Real-world Machine Learning lessons #2

In my previous post (Real-world ML lessons from GAFA, top startups & researchers), I shared a summary of some of the most interesting ML developments I've seen in 2016 and 2017. I also mentioned a few venues to find more content at the intersection of Research & Industry, Software Engineering & ML.

PAPIs is one of them, as it focuses on real-world ML applications and on the techniques, tools, and APIs that power them. For those with an industry background, it may look academic in its approach (CfP with blind reviews, proceedings, etc.), whereas for academics, it totally looks like the bias is towards industry — see the video teaser below ;)

I like to think that PAPIs has the best of both worlds — see the keynotes of our next edition for example (Cedric and D. are NIPS veterans):

• Current state of AI and ML in e-commerce and future disruption - David Drollette (Wayfair)
• Bringing powerful artificial intelligence to all developers - Cedric Archambeau (Amazon)
• Machine Learning, Technical Debt, and You - D. Sculley (Google)

D.'s talk will tackle testing and long-term usage of ML systems in your organization, which I think nicely complements another talk on widespread ML/AI usage in your organization: Putting the P in A(P)I: Why APIs are key to make AI scale by Tatiana Mejia (Adobe).

It's interesting to see that this year's ICML featured invited talks on AI Design and on real-world Reinforcement Learning. These seem to be trending topics, that you'll also find in the PAPIs '17 program:

• Behind the AI Curtain: Designing for Trust in Data Science - Crystal C Yan (FiscalNote)
• Supercharging Deep Learning with the Unity Engine - Arthur Juliani (Unity Technologies)

Heard of Neural Network-based style transfer on images? How about videos, and real-time (instead of sending an email when ready)? This poses a few ML and engineering challenges...



You'll also find topics that I mentioned in my previous post: Real-world Use Cases, Deployment, ML platforms, APIs, Software Engineering and Data Engineering. You'll hear about the integration of ML in cool applications:

• Learning artistic style for real-time stylization of video - Jeffrey Rainy (Element AI)
• Real world Turing test: when AI answers phone calls - Vincent Van Steenbergen
• Pitch Prediction for the World Series - Greg Michaelson (DataRobot)
• Rent, Rain, and Regulations: predicting crime using ML — Jorie Koster-Hale (Dataiku)
• Predicting Remaining Useful Life using IoT - Adarsh Narasimhamurthy (MathWorks)

Expect to also hear about productizing and deploying models within live applications. This will be further discussed in:

• Beyond prediction: structural modeling as a tool - James Savage (Lendable)
• Automated Machine Learning: Mostly Unhelpful - Charles Parker (BigML)
• APIs and DSLs for Building and Integrating Many Models - Harlan Harris (WayUp)
• Model as a Service up and running in AWS - Lia Bifano (Nubank).

A few of these talks will mention the platforms that power ML systems — in-house and commercial. Finally, some presentations will be focusing on some engineering aspects of ML platforms, such as Distributed Computing and Data Engineering:

• Flexible and Scalable Deep Learning with MMLSpark - Mark Hamilton (Microsoft)
• Dumpster Fire to Lit: Time-Series Data in Amazon DynamoDB - John Bledsoe (Nexosis)

Last year, PAPIs was praised for the quality of talks but also for its networking opportunities (one attendee said that "there were quality speakers and ample time and exposure for the conference attendees to speak with them individually"). Our conferences are probably similar in size to the most related workshops at top ML conferences (there are only about 185 seats at the PAPIs '17 venue at Microsoft NERD). But from my own experience, having a dedicated event helps build connections. As Denis Vanegas put it last year, "the intimate setting allowed for better conversation and relationship building." This is something we want to keep in the next editions!

You'll find more information about PAPIs '17, talk abstracts and registration at papis.io/2017

Also, you can win a free pass to PAPIs '17 (or free access to presentations after the conference), by liking this post and leaving a comment ("+1" will do)!