The Second Newsletter (May 2020)

Written by Jennifer Prendki

May 5, 2020

Hello again from Alectio! We’ve got some exciting stuff coming down the pike but we wanted to share a little of what we’ve been up to since the end of March in the meantime.

Here’s a collection of things we’ve been writing and some fun machine learning projects and resources that caught our eye this month:


How we can understand what data your model needs without looking at your model
The most frequent question we get is “how can you understand my model if you can’t see it or the data?” We understand why we get that question. After all, it’s a bit counterintuitive. We explain how in this post, with the help of a guy making a scene on the bus. It’ll make sense when you read it.

Here’s why you need a data collection strategy
Many businesses think “collect all the data” is a strategy. It isn’t. In fact, that’s a lot more like hoarding. In this post, we discuss how we leverage active learning to understand what data your model needs–and what data it doesn’t.

Using explore/exploit to build a better breed of active learning
Explore/exploit is a fascinating paradigm that’s central to reinforcement learning. But that’s not all it’s good for. Not even close. Find out how you leverage explore/exploit to order food and how we leverage it to create dynamic querying strategies and smarter models.

Just because the data is representative doesn’t mean it’s useful
When you’re training your model, are you listening to what it wants to learn from or making assumptions? Because if you’re just doing random selection or trying to tailor your training sets to be representative of the data you have, chances are, you’re doing it wrong.


Jukebox Neural Net
The folks at OpenAI published a paper on neural net that generates music. It’s a fascinating paper that comes complete with thousands of examples in myriad genres you can listen to. This Whitney Houston one is actually a jam.

AI Gahaku
Not a music fan? Fine. Here’s an AI that creates fine art. Or “fine art.” The results are pretty mixed but the best and worst attempts are a delight.

65 ML and Data Books: Free
Towards Data Science put out a great list of some free, downloadable machines learning resources. It’s a well-curated list that spans a lot of topics and, again, did we mention these are free?


That’s it for this month. Please don’t hesitate to reach out to this email or if we can help you with anything. Stay safe and stay healthy.

You May Also Like…

5 Pillars of Data-Centric AI

5 Pillars of Data-Centric AI

Currently, AI is the latest buzzword of the technology industry. Social media users and tech enthusiasts seem to...


Submit a Comment

Your email address will not be published. Required fields are marked *