<?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 Gabriel Harper on Medium]]></title>
        <description><![CDATA[Stories by Gabriel Harper on Medium]]></description>
        <link>https://medium.com/@gabriel.s.harper?source=rss-4535eb5ec07------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*wxOBpUs1SFO8xj_a.</url>
            <title>Stories by Gabriel Harper on Medium</title>
            <link>https://medium.com/@gabriel.s.harper?source=rss-4535eb5ec07------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 09:32:43 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@gabriel.s.harper/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[Why learn Mandarin?]]></title>
            <link>https://medium.com/startups-mandarin-and-computer-science/why-learn-mandarin-3163460420d4?source=rss-4535eb5ec07------2</link>
            <guid isPermaLink="false">https://medium.com/p/3163460420d4</guid>
            <category><![CDATA[language]]></category>
            <category><![CDATA[mandarin]]></category>
            <category><![CDATA[chinese]]></category>
            <dc:creator><![CDATA[Gabriel Harper]]></dc:creator>
            <pubDate>Thu, 11 Oct 2018 05:51:51 GMT</pubDate>
            <atom:updated>2018-10-11T05:51:51.328Z</atom:updated>
            <content:encoded><![CDATA[<p>Learning Mandarin might be one of the smartest career choices you make. The graph below depicts languages by percentage of world GDP from 2003 to 2010.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/836/0*zVTFrWEUKsf1_gkh.gif" /></figure><p>Source: <a href="http://unicode.org/notes/tn13/">http://unicode.org/notes/tn13/</a></p><p>After English, Mandarin is the clear winner. In 2010, Mandarin accounted for 22.8% of World GDP, up from 12.5% in 2002, reflecting China’s double digit compounded GDP growth. Humans are notoriously bad at long-term planning and we should consciously compensate for this. In fifty years I may not be programming but I will probably be speaking Mandarin. Being able to communicate with an additional 1,400,000,000 people in their own language has great returns. Whilst I was in Beijing this Summer, I decided to increase my student loan so that I could undertake private tuition. I figured that the compounded intellectual returns far exceed the present value of cash, despite marginal utility.</p><p>China is going to be more influential than any other country in the world in the next century. There is a Chinese idiom for those who study Mandarin — 锦绣前程. A bright future awaits.</p><p>Written in 2013.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3163460420d4" width="1" height="1" alt=""><hr><p><a href="https://medium.com/startups-mandarin-and-computer-science/why-learn-mandarin-3163460420d4">Why learn Mandarin?</a> was originally published in <a href="https://medium.com/startups-mandarin-and-computer-science">Startups, Mandarin and Computer Science</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why is Computer Vision hard?]]></title>
            <link>https://medium.com/startups-mandarin-and-computer-science/why-is-computer-vision-hard-af81194be933?source=rss-4535eb5ec07------2</link>
            <guid isPermaLink="false">https://medium.com/p/af81194be933</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[computer-science]]></category>
            <category><![CDATA[computer-vision]]></category>
            <dc:creator><![CDATA[Gabriel Harper]]></dc:creator>
            <pubDate>Tue, 27 Sep 2016 17:03:06 GMT</pubDate>
            <atom:updated>2016-09-27T17:03:06.725Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/636/1*8p-Rn_gOCdweU2BpkmgM6A.png" /><figcaption>Credit to <a href="http://xkcd.com/">xkcd.com</a></figcaption></figure><p>In 1966, Marvin Minsky, a Turing award winner, hired a first-year undergraduate student, Gerald Sussman, and assigned him a problem to solve over the summer: connect a television camera to a computer and get the machine to describe what it sees.</p><p>Half a century later, with research teams around the world dedicated to this problem, we still struggle.</p><p>The difficulty of this area piqued my curiosity so I set out to write a computer program capable of playing one of my favourite video games, Diablo II. I placed one constraint on the program — it only receives raw pixel data as input, so it perceives the game just like a human would. A few example problems of the challenge are: How do we determine if we have been to a place before? How can we detect a monster? Using the Computer Vision OpenCV library I was able to detect monsters, but struggled to implement melee combat. However, the program successfully detects the appearance of a monster and then uses area of effect spells to kill the monster.</p><p>My other Computer Vision challenge was to develop a Google Glass App which defines a word when you point to it. I had the fortune to be sponsored to fly out to JacobsHack in Bremen. I eagerly offered to implement the finger detection component in OpenCV for our team. After coding non-stop for twenty hours, at 6.00 AM it started to dawn on us that it wasn’t going to work. My teammate, who was pursuing a masters in machine learning, made a last ditch attempt to apply machine learning to detect a fingernail. I had been awake for thirty hours, and we were pretty desperate, so it seemed promising until we realized that there wasn’t a dataset of fingernails large enough.* Who would have known?</p><p>One of the things that I have found out from these experiences is that Computer Vision is really hard. Why?</p><p><strong>Problem 1: The failure of intuition</strong></p><p>The first problem is that intuition fails us. I often try to use intuition from real world images to inform the Computer Vision development process. How do I detect edges? How do I determine that I have been to this place before? One problem with this is that our brains automatically compensate for differences in the images that we see. Two images might differ in light intensity by a huge factor, but our brains will automatically compensate for this and so we won’t see the difference. For instance, a source illuminated by office lighting, the illuminance is 500 lux, whereas if it were illuminated by direct sunlight, the illuminance would be 50,000 lux.**</p><p><strong>Problem 2: Stochastic models</strong></p><p>We unconsciously think in stochastic terms but find it hard to do so consciously. Consider the scenario where you see two people approaching from a far distance. You might wonder if it’s your two friends Jeff One and Jeff Two. You know that One often hangs out with Two, so if one of them looks like One, then it’s probably One and Two. Furthermore, if you saw them earlier on then it’s almost definitely One and Two. Maybe all of these probabilities and weightings feed into each other.</p><p>Another example might be catching a ball. If someone throws a ball at you, you can probably catch it. However, suppose Jeff One throws a ball to Jeff Two. You might find it difficult if Jeff Two asks you where he should stand, to the nearest centimeter, to catch it. We can perform complex physics equations in real time subconsciously, but find it hard to reproduce these consciously.</p><p>I feel the same happens with computer vision and capturing these stochastic relationships in code can be tricky.</p><p><strong>Problem 3: Exponential relationship between reliability and difficulty</strong></p><p>It’s not hard to detect a fingernail sixty per cent of time, but it’s a million times harder to get it to work ninety per cent of the time. In Computer Vision, there is an exponential relationship between reliability and the difficulty in achieving that level of reliability. This is why mission-critical applications such as Google’s cars pose such a hard challenge. They face an extremely harsh regulatory environment without achieving an extraordinarily high reliability rate.</p><p>These are a few reasons why Computer Vision is hard. As it turns out, Gerald Sussman, the undergraduate hired by Martin Minsky, never worked in Computer Vision again.</p><h3>Footnotes</h3><p>*Part of the problem with finger detection is that it is not a researched or documented problem. Text recognition and face detection are heavily researched and documented, with various open source libraries offering these functionalities. They are significantly easier problems because we can leverage all the existing research. This isn’t really intuitive, is it significantly harder to detect a finger or a face.</p><p>**Source:<a href="http://en.wikipedia.org/wiki/Lux">http://en.wikipedia.org/wiki/Lux</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=af81194be933" width="1" height="1" alt=""><hr><p><a href="https://medium.com/startups-mandarin-and-computer-science/why-is-computer-vision-hard-af81194be933">Why is Computer Vision hard?</a> was originally published in <a href="https://medium.com/startups-mandarin-and-computer-science">Startups, Mandarin and Computer Science</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>