p><p><strong>Other tools:<br></strong>Other things we used were VS studio as the text editor, Github and VS Liveshare to work collaboratively.</p><p><strong>Don’t worry if none of this made sense:<br></strong>As I have also just started mobile app development, I don’t really understand the intricacies of native vs non-native code, nor do I understand at this stage all the technicalities of using Expo/ React Native/ other platforms. I don’t know what a CLI is or what an SDK is, and it was my first time using package managers like npm and yarn. You don’t need a deep understanding of all this to get started. <strong>I recommend just </strong><a href="https://www.youtube.com/watch?v=VozPNrt-LfE&amp;"><strong>following this tutorial </strong></a><strong>(or something similar) on downloading Expo/ setting up your coding environment and going straight to coding, and learning all the details later.</strong> The actual coding itself is not too bad at all.</p><p>You may find that you prefer app development as the style of coding is pretty different to something like learning Python scripts for the first time. <strong>Front-end engineering is learning to code visual things (which are essentially all just boxes and colours)</strong>, as opposed to heaps of abstract or mathematical reasoning. JavaScript may be confusing at first if coming from other coding languages but so many answers exist on the internet. I also highly recommend the <strong>React Native documentation</strong> (<a href="https://reactnative.dev/docs/navigation">such as pages like this</a>) for learning as the writing style is actually very clear. Astoundingly, <strong>ChatGPT </strong>is an amazing resource for learning React Native. It came out in the last week of this project and I found myself asking ChatGPT how to code many features. It took some knowledge to ask the right thing, which is a learning process in itself. For some basic features, ChatGPT could successfully write ~90% of the code.</p><p><strong>Folder organization:</strong></p><p>The rough folder organization of an app looks like this:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*okRE1T8spNf1frxpVcJOyg.png" /><figcaption>Folder organization of React Native code for Dishcovery on <a href="https://github.com/janetzhong/cs47-dishcovery">GitHub</a></figcaption></figure><p>The <strong>app.js</strong> file is the home base of all the code. It is normally where all the navigation code goes and where you import all the screens. Different screens are normally in different .js folders in some folder aptly named ‘screens’. <strong>Assets, components, constants</strong> can be use for <strong>global styles, helper functions, image files, data files etc</strong>. <strong>Package.json</strong> has all the installed packages which is how it knows which packages to install when you run ‘npm install’ or ‘yarn install’. Sometimes installing all the packages might have various dependence issues. Coding sometimes felt like 80% fixing a package or dependency issue, 10% getting Expo to work, 10% actual coding.</p><p><strong>Basic file structure of a .js file</strong></p><p>At the top of .js files, one typically imports packages. After that is the main JavaScript code for the file. At the bottom where it says const styles are where all the styling (UI elements) of the code goes. One can do this inline in the JavaScript code as well (where it says style ={{…}}), but it’s normally quite long and looks better at the bottom. Some people separate the style code into different files, but a lot of people in convention also leave it at the bottom of the .js file it applies to. You can also import global styles from another .js file which also helps keep things consistent.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EOmKKI4MwkafpEmJY6yx1g.png" /><figcaption>Example app.js file in an <a href="https://snack.expo.dev/@drorbiran/flex?platform=ios">Expo Snack</a></figcaption></figure><p><strong>Some key components of React Native code:</strong></p><ul><li><strong>&lt;View&gt;</strong> is how you make boxes or containers of different elements. Everything in visual design is just arranging boxes in rows or columns with different spacing.</li><li><strong>&lt;Text&gt;</strong> is how you write text</li><li>Visual elements of the app are styled using <strong>‘flex’, ‘flexDirection,’ ‘justifyContent’ and ‘alignItems’.</strong> These can style most things including &lt;View&gt; boxes. See this <a href="https://medium.com/wix-engineering/the-full-react-native-layout-cheat-sheet-a4147802405c">cheat sheet</a> for the commands. Once you do it a few times you get the hang of it, but it can be a bit fiddly.</li><li>Good practice means having styling commands that are agnostic with different screen sizes — typically this means using percentages instead of fixed heights and widths.</li><li><strong>‘useState’</strong> is used for components that can change value. This is used for things such as basic like counts (the state which is the number of counts changes), whether a button is pressed or not pressed etc.