Nice WiReD "5 Levels" vid on ML

Passing along latest WiReD “5 Levels” video, this on: Computer Scientist Hilary Mason Explains Machine Learning in 5 Levels of Difficulty

1 Like

I have seen it also. One thing that strikes me particularly is the phrase, “uneven applications of resources to problems”. The cost for R&D in advanced AI/DL is so prohibitive such that commercially valuable problems get most attention. Yet, for tinyML engineers, the most advanced models (with the greatest number of parameters or uses the most computational resources to train) are not usually of interest. TinyML concerns practicality, usability, efficiency, deployability… Hopefully, tinyML could be used to solve some problems that are not the most commercially valuable, but perhaps socially or humanitarian-ly.

My first take-away was Hilary Mason’s engaging interaction with the youngest participant. I thought she did a very good job motivating ML on a high level in an understandable way that seemed to resonate with that young girl.

The most interesting part for me was the last part — “Expert-to-Expert” — in which the ladies broached the point you made. (BTW, I wish that part hadn’t been edited as tightly as it was to reduce it to only ~8 minutes. They touched on a range of interesting topics…and then the editor cut to a later part of their conversation without a smooth seque or a satisfying resolution. I’d like to see the full Level 5 video.)

My main take-away there was something hammered home in the TinyML course: The absolute need for large amounts of good data; and not just collecting reams of it and then feeding it straight into a model, but first doing the unglamorous grunt-work of understanding the data’s provenance and exposing its biases. (And as my domain’s aquaculture, I was interested that they raised a related domain – smart agriculture.)

Regarding your point on the allocation of funding, that’s a challenging real-world problem across disciplines. I don’t know what sectors/applications you have in mind, but as you mentioned social/humanitarian problems, perhaps you can find a fit with some of the social investment groups that look beyond traditional measures of profit as a criterion for their participation? (Sorry, but I have no suggestions there.)