Five Machine Learning Projects I Want to Be Working On Now

There’s only so much data we can look at and still make sense of it. Even if we have acute pattern-matching genes and excellent data wrangling skills. There comes a point where time and bandwidth become an obstacle.

But gobbling copious amounts of data and making sense of it is something computers are really good at. Garbage In — Garbage Out still applies, but at least you’ll get your garbage quicker.

In the last year, Google has been making great strides towards making machine learning and artificial intelligence available to all, as well as incorporating ML and AI into many of their consumer-facing products.

Ok, let’s get on with the show. Here’s a list of five problems I want to tackle with ML. Join me if you like ;-)

1. Analyzing our humongous video library

I have over 17,000 hours of broadcast quality video, already digitized and up on the cloud. Many of these shows have over 150 episodes, each one 45 minutes long! I’d love to make compilations of the best scenes of these shows, but looking for them is a daunting task. Google’s machine learning infrastructure could help, by analyzing each scene and tagging it with metadata: is it indoors or outdoors, are there people in the scene, are they happy, angry, fighting? kissing? Are there any pets? Brands? Can we recognize the actors?

2. Creating a recommendation engine for our video viewers

One of the most interesting problems we face is recommending users more content to watch, since the more you watch the longer you stick around. There are many ways to do this and Netflix has probably paved the way. Now, imagine if you took each user’s watch history and their search queries, added some external databases like Wikipedia and IMDB, maybe looked at the campaigns that brought each user in, sprinkle some Facebook and social media pixie dust and you can begin to figure out not only what to recommend each user to watch next, but you also get some valuable information about which content to license or produce for your users.

3. Analyze your onboarding funnels

When you’re paying top dollar to acquire customers, you need to make sure they’re not leaking out of your funnel. Common sense and best practices will take you a long way towards an efficient funnel (I quadrupled our funnel’s effectiveness by taking the time to analyze it, compare it and optimize it), but there’s still a lot of data to go through: where are your best customers coming from, how quickly do they start watching videos, how many searches do they make, when do they drop off, why do they stick around?

Modeled appropriately, all these data can tell you when a user is about to become a long term subscriber (or when he’s about to be gone, so that you may bring him back), or if there is a campaign, or time of day, or other obscure variable, that makes a difference? As your customer base grows, analyzing these data on Excel, Sheets or MixPanel (to name a few), becomes increasingly more complicated and time consuming. You want your analysis to scale and your results to be actionable.

4. Analyze my multi-channel network

You can probably tell by now that I like to look at data. Sometimes I don’t even know what I’m looking for. Like a sculptor, I like slicing and dicing data to see what’s there. You still have to be careful not to draw the wrong conclusions, but it’s very exciting when you see a pattern appear that might open your eyes to new insights.

I’ve been running a multi-channel network on YouTube for a number of years now. We’ve grown every year and I continue to see opportunity in the space. I’d like to sit down, look at the data we have and look for patterns that indicate potential success — not the next viral hit, but sustainable growth… but it’s a lot of channels to peruse. And we can use what we learn in #1 above as well.

5. What else do you think I should work on?

What other cool ML opportunities do you see?

I’m more of a PM than a coder, since my coding chops are uber-rusty (but I’m working on it), so this may take me longer than I’d like. But let me know if you’d like to work on this or any other cool projects. And if you think any of these are not good candidates for ML/AI, I’d love to know why you think so too ;-)

Thanks!

Originally published May 23, 2017 on Medium.

Want to see how these predictions aged? I revisited this list nine years later — read the 2026 update.