</li><li><strong>&lt;Button&gt; </strong>is the basic button but it is not so customizable. <strong>&lt;Pressable&gt;</strong> or <strong>&lt;TouchableOpacity&gt;</strong> is much more customizable in terms of styling. Typically one adds functions like onPress=navigate.navigate… to move to different screens or pass parameters on the press of the button</li><li>Curly brackets {} normally mean something is being evaluated/ calculated</li><li>There are <a href="https://reactnavigation.org/docs/stack-navigator/"><strong>stack navigators</strong></a><strong>, tab navigators, drawer navigators </strong>for different navigation screens (just follow the documentation and adjust the examples)</li><li>There are many packages one can install for more specialized things. Examples include <a href="https://ionic.io/ionicons"><strong>Ionicons</strong></a><strong> </strong>for icons (easier than finding icons on Figma!). We also used <a href="https://www.npmjs.com/package/react-native-country-flag"><strong>country flag icons</strong></a> for Dishcovery.</li><li>For more sophisticated functions one can use <strong>API’s</strong>. An example is that our image recognition was done using the <strong>Clarifai food recognition API</strong>. This meant we didn’t have to write any of the machine learning code for image recognition, we just had to plug in the API (though this can also be fiddly). Other examples is a useful natural language processing <strong>search bar is Algolia API</strong>. One can use summarizer or generative API’s like <strong>GPT-3</strong>, the <strong>Spotify API</strong> and so much more. The more functional parts of the code is called <strong>back-end engineering.</strong></li><li>useState is pretty good for storing some values that change, but when you have a lot of data you may need a better storage solution. A local storage solution is <strong>async storage</strong>. On the web, you can authenticate accounts and send / receive data using <strong>Firebase </strong>or <strong>Supabase</strong>. This is how many things such as saved galleries, or chat apps are made, by sending it to a Firebase or Supabase data base.</li><li>Be careful of the <strong>privacy and security </strong>of your app, especially with Firebase and Supabase. There was not a huge amount of back-end engineering in this app, so I have yet to learn what this entails, but it is pretty risky and unethical if you are not careful.</li></ul><p><strong>Workflow for Dishcovery</strong></p><p>The rough timeline of how things went looked like this:</p><ul><li>Set up the basic folder structure of the app</li><li>Set up bottom tab navigator (three tabs), some of the tabs like the Explore tab had multiple pages so they had a nested stack navigator</li><li>Did the image recognition. At first, we were going to fake or hard code the result for the demo as we only had limited time. However, connecting to the Clarifai AI API turned out to be better than expected, so we were pretty happy to have the image recognition working</li><li>The image recognition results gave ~20 results, so we restricted it to only show the results of items that were in some custom specified dictionary saved locally (in JavaScript, dictionaries are called objects).</li><li>We added cultural context, nutritional information to the dictionary of the custom food items. Upon scan, you can navigate to pages where it shows this information. We also made a scanning gif which and a scan failed page</li><li>Then we made scrolling cards for the explore pages and hand sourced 5 bitter melon recipes for the demo and 5 Christmas recipes. Surprisingly, some of this information was most easily sourced through ChatGPT, but ideally if Dishcovery were real this would be sourced from people of the culture being represented.</li><li>Then we made a lot of the UI screen agnostic and added fake search results, profile and filter pages for the demo. If given more time, using Algolia search API would have been nice</li><li>In the last day, I also got halfway through adding the liked feature using async storage but didn’t finish it. This could also be done with Firebase or Supabase. We added a little UI feature that said ‘countries dishcovered’ and showed the flags of the cuisines explored in the liked recipes pages. The liked recipe page had the same gallery code as the search results page.</li></ul><p>There are still many limitations of Dishcovery, but this is what we achieved in three weeks, which I am proud of. The process of coding the app also takes some strategy, problem solving and inventiveness, which I hadn’t considered as much prior to coding.