tl;dr The Meeshkan Machine Learning Public Beta is officially live. Go learn something on it!
In 2010, Cloudera played one of the Internet’s great April Fools jokes. Called Pushing the Limits of Distributed Processing, they claimed that Apache Hadoop was being run on 1,000,000 Nexus devices to help Google process images. Little did they know that, as mobile computing got more powerful, this is exactly what companies would do all the time — using the machines that connect to their network as an extension of their servers.
Putting aside the Orwellian overtones, this brings up a great question — what if we actually worked together to make distributed systems that could help people crunch data while paying device owners for usage time? Of course, most people in 2010 did not have access to massive amounts of data, let alone the knowledge to analyze it. But it’s 2017. Andrew Ng has published his second Machine Learning course on Coursera. Companies are hiring dedicated ML engineers and using Machine Learning as a Service (MLaaS) in their workflows. Artists are using Machine Learning to create crazy cool work. Individuals’ personal devices, even wearables, are getting more and more powerful. People fly GPUs across the globe to mine virtual currency. The game is different…
A World of SETIs
The SETI@home project describes itself as a network of “Internet-connected computers in the Search for Extraterrestrial Intelligence (SETI). You can participate by running a free program that downloads and analyzes radio telescope data.” Your computer helps SETI sift through an obscenely large, ever-increasing dataset to find aliens. How cool is that?!?
In 2017, the business world is starting to look a lot more like a bunch of mini-SETIs. With techniques like data augmentation as well as awesome projects like the IOTA Data Market and OpenMined, people can run their Machine Learning models on giant open datasets and data-streams. So while data is becoming accessible, the problem is that most companies and individuals cannot afford experimenting with Machine Learning in a sandbox environment to gain intuition and understanding about their models.
Enter Meeshkan. Meeshkan is a network where anyone can train and test Machine Learning models in a massively parallel way on participating host devices. In return, the hosts make a few bucks. This brings Adam Smith’s invisible hand into GPUs and CPUs and allows people to work with big data without the big price tag. Of course, a Raspberry Pi can never be as fast as an NVIDIA 1080Ti. However, thousands of these little guys will, like the African Wild Dog, divide and conquer to get things done in a reasonable time. As device owners make money, businesses can test out ideas cheaply.
Meeshkan: What’s in a Name?
Meeshkan is short for the Hebrew word meeshkenotecha, which roughly translates to “your dwelling places” and is found in Numbers 24:5. Balaam, unexpectedly moved by the Israelites’ makeshift encampment in Sinai, proclaims “How beautiful are your tents, Jacob, your dwelling places, Israel!” (Ma tovu ohalecha Ya’akov, meeshkenotekha Yisrael).
Meeshkan takes this small anecdote as a starting point for the entire service. The company’s mission is to make something beautiful out of our scattered digital dwelling places working together.
Let’s think about who wins and who loses…
- Device owners, called “sellers” in Meeshkan, win because they make money just by connecting to the network.
- Machine Learning practitioners, called “buyers” in Meeshkan, win because they can run massively parallel jobs at a fraction of the cost of services with their own machines. This is because a market of thousands of devices dictates the price.
- Meeshkan wins because our company takes a small fraction of each transaction from buyer to seller, not unlike other P2P and B2B services like eBay, Uber or AirBnB.
- Companies that overcharge you for Machine Learning lose. No, just kidding, everyone wins, so they win too because they can lower their prices. In fact, in our private alpha, a Machine Learning consulting company automatically drew extra capacity from the Meeshkan network when demand was high.
Go Learn Something!
A few random musings:
- We give you 100 free hours of ML on Meeshkan. Use it, break stuff, report back to us, and help us make Meeshkan better.
- Once you’re in Meeshkan, click on Settings, scroll to the bottom, and sign up for our Slack #public-beta channel.
- Subscribe to @meeshkan on Medium and @MeeshkanML on Twitter to watch our first end-to-end tutorial, called “Machine Learning the Github API,” which will be available next week.
- We are open-sourcing our codebase as fast as possible. Check out our projects on https://github.com/meeshkan.
- Do you like coffee? Do you want more money to buy some? Put your device on Meeshkan! Or download our Android App and we’ll transform your droid into a Deep Learning workhorse and pay you for usage time with Iota.
- Do you really like coffee? Do you want more money to buy a lot of it? email@example.com
What are you waiting for? GO LEARN SOMETHING on Meeshkan!
Your partner in machine (and human) learning,
C.E.O., Meeshkan Machine Learning
P.S. Did anyone else pull an all-nighter last night to finish Convolutional Neural Networks before the deadline? I haven’t been this excited for new installments of a series since the final Harry Potter books. BOOYAH!