Archives for April 2014

I spoke at HTML5 App Developers in New York on April 16th. John Paul approached me two months ago and asked if I would present. Until then, I hadn’t considered speaking at a meet-up because I didn’t have anything to share. When John Paul asked me, I said yes because I knew it would be a good opportunity. I had never spoken on a technical topic, but I had been to enough to know what to prepare.

Upon submitting my talk topic a month ago, I decided on Computer Vision (CV) and Machine Learning (ML) in JavaScript. I never studied complex math or computer science in school, so I needed to research the topic to preparing a talk.

Before picking my talk title, I did my share of Google-ing around for potential leads. I made sure Javascript and ML were existing topics. I also had read of Computer Vision applications to e-commerce filtering stores. I made sure people had written on the topic and looked for codebases I could learn from.

I collected all the resources I found in a Evernote pad. I used Evernote for the ease of sharing between devices. I have a work computer, personal laptop, tablet, and cell phone. Whenever I had time, I looked for links and added them to my document. With ever link, I added a short description of the content. I would reorganize the document in priority of usability and value.

Once I had enough resources to pick from, I began gaining a high level picture. I only Googled and collected links on first impression. After looking further into the topic, I realized I made a huge mistake. The resources I found were of ML and CV with JavaScript, but not with the browser. The resources were Node applications. They were no different from the C++/Python/MatLab/Java alternatives. This was a problem.

Talking to people and Github search saved me in the end. I also looked on HackerNews and sought out email newsletters on ML. I used Github repository search and filtered out backend languages and found JavaScript-only projects. Also through reaching out to people through their blogs and Twitter, I was able to get insight into resources to look into. Talking with people helped speed up my ability to identify the “best” resources available.

After securing resources, I focused on gaining a lower-level of how the resources functioned. I was able to read and understand the JavaScript, but I didn’t understand the principles. I found an online course held by Stanford’s CS program. The resources I found mentioned the class. The class provided a full-spectrum of the ML mechanics and familiarized me with the vocabulary.

Through reviewing the class, I realized it was impossible to grasp the topic before the presentation. There was too large of a gap between what I understood and what I needed to know. I didn’t understand linear algebra.

I spent time to understand the primary concepts, but ignored mastering them. I looked at matrix math, linear digression, and the referenced algorithmic methods. I didn’t understand the depth of the algorithms, but I understood what they did/didn’t have the capacity of doing. Through this, I was able to synthesize the concepts using analogies. Instead of presenting the terse concepts in full, I could present the analogies.

Finally, I decided how I would present. I synthesized the materials I found. I separated the materials into a presentable form. I started with the broad concepts. Then I moved to explaining low-level behavior. Finally, I planned to present examples. I ran the presentation structure by John Paul. After hearing his thoughts, I restructured the presentation to focus more on code and less on concepts.

Meet-up talks are on a technology/library the presenter made. In my case, I didnt have anything to showcase. I felt the libraries and topics were obscure enough that people would value from the exposure. After restructuring the material, I planned to make a photo-rich presentation.

The presentation itself was first written out. I wrote out one line-per-slide in a Word document. Once I had the majority of the presentation, I started looking for images to go with each line. I had an idea of what I wanted to show for each line, so this process was a matter of collecting resources. After I collected all the images and made the respective gifs, I slapped it all in a Powerpoint presentation. I printed out the text to read, so the computer would only present slides.

The preparation helped me during my actual presentation. Although I was nervous, I knew exactly what to say. I set up a timer for myself when getting on stage, so I knew if I was behind schedule. By the half-way point in my presentation, I stopped reading from the prepared text. I had already gotten in the zone and didn’t need to have a guide. The talks delivery was key, so even though I didn’t use the prepared text, it comforted me to have.

I didn’t know anything about ML or CV. I couldn’t have be happier for accepting to talk. I am confident through the preparation, I become more familiar with ML and CV than I could ever have expected to be. The research also gave me some good ideas for future projects and talks. I realized through researching that there were a lot of applications of ML and CV that I expected, but couldn’t find. The topic’s complexity is a barrier for applying the concepts to everyday problems people would enjoy.