</p><h3>Closing thoughts</h3><p>Overall, it was so much fun to have such an all encompassing project from design, to coding, to interviewing people. I also had the most amazing team, and felt so happy to work with them. The timeline set out in this summary is a bit arbitrary if someone wanted to replicate it. You could do things however you like. In some ways we had so many assignments in the class and extra things not included in this summary (such as experience prototypes or a concept video) that I didn’t include in this list which added time. In other ways, having a great team, and so many resources like a TA to help us certainly helped save us a lot of time. So the time taken for each section might be totally different for anyone else doing such a project. I must also emphasize that this summary is my interpretation of the things I liked best from the class which is very subjective.</p><p>Anyway, what I appreciated most of this class is that it had a pretty strong focus on some sort of social impact or purpose for building things, and it certainly reshaped how I view a lot of things in technology. I had so much fun in this project and I now feel like I have the base skill set to make anything else I might want to make, and that’s a powerful feeling.</p><h3><strong>Appen-dish</strong></h3><p>These are the evaluation heuristics we used in class when evaluating prototypes. We made a list of the violations according to the H category and also rated the severity of the violation, but you can do whatever you like with it.</p><h4><strong>12 Evaluation Heuristics from Stanford CS147 class:</strong></h4><ul><li><strong>H1. Visibility of system status <br></strong>The system should always keep users informed about what is going on, through appropriate feedback within reasonable time.</li><li><strong>H2. Match between system and the real world <br></strong>The system should speak the users’ language, with words, phrases and concepts familiar to the user, rather than system-oriented terms. Follow real-world conventions, making information appear in a natural and logical order.</li><li><strong>H3. User control and freedom<br></strong>Users often choose system functions by mistake and will need a clearly marked “emergency exit” to leave the unwanted state without having to go through an extended dialogue. Support undo and redo.</li><li><strong>H4. Consistency and standards <br></strong>Users should not have to wonder whether different words, situations, or actions mean the same thing. Follow platform conventions.</li><li><strong>H5. Error prevention <br></strong>Even better than good error messages is a careful design which prevents a problem from occurring in the first place. Either eliminate error-prone conditions or check for them and present users with a confirmation option before they commit to the action.</li><li><strong>H6. Recognition rather than recall</strong> <br>Minimize the user’s memory load by making objects, actions, and options visible. The user should not have to remember information from one part of the dialogue to another. Instructions for use of the system should be visible or easily retrievable when appropriate.</li><li><strong>H7. Flexibility and efficiency of use </strong><br>Accelerators — unseen by the novice user — may often speed up the interaction for the expert user such that the system can cater to both inexperienced and experienced users. Allow users to tailor frequent actions.</li><li><strong>H8. Aesthetic and minimalist design </strong><br>Dialogues should not contain information which is irrelevant or rarely needed. Every extra unit of information in a dialogue competes with the relevant units of information and diminishes their relative visibility</li><li><strong>H9. Help users recognize, diagnose, and recover from errors </strong><br>Error messages should be expressed in plain language (no codes), precisely indicate the problem, and constructively suggest a solution.</li><li><strong>H10. Help and documentation </strong><br>Even though it is better if the system can be used without documentation, it may be necessary to provide help and documentation. Any such information should be easy to search, focused on the user’s task, list concrete steps to be carried out, and not be too large.</li><li><strong>H11. Accessible design </strong><br>Users can interact with the system using alternative input methods. Content is legible with distinguishable contrast and text size. Key information is upfront and not nested for screen readers. Purely visual or auditory content has text-based alternatives for users with low vision and low hearing.</li><li><strong>H12. Value alignment and inclusion </strong><br>The design should encode values that users can understand and relate to. It should make a diverse group of users feel included and respected. The design should prevent the reproduction of preexisting inequities and not create additional burdens for disadvantaged populations.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=349c7ce6a49f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[My Two Month Deep Dive into Natural Language Processing]]></title>
            <link>https://medium.com/@janetzh/my-two-month-deep-dive-into-natural-language-processing-537e69f4932c?source=rss-253cdf5852a2------2</link>
            <guid isPermaLink="false">https://medium.com/p/537e69f4932c</guid>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[nlp]]></category>
