An Introduction to Hashing in the Era of Machine Learning

Locality sensitive hashing (what you describe) is really useful for clustering and doing nearest-neighbor searches. Its used in digital fingerprinting, genome sequencing, and other interesting areas.

it's called semantic hashing: https://www.sciencedirect.com/science/article/pii/S0888613X0...

also see https://en.wikipedia.org/wiki/Sparse_distributed_memory

That's what word2vec is trying to model, it cannot do it because natural language semantics are person-2-world relationships.

It groups words together based on their distributional characteristics, and with the right parameter tuning, that can be a good-enough model of similarity.

Oh, cool idea. I had not heard the term Semantic Hashing before, but I'm going to do some research now.

Thanks for sharing!

Are you talking about cache utilization in the context of collision handling, or hash functions themselves? I'd love to learn more about cache-conscious hashing, could you suggest a paper to read, or algorithm to search for?

I Googled "Cache-Conscious Hashing" but didn't quickly find anything promising :(.

Thanks!

Submitted a PR for your SMHasher repo: https://github.com/rurban/smhasher/pull/41 .

I can't see how your statement can be true:

"In response to the findings of the Google/MIT collaboration, Peter Bailis and a team of Stanford researchers went back to the basics and warned us not to throw out our algorithms book just yet. Bailis’ and his team at Stanford recreated the learned index strategy, and were able to achieve similar results without any machine learning by using a classic hash table strategy called Cuckoo Hashing"

"We can do your ml hashing with ordinary hashing" is a statement about both ordinary hashing and ml hashing, indeed a fairly strong statement about each.

Whether it's true is a different matter but to say "nothing about ml hashing" seems unsupportable.

It's oddly formatted (I thought it was a quote at first), but the article includes this:

>(If you’re already familiar with hash tables, collision handling strategies, and hash function performance considerations; you might want to skip ahead, or skim this article and read the three articles linked at the end of this article for a deeper dive into these topics.)

The cuckoo hashing was new to me.

Some extra experimental evaluation of learned-indices: https://dawn.cs.stanford.edu/2018/01/11/index-baselines/

These structures are particularly interesting as they lend themselves well to visualization and interactive navigation. I was happy to see that Ed Chi is one of the co-authors and comes from an HCI / Info Vis background.