Motivating the rules of the game for adversarial example research

Motivating the Rules of the Game for Adversarial Example Research is one of the most level-headed things I’ve read on AI safety/security in a while. It’s 25 pages, which is long for a machine learning paper — but it’s worth it. My brief take-away from the paper, which I totally support:

Conceptual issues in AI safety: the paradigmatic gap

Mid-term AI safety research fails to account for a potential paradigmatic gap: a major change in the concepts and algorithms which underly our learning algorithms. I argue that our beliefs about the value of present-day safety work ought to incorporate the possibility of a paradigmatic gap.

Do brains represent words?

An introduction to the problem of neural representation.

This is not an academic post

It's about breathing.

I saw a dog

How does our brain "fill in" the sparse visual signals provided by the eyes? A meditation on top-down vs. bottom-up perception.

LeCun: Language is the next frontier for AI—or not

A change of opinion hints at a change of fortune in the NLP community?

How to prepare for a Vipassana retreat

I'm not going to talk about my Vipassana experience. Here's why.

Sunday links

Week of 26 February

Bridging principles

I think it's necessary that linguistic meaning be derived from sources that are nonlinguistic. I call this claim a "bridging principle," and demonstrate parallel bridging principles in several subfields of philosophy.

Sunday links

Week of 6 November

Situated language learning

A paradigm for bringing about language acquisition in artificial agents

Sunday Links

Week of 18 September

On "solving language"

Stop telling me language is about to be "solved": We haven't even found the right tasks yet.

Hybrid tree-sequence neural networks with SPINN

An introduction to the SPINN model family described in our ACL 2016 paper. SPINN enables a novel hybrid tree-sequence (recursive-recurrent) neural network architecture.

Sunday Links

Week of 19 June

Life update

Coming back to blogging after another long hiatus

Conditional generative adversarial networks for face generation

I describe a recent project using generative adversarial nets to learn to draw human faces.

Machine learning and technical debt

Machine Learning: The High Interest Credit Card of Technical Debt is a great non-technical paper from this past NIPS conference. The paper claims that because machine learning systems are fundamentally different from traditional software, there are unique problems in incorporating them in otherwise maintainable, robust architectures. This analysis is useful for both the researcher and the practitioner.

A GloVe implementation in Python

More adventures in the land of word embeddings