            <category><![CDATA[stanford]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[naturallanguageprocessing]]></category>
            <dc:creator><![CDATA[Janet Zhong]]></dc:creator>
            <pubDate>Thu, 31 Mar 2022 00:49:43 GMT</pubDate>
            <atom:updated>2022-03-31T08:02:34.514Z</atom:updated>
            <content:encoded><![CDATA[<p>Today I felt like reflecting on my two month deep dive on Natural Language Processing (NLP). <strong>If you had asked me two months ago what NLP is, I would have told you that I have no clue!</strong> I have since gone on one huge deep dive in it, which is quite a fun ride, so I thought I’d briefly document it.</p><p>The topic of this blog post is a bit of a mix: it is 30% an overview of NLP and NLP start ups, 30% about taking classes related to NLP and 40% random personal anecdotes of learning NLP. I probably should have picked one of these topics, but it’s too late now so I guess this blog post is the amalgamation of all those things.</p><h3>What is Natural Language Processing?</h3><p>NLP is the combination of artificial intelligence (AI) and human language. This means it is the combination of <strong>something technical</strong> (coding, logic, maths) and <strong>something very human/ people-oriented</strong> (language). A very simplified explanation of NLP is that it is the study of computer programs that can understand and manipulate text, speech or language. (Since text/ speech/ language are ubiquitous — so too are the applications of NLP). I am used to physics research which is very devoid of people and I am very people-orientated so this was an interesting shift.</p><h3>My background / journey with NLP these last two months</h3><ul><li><strong>My coding background</strong> <br>I have a lot of scrappy self-taught python skills for physics built up over the years, but prior to these two months I had never done any machine learning.</li><li><strong>MS&amp;E275 start-up class</strong> <br>I took up a start up class at Stanford called <a href="https://explorecourses.stanford.edu/search?view=catalog&amp;q=MS%26E+275">MS&amp;E275 Intelligent Growths in Start-ups.</a> We had to pick a topic or sector to focus on, and my partner who was also physics-y picked NLP. I had no clue what this was, but once I googled it, I was very intrigued. We analysed the market of NLP, how it is increasingly adopted into society, why it is going through big improvements due to shifts in technology (GPT-3, transformers) and we also interviewed 13 NLP start-ups.</li><li><strong>CS224N computer science class</strong> <br>After picking NLP for my start up course, I also ended up joining the actual computer science NLP class at Stanford <a href="http://web.stanford.edu/class/cs224n/">CS224N</a> two weeks after the quarter had already begun. This was bold, since I have little experience with machine learning (ML), but the class was so good. It had 500+ students, and they have to revise the syllabus every year to keep up with the rapid changes in NLP. Because of my minimal background in ML, the class felt a little bit like getting slammed in the face with a hardcover textbook — I found it difficult but I am glad I did it. Bonus points because Chris Manning is Australian.</li><li><strong>Balance of technical vs high-level topics</strong> <br>I found that analysing the high level trends of the NLP industry in the start-up class was so much fun, and made me dream quite broadly of what NLP could do. Doing the actual NLP class was then grounding, as I quickly realised that NLP is quite hard (lol). I found that having both was a good mixture — the high-level stuff motivated me, whereas the technical stuff gave me a more pragmatic view of what’s possible and what’s not. The technical content could also be satisfying / imaginative in a different way.</li><li><strong>The joy of novel research vs the joy of novel NLP startups</strong> <br>The appreciation and joy of novel research in NLP (as I slowly gain familiarity with it) and the appreciation and joy of novel NLP start ups that I think are really valuable to people are two different kinds of enjoyment. But both are feelings that I chase — some kind of appreciation for beautiful ideas.</li></ul><h3>What is Natural Language Processing used for?</h3><p>It is probably easier to get a gist of NLP by hearing about what it is used for. Some of the most common applications of NLP are:</p><ul><li><strong>Translation</strong><br>How do you write a program that understands the meaning of text in order to convert it into another language? Turns out AI can make this better and better.</li><li><strong>Speech recognition/ voice assistants/ chatbots<br></strong>Think Siri, Cortana, Alexa etc. A lot of us may have had clunky experiences with chatbots, but soon their capabilities may be much more seamless and/ or ubiquitous.