Episode 154 – Roaring News

In this News installment, coming to you courtesy of Dave and Endgaget, we talk about how the Osaka track is paving the way for data free flow across borders and take a look at the alleged problem with the UK’s facial recognition system. Ending on a “high note”, we discuss how Facebook and Googles (and a lot of other’s) MLPerf benchmark is going to change the way we look at our machine learning setups. Or not…

Data without frontiers!

Have world leaders woken up and smelled the coffee that is reality or do they still believe data flow is something that stops at a border check point?

The Osaka track that was initiated by Japanese Prime Minister Shinzo Abe seems to indicate enlightenment is at hand.

However, real concrete information is rather hard to find so we engage in a bout of theory-crafting on this subject in the hopes of coming to a useful conclusion.

Wish us luck!

 

I’m sure I’ve seen that face before…

This article on how some people claim the UK’s facial recon system is quite bad and other people say it’s wonderful is a prefect excuse to talk about how context and full understanding of the underlying technology is often a “good” think when reading up on statistics and benchmarks.

It feels like every time somebody writes up an evaluation, they use (and abuse) the data such that it makes the point they are looking for. And can you blame them? of course you can, though that does not excuse us form getting educated on a subject before buying in to the rhetoric, no matter how fit-for-purpose it seems to be.

So yea: “lies, damn lies and statistics” is still very much alive it would seem…

On the subject of lies, damn lies and benchmarks…

Following SPEC and TCP, machine learned experts decided they need a propper benchmark suite of their own, thank yo very much indeed!.

Now the actual value to the end consumer of Machine Learning Goodness(TM) is more often than not negligeable, but “industry standard” benchmarking has definitely earned its place in the world of RFIs and RFPs…

This endeavor in benchmarking does seem to be a very ambitious project encompassing all of hardware, software and services (think cloud SaaS). Does this make MLPerf brillant, or doomed to failure?

And do we need benchmarking after all?

All good questions and the answers can probably not be found in our discussion on the matter…


Please use the Contact Form on this blog or our twitter feed to send us your questions, or to suggest future episode topics you would like us to cover.