One way to do BUSINESS DEVELOPMENT could be through grant development to help bootstrap initial outputs.
A sensible place to get some grant funding connected into the themes around this work would be through Joe’s appointment as a Research Fellow at Oxford Brookes University’s Ethical AI Institute.
One of the constraints here is that that’s a 2-year appointment, and some grants (but not all) require the PI to be a permanent employee. But this constraint is something we could relatively easily work with.
EPSRC: Mathematical Sciences Small Grants
This is for projects of up to £80K.
"Inducing a mathematics encyclopedia out of Arxiv"
We talked about following on from the Open Information Extraction theme possibly in connection with the new preprint “Language Models are Open Knowledge Graphs” (and the associated video review). This is close to Deyan’s expertise and would build on his previous work.
Deyan: "Accurate entity linking is desired." (video min 43) -- hey, they honestly admit this in the paper! This was my stopping point with the jupyter notebook as well i believe, I would like to get the neural model to "ace" that part of the job before moving further. [...] ok, so after watching the entire review, I can conclude if I arrive at a reliably successful method following my initial notebooks, and apply them to arXiv and/or SE there's certainly a paper in there. Their suggestions from the paper aren't actually that useful sadly, just confirming some basic practical tricks (using last attention layer, constraining with beam search and using language priors for the triple content) that were already known
"Mining informal proof structure"
This would be a somewhat similar line of work but more along the lines of what we did in the FARM paper. We discussed the advant
Maybe we could move in another more formal (category theoretic) direction
Maybe we could build out some of the ideas that Zans and Ray have been discussing. Potentially this could also involve