</li><li><strong>Sentiment analysis<br></strong>This is analysing trends in feelings towards things using NLP. For example, we can detect depression from social media posts, or how people feel about certain topics such as COVID or stocks, based on Twitter trends etc. Machines and models are never perfect, but it can tell us a lot — often surpassing what humans can tell, and it is a powerful predictive tool, that we can use to build or understand things that we care about.</li><li><strong>Knowledge graphs<br></strong>This is a way of organizing data. The side panels from google are actually made from knowledge graphs.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*PaJJqGwG2o5XF17q" /><figcaption>Google’s side panel is made with a knowledge graph.</figcaption></figure><p>These are just some broad categories, there is a lot more.</p><p><strong>NLP start ups</strong></p><ul><li>One NLP start up that nearly everyone knows is <a href="https://www.grammarly.com/&#39;"><strong>Grammarly</strong></a>, which is an AI-assisted proof-reading tool. Other examples include <a href="https://primer.ai/"><strong>Primer</strong> </a>(which can summarize and find key information from large amounts of textual data), <a href="https://verbit.ai/"><strong>Verbit</strong> </a>(automatic transcription and captioning), <a href="https://unbabel.com/"><strong>Unbabel</strong> </a>(translation), <strong>Emotive</strong> (eCommerce messaging), <a href="https://www.notablehealth.com/"><strong>Notable</strong> </a>(automatic transcription of doctor-physician conversations), <a href="https://www.luminance.com/"><strong>Luminance</strong> </a>(helps lawyers read and understand legal documents more quickly), <a href="https://www.sesamm.com/"><strong>SESAMm</strong> </a>(fintech company on big data and AI), <a href="https://www.koahealth.com/"><strong>Koa Health</strong> </a>(using NLP for mental health services) and more.</li><li><a href="https://www.failory.com/startups/natural-language-processing"><strong>Here is a list of 42 NLP start-ups</strong></a><strong> </strong>— it was actually one of the first articles I read about NLP. <a href="https://www.forbes.com/sites/robtoews/2022/03/27/a-wave-of-billion-dollar-language-ai-startups-is-coming/?sh=67d3ab802b14"><strong>Here is a more recent / comprehensive Forbes article</strong></a> on NLP start-ups. After this two month journey, it was quite fun to realize that I recognized a lot of the start-ups mentioned in the Forbes article, both from class and from organically learning about it.</li></ul><h3>Interviewing 13 NLP start-ups</h3><p>For my <a href="https://explorecourses.stanford.edu/search?view=catalog&amp;q=MS%26E+275">MS&amp;E275 class</a>, my class partner and I had to interview start ups on our topic. We ended up interviewing 13 start ups:<a href="https://otter.ai/"> <strong>Otter.ai</strong></a> (automatic transcription), <a href="https://levity.ai/"><strong>Levity</strong> </a>(no-code AI), <a href="https://www.deepset.ai/"><strong>deepset</strong> </a>(open-source NLP models), <a href="https://www.knockri.com/"><strong>Knockri</strong> </a>(AI video hiring with diversity and inclusion in mind), <a href="https://www.othersideai.com/"><strong>OthersideAI</strong> </a>(text generation), <a href="https://golden.com/"><strong>Golden</strong> </a>(map of all human knowledge), <a href="https://graphaware.com/"><strong>GraphAware</strong> </a>(knowledge graphs), <a href="https://linguix.com/"><strong>Linguix</strong> </a>(AI-assisted editing), <a href="https://www.canaryspeech.com/technology"><strong>Canary Speech</strong></a><strong> </strong>(NLP for diagnostic healthcare), <a href="https://mem.ai/"><strong>Mem</strong> </a>(automatic-self-organizing notes/ workspace), <a href="https://vivatraduce.com/en/"><strong>Viva Translate</strong></a> (AI / context aware translation), <a href="https://www.vannevarlabs.com/"><strong>Vannevar Labs</strong></a> (NLP for defence) and <a href="https://www.neuralspace.ai/"><strong>NeuralSpace</strong> </a>(NLP for low-resource/ non-European languages).</p><p><strong>Reflections:</strong></p><ul><li><strong>The personalities of CEO’s /the feel of the NLP start ups varied a lot<br></strong>We talked to a wide range of people, from interns to CEO’s. Of the founders we talked to, I found that their personalities varied so much: there were more risk-averse/ steady business CEO’s, vociferous young start-up founders, more altruistic-inclined founders, straight-forward academic types, nonplussed/ impenetrable expert salesman (who are highly effective) and more. I am thankful for everyone I spoke to — they really didn’t have much to gain from calling us, and I learnt a lot from them.</li><li><strong>Have to cut through the marketing to figure out what’s real<br></strong>It’s easy for there to be a lot of hype in the start up world. Some of the excitement that people convey is real, but some of it may be professionalism too. A start up’s website is sometimes the worst way to learn about a product. The best way to cut through the hype is to just try their product yourself.