Lambda the Ultimate

A new paper from Google deep mind; abstract:

We propose the neural programmer-interpreter (NPI): a recurrent and compositional neural network that learns to represent and execute programs. NPI has three learnable components: a task-agnostic recurrent core, a persistent key-value program memory, and domain-specific encoders that enable a single NPI to operate in multiple perceptually diverse environments with distinct affordances. By learning to compose lower-level programs to express higher-level programs, NPI reduces sample complexity and increases generalization ability compared to sequence-to-sequence LSTMs. The program memory allows efficient learning of additional tasks by building on existing programs. NPI can also harness the environment (e.g. a scratch pad with read-write pointers) to cache intermediate results of computation, lessening the long-term memory burden on recurrent hidden units. In this work we train the NPI with fully-supervised execution traces; each program has example sequences of calls to the immediate subprograms conditioned on the input. Rather than training on a huge number of relatively weak labels, NPI learns from a small number of rich examples. We demonstrate the capability of our model to learn several types of compositional programs: addition, sorting, and canonicalizing 3D models. Furthermore, a single NPI learns to execute these programs and all 21 associated subprograms.

Seems much closer than before!

A blog post on how programmers think. Tidbits:

Everybody in this country should learn to program a computer… because it teaches you how to think.

Nobody would argue with that.

This may be why brain imaging of programmers shows that programming relies more heavily on the language centres of the brain versus the areas associated with maths. Programmers will rely on the ventral lateral prefrontal cortex, alongside their working memory.

Something that will debated forever. That programming is more about language (words and such) than math is why I believe OOP continues to be so successful. Programming simply isn't math.

All this happened in my head. When I got home, I simply wrote out the code, copy and pasted some parts I couldn’t remember and ironed out the bugs until it worked smoothly. The actual hard part happened away from the computer in this case. You even ‘run’ the programs in your brain by following the logic through step by step to see if it would work – you can this way flag up bugs (gotta’ catch ‘em all!) before actually creating the code.

Programming environments aren't really optimized for thinking. It would be interesting to see if "thinking" and "programming" could be combined for a more augmented experience.

We use our working memory to store information, so when you’re imaging what a line of code does, you have to store the variables and the ideas you’re testing out, there. So when you’re thinking of a sequence of events, you need to keep the line of logical reasoning held in your working memory – which is where the similarity to math comes in.

...

Here’s also where things get more interesting still: because the brain areas involved with abstract thought are actually the same as those associated with verbal semantic processing (1). But then of course it does – language represents ideas just as code represents logic.

True, I wonder if our own internal working memory can be reflected back into the computer...again augmentation...this is the main point of live programming.

It’s probably down to the very immediate feedback loop that coding provides: the ability to test what you’ve written and immediately see the results is incredibly gratifying. Dopamine is the neurotransmitter that our brain releases in anticipation of reward and I’d argue coders have got dopamine to spare. This is the definition of ‘the work being its own reward’.

Again, there is great value in even better more immediate feedback loops. Feed the Dopamine! And ya, lots of caffeine.