Link to the talk

I was born in a generation of the dying video rental stores. Blockbuster’s business peaked in 2004. Then it declined. The rise of RedBox, Netflix, and the On-Demand alternatives crushed the rental business. In middle school, I spent afternoons roaming the aisles of my local video stores. I could stop in once section and see all the tapes in a single series. I spent many nights binge watching all the original Star Trek, Indiana Jones, and Star Wars films. At the video store, I would find a section I liked and grab as many videos I could fit in my arms. I used genre to discover videos.

I was also born into a generation with frictionless access to music. My siblings’ generation were apart of the Walkman revolution. I was apart of the music industries transition to CDs. The rise of Virgin Music. This was the shift of the vinyl from a music listening device to a vintage item. This was the Napster generation. The rise of Kazaa. And the growth of iTunes. My ability to access music changed over a few years.

In elementary school, I couldn’t answer questions about music. My classmates would have favorite bands. They would talk about the concerts they went to with parents. I couldn’t relate. My favorite CD’s were movie soundtracks. I listened to the Matrix soundtrack on repeat for months. Not because I though it was the best music, but because it was all I knew about.

When P2P networks began to rise, I started discovering music through individual song downloads. I would search for bands I had heard of, and download any of the returned results. In middle school, I discovered House music and started downloading music based on genre. I would search in mass for keywords associated to my taste. After downloading songs, I would burn it on a CD and listen to it on the way to school. I rarely shared CDs. I felt like the music I listened to was unique to my tastes.

While driving with my friends parents, we would listen to the radio. We would listen to the Oldies channel. I rarely heard anyone who listened to the House or Techno music I liked.

Torrenting changed the way I consumed music. Instead of downloading individual songs, I began downloading discography’s of bands. I wouldn’t download the individual songs I liked, but instead I would have every song ever released by an artist. I remember going through an intense Rage Against the Machines, Wu-Tang Clan, Nirvana, and MF Doom phase. This was because I was able to get the discography and listen to every one of their songs.

By having the discography, I found it difficult to find the quality songs by a brand. P2P services acted as a indirect rating service through the number of upload/downloads happening. Torrenting services acted the same way, but the files downloaded were much larger. Through a P2P service, the trafficked files were important. Discography’s didn’t show me the best songs or albums by an artist. As a result, I started having more music than I would ever listen to. I even had bands that I didn’t like because I never heard their ‘good’ songs.

My problem wasn’t about access, it was about identifying quality. P2P networks worked because people download one song at a time. If there was high traffic on a song, it was good. With Torrents, the good songs were no longer clear. I didn’t care enough to research a band after downloading their discography. I was more interested in the feeling of having all the songs.

iTunes was a game changer. It wasn’t about making music accessible. I had all the access I needed. It wasn’t about the instant access. As a high schooler, I had all the time in the world to find how to download an album or song. I would scour IRC boards, download forums, or torrents. iTunes was a game changer for me because it reintroduced the ability to distinguish an artist’s quality songs. This was missing before.

I had a discovery problem. In video stores, I discovered content based on genre. I would go to the “editor’s picks” section if I was in need of inspiration. For music, I started out with movie soundtracks. Napster and Kazaa made music accessible to me. Before, I felt that music was a foreign world. After, I could learn from the presence of an existing community of consumers. Torrents gave me access to everything I wanted in bulk. My issue was not about access. I didn’t know how to identify the quality. iTunes reintroduced the ability to discovery quality.

When I discovered Hype Machine, I became ecstatic. Hypem surfaces the internet’s most popular songs by tracking music blogs. Hypem lets you listen to music through a online stream, without advertisements. I would listen to the Hypem most popular and if I heard something I liked, I would find the artist in iTunes and buy their songs. I discovered many artists I love through this methods.

Discovering new music is great. I love finding a new band that I can share with my friends. I also love having a variety of music playing in the background. When Im in the mood for something new, I use Hypem. This is the best part of the passive music listening process for me.

And now, Spotify is my new boat. A band I haven’t listened to in a long time will resurface in my memory. When this happens, I turn to Spotify. Spotify is my go-to music access tool. In contrast to Hypem, I use Spotify when I know exactly what I want to listen to. Today it was Spoon. Yesterday it was The Faint. Tomorrow, who knows. Regardless, having a tool to discover new music and another tool to resurface old music are crucial.