</li><li><strong>Open-source is nice</strong> <br>I also found the open-source community to be a beautiful concept (so many people in research use <a href="https://huggingface.co/">Huggingface</a> models, which is a hub for open-source NLP models).</li></ul><p><strong>Future trends/ outlook on NLP</strong> <br>I tried to ask a lot about predictions for the future of NLP in the interviews. Some trends include:</p><ul><li><strong>No code AI</strong> <br>This is AI or ML platforms that are built so they do not require any programming skills. This makes AI or ML more accessible and easier to implement for people and businesses that can’t afford hiring AI/ ML engineers.</li><li><strong>Improvement in NLP models are not slowing down anytime soon<br></strong>Another overall sentiment that NLP models are getting more and more powerful. GPT-3 was a huge jump, and other bigger language models are being built which will soon surpass this.</li><li><strong>More powerful text generation</strong> <br><a href="https://www.othersideai.com/">OthersideAI</a>, <a href="https://you.com/">You.com</a> (YouWrite), <a href="https://cohere.ai/">Cohere</a> and more all have similar text generation tools that can convert bullet points, or notes into full paragraphs. Text generation still has errors, but it will continue to get much better with more powerful NLP models. Can NLP replace copy-editors, authors, bloggers? Interestingly, different people have different views. Some think it is revolutionary while others are more skeptical and believe that it is hard to replace human creativity.</li><li><strong>Text style transfer</strong> <br>This is similar to text generation but it converts writing styles. For example you can convert informal to formal language using NLP. Another interesting use case is that NLP could be used to anonymize text. People have distinct writing styles, and text style transfer could be used to remove these linguistic styles to prevent identification. I guess this could be used for bad by criminals, or for good by journalists, etc.</li><li><strong>Live-speech translation</strong> <br>This is being able to translate speech in real time. The applications would be huge such as for people that can’t speak English in an English speaking country. Maybe it would help prevent the loss of dialects in some way, if it got really good.</li><li><strong>Product collaboration tools</strong> <br>NLP can summarize and annotate content automatically. It can organize content in a way that makes it more shareable with friends or colleagues. We can use NLP to learn better together. These are some of the aims of start ups like <a href="https://otter.ai/">Otter</a> and <a href="https://mem.ai/">Mem</a>.</li><li><strong>Multilingual NLP and multi-modal AI</strong><br>~90% of NLP solutions are only for European languages. Adapting NLP models to other languages will be an important task in future, especially as NLP is increasingly adopted. This is the aim for startups like <a href="http://neuralspace.ai/">Neuralspace</a>. The two big areas of AI are NLP (language/ text) and computer vision (images). Powerful tools will likely combine the two which is multi-modal AI.</li><li><strong>Widely adopted knowledge graphs</strong> <br>There are start ups like <a href="https://golden.com/">Golden </a>that are trying to “map all human knowledge,” using knowledge graphs. People increasingly use Google’s zero-click answers which are made from knowledge graphs over actual websites like Wikipedia. Where else will knowledge graphs go from here?</li></ul><h3>Taking the NLP class at Stanford — my first machine learning course (SavedYouAClick project)</h3><ul><li><strong>The NLP class at Stanford is very good<br></strong>It’s also all online, which is nice. I would often use the public access <a href="https://www.youtube.com/watch?v=rmVRLeJRkl4">Youtube</a> videos instead of actual class recordings.</li><li><strong>SavedYouAClick project</strong> <br>The class had a <a href="http://web.stanford.edu/class/cs224n/project.html">big final project</a>, which I love because I learn much better from projects than assignments. Maybe this is because they require more creativity, or maybe because they give more autonomy. Turns out, I am an autonomy junkie. For our project, my class partners and I sought to automate the Reddit <a href="https://www.reddit.com/r/savedyouaclick/">SavedYouAClick</a> forum. In other words, we used NLP to automatically read clickbait articles and give the answer to the clickbait title hook of the article using the article text. This essentially saves you the pain of having to read the article yourself. Interestingly, no one seemed to have made this, but after we submitted our class report, we found that a German research group had just submitted a <a href="https://arxiv.org/abs/2203.10282">paper</a> on the same topic to a top conference. Pretty neat!