Following the recent controversy around the announcement of OpenAI’s “Solving the Rubik’s cube”, Vicki started investigating to figure out for herself whether the cube was hype or not. Doing so she got sidetracked and looked deeply into OpenAI.

As machine learning becomes more automated, human judgment still matters.

Glue work is valuable - without it, projects fall apart - tasks get dropped, teams miscommunicate, and it is just harder to get things done. However, there is a difference between valuable and valued. If glue work makes the team function more smoothly and effectively, but it is not valued, then that is a problem.

AI warfare is beginning to dominate military strategy in the US and China, but is the technology ready?

An overview of the ecosystem and tools shaping the future of notebooks.

Good analysts are a prerequisite for effectiveness in your data endeavours and are currently often under-appreciated: relentless curiosity, willingness to question, commitment to learning and communications skills are among the skills you should be looking for.

Hyperopt is an hyperparameter optimization library in Python. This new release now includes a SparkTrials class that allow you to scale out a hyperparameter tuning across a Spark cluster. This obviously covers random search, but the Tree of Parzen Estimators (TPE) is also included, which is no easy feat considering TPE is a Sequential Model-Based Optimization. Looking forward to give it a try.

Udacity just launched a new nanodegree to learn the fundamentals of stream processing, including how to work with the Apache Kafka ecosystem, data schemas, ApacheAvro, Kafka Connect and REST proxy, KSQL, and Faust Stream Processing.

A sharpening of your understanding of the difference between explainability and interpretability, and why the former may be problematic.

And some great pointers to techniques for creating truly interpretable models.

This short post explores the implications of a recent research paper about using an exponentially growing learning rate for training a neural network. You might be wondering how is this not going to blow up the weights of the net: it all comes down to the Scale Invariance induced by Batch Normalization.

This post explores a few cases where K-Means Clustering does not perform well or may produce unintuitive results: wrong guess on the number of clusters, highly dimensional feature space and clusters in strange or irregular shapes.