ML’s lament

ML’s Lament 

(to the tune of ‘part of your world‘, from The Little Mermaid)

Screen Shot 2017-07-13 at 1.45.40 PM

Look at this net, isn’t it neat?
Wouldn’t you say that it’s Turing-complete?
Wouldn’t you say that this net
This net can learn anything?
Look at this graph, tensors unfold
How many layers can one network hold?
Lookin’ round here you would think
Sure, it can learn everything

I’ve got ReLUs and batch-norms aplenty
I’ve got convos and poolings galore
You want GPU stacks?
I got twenty
But who cares?
No big deal
I want more


Read more

Un Surprise

[This is not a post about quantum mechanics]

Imagine a world in which humans understood quantum mechanics.

Understood it intuitively, that is. Understood it in the same way that you can look at a face and recognize that it is not your mother. There are a lot of neurons in your brain working up a storm, going through the calculations that end up with you knowing it is not your mother. But “you” (the conscious thinking part of you) have no direct access to these computations, you are just aware of the end result.


Read more

[Cute Title]: [Serious Title]

Science paper titles are dusty affairs as far as the general reader is concerned. Take “Structural basis of lipoprotein signal peptidase II action and inhibition by the antibiotic globomycin”, an actual recent title. To people in the know that might sound terribly exciting, but outside the field it falls a bit flat.

Some authors try to spice their titles up, using a familiar phrase or riffing on one. This is followed by the but-seriously-colon, and a more solemn title. My current favorite is “Waiting for Merlot: Anticipatory Consumption of Experiential and Material Purchases” (Kumar et al. 2014).


Read more

A modest proposal for new journals

Three suggestions for journals with a different incentive structure than the current scientific publishing model.



Assume that ONCE is an established scientific journal that publishes broadly in the areas of physics, chemistry, biology, astrophysics, computer science, social science and so on. The journal uses basic peer-review that keeps out total nonsense. But beyond a sanity check and technical soundness review, almost all manuscripts get in.

The thing is, you can only publish in ONCE, once.


Read more