</li><li><strong>Being consumed by projects is fun, but all that matters is friends<br></strong>This is a random anecdote, but towards the end of the project it got very busy and there was a lot to do. I spent a few nights staying up late finishing things. There is a feeling straight after an intense project finishes which is almost like a void. Projects are very fulfilling in the moment, but after they are done, the sudden loss of structure (or purpose) can be jarring. This was just a class project, so this void feeling was not huge. I also tried my best to maintain my friendships as term got busy, which avoids this. I have felt something like this in the past before. It reminds me that no matter how fun or engrossing a project is, how meaningful it might feel or how ambitious I might want to be, nothing truly matters in the end for me but friends, because that’s all I crave in the moment straight after a project finishes. Otherwise, what is the point of everything? Didn’t expect this NLP blog post to get existential, did you?</li></ul><h3>Reflections on NLP journey</h3><ul><li><strong>Some friends are tech-lovers, and some are very tech-averse</strong> <br>There are lots of cases where NLP is good, but there are definitely cases where it could be bad too. Cambridge Analytica used a lot of NLP tools. Some of my friends love tech, and adopt it quickly to improve their lives. Some of my friends are much more skeptical or tech-averse. Not every problem in life needs to be solved by some complicated ML algorithm. It’s easy to get swept into this new bubble, but one must take a step back and evaluate where it’s all going.</li><li><strong>Am I just riding a Stanford wave? What is autonomy?</strong> <br>Trying a new start up class and artificial intelligence class was super fun. I felt my world broaden lots with both classes, in a way that it hasn’t in a while. Perspective broadening shifts are one of my favourite things in life. However, I also can’t help but think that this is a very Stanford/ Silicon Valley bubble to get swept into. It is so close to where new developments happen, so of course I am excited by it all. I don’t think it’s a bad thing to ride the wave, but it is amusing to think about — my stance on things are unsurprising given what I am exposed to.</li><li><strong>Listening to NLP podcasts to sleep</strong> <br>This is another random, personal anecdote, but when I first learned of NLP, all of a sudden all I could consume was content about NLP. I was so excited by all the big questions such as what NLP can do and how NLP intersected with society. I listened to many <a href="https://podcasts.apple.com/us/podcast/gradient-dissent/id1504567418">Gradient Dissent</a> and <a href="https://podcasts.apple.com/us/podcast/the-pulse-of-ai/id1494981844">Pulse of AI</a> podcast episodes on NLP. I have memories in those first few weeks falling asleep to NLP podcasts, very happy. I go through spurts of addictive new content consumption, so I can’t tell how long things stick, but it is fun to recollect.</li><li><strong>The overall takeaway — NLP world is booming and undergoing a lot of change</strong> <br>I looked at NLP from the vantage point of interviewing NLP start ups and the vantage point of taking the NLP class at Stanford. Both convey this feeling of huge potential and growth in NLP. So many new start ups are popping up in NLP and the technology is only getting better. Similarly, NLP is so young in research years that there is so much more to uncover. Obviously, my vantage point is imperfect and I can’t see everything, but that’s the overall feeling I got these last two months. Perhaps it is similar across other artificial intelligence fields.</li></ul><h3>Where to next? What am I interested in?</h3><p>Now that both classes are over, it is hard to say where I will go with NLP from here. I hope to still learn more about it. I still have questions that are fun and some are perhaps small side projects. An example is understanding the spread of negativity of social media. This can probably be tracked pretty well with NLP. Is Twitter, Facebook or Reddit more likely to <a href="https://www.nature.com/articles/s41598-022-05359-9">spread negativity</a>? What else can be learnt from NLP about the way political views or (mis-)information spreads? I have used a chrome extension in the past called <a href="https://chrome.google.com/webstore/detail/official-media-bias-fact/hdcpibgmmcnpjmmenengjgkkfohahegk">Media Bias Checker</a>, but more powerful political bias checkers exist with NLP. Another random application was using NLP on my journal entries, to summarize trends in emotions, feelings or concepts. I also want to learn about knowledge graphs. When I first heard of NLP, one of the first things I pictured was Google as a 2D projection of a knowledge graph, with a concept map of links of blog posts, videos, websites etc. that increase in size depending on relevance, like this <a href="https://www.connectedpapers.com/">connected papers</a> website for journals. Turns out all this machinery probably already exists, Google already has a huge knowledge base that can make knowledge graphs. Maybe they don’t present it visually like a 2D graph because us simpletons prefer plain lists? Who knows. Another thing I was interested in was using NLP to understand <a href="https://arxiv.org/abs/1902.08654">conversation quality</a> and meaning. What makes a conversation meaningful? Are there NLP assistants that could help people socialize?</p><p>Anyway, that’s my NLP journey! It is likely that some of my views may change. Hope you enjoyed the ride!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=537e69f4932c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Welcome to my blog about life as a Physics PhD Student at Stanford!]]></title>
