I was learning more about CODEX from Algorithmia, their enterprise platform for deploying machine learning API collections on premise or in the cloud. Algorithmia is taking the platform on which their algorithmic marketplace is deployed and making it so you can deploy it anywhere. I feel like this is where algorithmic-centered API deployment is heading, potentially creating some very interesting and hopefully specialized collections of machine learning APIs.
I talked about how the economics of what Algorithmia is doing interests me. I see the potential when it comes to supporting machine learning APIs that service an image or video processing pipeline — something I’ve enjoyed thinking about with my drone prototype. Drone is just one example of how specialized collections of machine learning APIs could become pretty valuable when they are deployed exactly where they are needed, either on-premise or in any of the top cloud platforms.
Machine learning marketplaces operated by the cloud giants will ultimately do fine because of their scale, but I think where the best action will be at is delivering curated, specialized machine learning models tailored to exactly what people need, right where they need them — no searching necessary. I think recent moves — such as Google putting TensorFlow on mobile phones and Apple making similar moves — show signs of a future where our machine learning APIs are portable and are operating on-premise, on-device, and on-network.
I see Algorithmia having two significant advantages right now:
They can deploy their marketplace anywhere.
They have the economics, as well as the scaling of it, figured out, allowing for specialized collections of machine learning APIs to have the metering and revenue generation engines built into them.
Imagine a future where you can deploy any machine learning and algorithmic API stack within any company or institution, or the factory floor in an industrial setting, and out in the field in an agricultural or mining situation, processing environmental data, images, or video.
Exploring the possibilities with real-world use cases of machine learning is something I enjoy doing. I’m thinking I will expand on my drone prototype and brainstorm other interesting use cases beyond just my drone video. Thinking about how I can develop prototype machine learning API collections, that could be used for a variety my content, data, image, or video side-projects. I think when it comes to machine learning, I’m more interested in specialty collections over the general machine learning hype that I'm seeing peddled in the mainstream right now.