3 ms·
Same or Different? AI Can’t Tell
- PaulHoule 5y agoFor a long time I have been interested in developing better similarity functions (e.g. are these two documents or images similar? is this document similar to that query?) The status quo is "glass half empty" from my point of view and methods under research such as Siamese networks have not manifested the promise they might have. Roughly, measures like tdfif/bm25/dfr don't know the difference between feature similarities that matter vs. ones that don't. I was the bagman for the data for one of the early learning-to-rank papers and it earned the PI a trip to the Googleplex but, no, I don't think that is good enough.