            <link>https://medium.com/@janetzh/welcome-to-my-blog-about-life-as-a-physics-phd-student-at-stanford-2c1a86ee40e7?source=rss-253cdf5852a2------2</link>
            <guid isPermaLink="false">https://medium.com/p/2c1a86ee40e7</guid>
            <category><![CDATA[physics]]></category>
            <category><![CDATA[stanford]]></category>
            <category><![CDATA[grad-school]]></category>
            <category><![CDATA[phd]]></category>
            <dc:creator><![CDATA[Janet Zhong]]></dc:creator>
            <pubDate>Sun, 20 Feb 2022 21:35:02 GMT</pubDate>
            <atom:updated>2022-02-20T22:37:37.269Z</atom:updated>
            <content:encoded><![CDATA[<p>Hello! My name is Janet (pronounced Janette because my parents spelt it wrong — immigrant parent things). I am a second year Applied Physics PhD student at Stanford. I decided to start blogging my experience about half an hour ago! I was thinking about my friend Claire’s recent new <a href="https://medium.com/@professional_pre-occupation">blog </a>and how I found it very endearing.</p><p>I have benefitted a lot from reading and hearing other people’s experiences — I would never have done a PhD had I not heard more personal recounts from friends. Having these personal insights from friends helps a lot and it is nice if it is even more accessible if you may not have these networks.</p><p>Moving from Australia to America for this PhD will undoubtedly change me — Stanford is a whole whirlwind in itself. There are many soft-life influences that are already shaping my world view. I think blogging is a great way to share that, and to bring others along with me on the journey.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VIo3mZB60ZlCv8XKCvqU9A.jpeg" /><figcaption>Me arriving at Stanford last September — very jet-lagged!</figcaption></figure><p><strong>Examples of things I might write about are:</strong></p><ul><li>My relationship with physics — from feeling existential (“what’s the point of it all?”) to loving it and even finding solace in it</li><li>How do I rate my first year of my PhD? (In terms of classes, research, social life and general life)</li><li>Reflections on stepping into the Stanford start-up culture (fun!)</li><li>The friends I meet and the ways they might change how I think (interestingly, different cohorts have different norms — physics people are quite different to start-up people)</li><li>Things I learn from classes, both technical things and things I notice about my peers/ lecturers and how they influence me</li><li>Reasons why I love research — what does my everyday look like as a theoretical photonics PhD student?</li><li>Discovering Stanford skateboarding club!! Thanks to my friend Emrys</li><li>How best to make friends or settle into a PhD</li><li>Foraying into computer science — discovering and struggling with my new CS class (Natural Language Processing — the intersection of AI/ human language: it is a very nice combination of something very human and something technical)</li><li>How does Stanford reshape my ambitions? Stanford certainly increases this feeling of life mobility. Do I like it? (yes) What balance leads to a most fulfilling life?</li><li>How does Stanford intersect with my identity? I think there are balanced ways to approach prestige.</li></ul><p>It will probably change a lot! But this is a start. And why not? The best thing that can happen is that former students I might have tutored at camp might google me or something and see what I am up to. I also want to learn how to write!</p><p>I remember a very random podcast with advice about how the best thing to do is to “collide your ideas with reality*,” — to decrease the distance between idea and action. So even though this blog is not well thought out, I’ve decided to start anyway! Hope you enjoy :-)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*K3cms_LVQ3ptpg3VFq1PWw.png" /><figcaption>Physics camp!</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*iOKmz2eJYOOHDCqWNi7CBw.png" /><figcaption>This is a view of my dorm room — credits to Supipi for helping me set it up</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Hk6-kSob6yYzZCoweI-ifQ.jpeg" /><figcaption>Skateboarding on campus, photo taken by my best friend Issy</figcaption></figure><p>*I think it was from a podcast with Marc Randolph, one of the cofounders of Netflix</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2c1a86ee40e7" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>