OLDaily

What is federated learning?
Ben Dickson, TechTalks, 2021/09/01

"The main idea behind federated learning," writes Ben Dickson, "is to train a machine learning model on user data without the need to transfer that data to cloud servers." We can see immediately the advantage of this. It creates a scenario where it is possible to enjoy the benefits of AI without the cost of surrendering your privacy to an all-seeing central server. Now this isn't automatic. There's a need for a back-and-forth exchange of data to ensure that the machine learning taking place at the edge is consistent with the model at the center. But, "advances in edge AI have made it possible to avoid sending sensitive user data to application servers. Also known as TinyML, this is an active area of research."

Web: [Direct Link] [This Post]

Sense Data
Gary Hatfield, Stanford Encyclopedia of Philosophy, 2021/09/01

I see something in front of me; I recognize it immediately as a yellow sponge. But I do not know it immediately as a yellow sponge; I have to know about things like 'yellowness' and 'sponges'. So there's something that is what I perceive, but is not itself a perception of a yellow sponge. What this is, according to the theory, is 'sense data'. But what is sense data? Is it direct awareness of a patch of colour? Of a three-dimensional object? This article explores the concept through history and gives us a sense of how people have struggled with the concept.

Web: [Direct Link] [This Post]

I wish the fediverse had ‘circles’.
Ru Singh, 2021/09/01

I've thought this too. "You get to decide who is in a circle and then you decide who you share a post with, which could be circles, individual people (maybe also referenced by email) or make it public." It's hard to make work, though. "It did sound great to all of us, but failed because the bookkeeping was unmanageable and a detriment to socializing…on a social network!" I think we'll get it right when we're able to employ a personal AI to do the hard work of defining circules and assigning the right posts to the right people. We're probably not far away.

Web: [Direct Link] [This Post]

Commercial applications of quantum computing
Francesco Bova, Avi Goldfarb, Roger G. Melko, EPJ Quantum Technology, 2021/09/01

NRC has recently launched a new Applied Quantum Computing Challenge program, led by Phil Kaye. I attended a number of his talks on the subject over the summer and can say, based on what I saw, that quantum computing is (a) a real thing, and (b) potentially useful. This article highlights the commercial possibilities of quantum computing. "The biggest promise of such computers," write the authors, "lies in solving large combinatorics problems.... combinatorics problems ask the question 'how many ways can this set of objects be combined?'" Which (surprise!) takes us straight into graph theory. Now while this work is all beyond what I've studied thus far (an advanced degree in mathematics would be helpful) I am nonetheless led to speculated about the possibility of quantum learning theories in neural networks, and hence, new ways of looking at learning in general. Alas, I do not have a second lifetime to devote to this...

Web: [Direct Link] [This Post]

Must-Read Books on Online Learning
TeachOnline.ca, 2021/09/01

This is a looong list of 173 books, organized alphabetically. None of them are mine, though to be fair, this is a list of real books, published by real publishers, and not self-published collections of blog posts. And I'm in good company; there's nothing by Tony Bates in there either. Nothing by Terry Anderson. Nothing by George Siemens. Clayton Christensen, Salman Khan, Will Richardson, Gary Stager, Seymour Papert, Jay Cross, Don Tapscott (to name only a few). Curt Bonk is there, but only for an edited collection. Weller is there but missing is the much more important The Battle for Open. So, while it's a long list, it also strikes me as an odd list.

Web: [Direct Link] [This Post]