David Schönleber

My new blog post on why many organizations fail to be data-driven is out: dschoenleber.github.io/challenges-bui… Taking inspiration from the book "The Art of Action", I trace back several shortcomings to inadequate org structure and muddled focus of work.

Fantastic post on the connection between statistical physics (Ising model) and probabilistic inference in machine learning (Evidence Lower Bound): jaan.io/how-does-physi…

An overview of hyperparameter selection & algorithm selection with big and small data by @dwschoenleber 👌 bit.ly/2L0n1o3

Another great overview over the modern data ecosystem (spanning both BI and AI) by a16z. It's remarkable in that it takes on a bird's eye view while not glossing over the complexity involved. a16z.com/2020/10/15/the…

Great overview over the 2020 Data & AI Landscape: mattturck.com/data2020/ I'm not referring to the picture here, but to the blog post, which reviews the "modern data stack", (new) tools for managing its complexity, and recent developments in the (enterprise) AI sector.

Extensive (> 150 slides) report on the development in AI during the last year along the dimensions research, talent, industry, and politics. If you're not yet sure whether you want to delve deeper, there's an executive summary on slide 7 :)

Great essay on the dependence of algorithmic progress on available hardware & software. It also touches on the "bigger is better" paradigm in deep learning. Recommended read!

Great post on the difficulties involved in creating enterprise ML platforms. The proliferation of ML tooling observed today is an indicator for a lack of agreement on the (minimal) set of capabilities required in a mature platform. Maturity takes time. towardsdatascience.com/the-problem-wi…

I’m not sure how to best tackle this problem, but regulation certainly is a part of it. (Maybe this will be Europe's contribution to shaping the future of AI.) newyorker.com/tech/annals-of… 4/4

At a time where GAFAM (the five biggest tech companies) make up for 20% of the stock market’s total worth, it’s worrisome if best-in-class AI models are build on the premise of scale - technologically and economically. nytimes.com/2020/08/19/tec… 3/4

With respect to (deep learning based) AI performance, it currently seems that “bigger” is the “best” (fastest / easiest) way to “better”. incompleteideas.net/IncIdeas/Bitte… (make sure to also read the response rodneybrooks.com/a-better-lesso… to the “bitter lesson”) 1/4

Insightful post by a16z on the challenges in ML arising from long-tailed data distributions (vs. traditional software engineering) and useful approaches on how to tackle them: problem scoping & break-down, reduction of model diversity (amongst others) a16z.com/2020/08/12/tam…

A growing body of work focuses on striking differences between current ML models and biological intelligence. We review the literature and argue that many of the most iconic failures can be understood as a consequence of the same underlying principle: “shortcut learning”

“Much of the time no one steps in to say, ‘Wait—we just did this five years ago.’ If they did, perhaps we’d see tech stacks change at a more reasonable pace—and work with more stable architectures as a result.”—@vboykis increment.com/software-archi…

If you're interested in the differences between traditional software and AI businesses, you should read this great post by A16Z on this topic: 👇 a16z.com/2020/02/16/the… "The fact that we’re seeing unfamiliar patterns in the data suggests AI companies are truly something new"

Pandas 1.0 is here! * Read the release notes: pandas.pydata.org/pandas-docs/ve… * Read the blogpost reflecting on what 1.0 means to our project: dev.pandas.io/pandas-blog/pa… * Install with conda / PyPI: pandas.pydata.org/pandas-docs/ve… Thanks to our 300+ contributors to this release.

Bainbridge's 1983 classic on the "Ironies of Automation" seems more relevant than ever as our automated systems continue to grow in scale and complexity: blog.acolyer.org/2020/01/08/iro…

Timely, relevant post on the importance of career ladders (and, in upcoming posts, on advices in this regard). Particularly in the field of data science with its plethora of incommensurable job titles, it’s a good indicator of team strategy and maturity. locallyoptimistic.com/post/career-la…

We are very excited to announce that we have open-sourced Metaflow, our human-friendly Python library for Data Science! We worked together with @awscloud to ensure that the open-source release is easily usable outside @Netflix. Learn more at metaflow.org