In April 2023 I participated in the Hypermatrix Workshoparity.science organized by my friend and colleague Carlos Zapata-Carratala, and hosted by the Wolfram Institutewww.wolframinstitute.org. Other participants include Edinah Gnang at Johns Hopkins, Luke Oeding at Auburn Unirsity, and Joshua Grochow at the University of Colorado, Boulder.
My short talk was on “rainbow arrays”, a different paradigm for how to think about array programming and tensors that mirrors changes being advocated in the deep learning community under the rubrik of “named axes”. After setting up a functional perspective on arrays, the basic idea is to see the axes of a multidimensional array as being organized not in an ordered list or tuple but as a record with named fields – doing so eliminates the need for explicit transposition and broadcasting operations.
I have since developed this talk further into a blogpost series that starts heremath.tali.link.