When I go to a bookstore, unless I'm checking to see if the new Analog and Asimov's are finally out yet, I don't know what I'm getting or even if I'm going to get anything. That's even more true when I visit a used bookstore, the sort of place that made me aware of the possibilities that were out there to begin with. Walking the shelves in a magazine store leaves you open to discovery, primed for finding things you didn't even know you were looking for. In a store you look this way and that, bouncing from thing to thing, making discoveries you would never have thought to look for yourself. Perhaps you didn't even know they existed.
In contrast, the online catalogs I've used on the exceedingly rare occasions I've bought books off the internet - cases in which the particular book I was looking for just couldn't be found anywhere in the physical realm - have been set up with the assumption that you already know what you want. None of the online directories I have experience with are able to replicate the ease of browsing, of having your attention drawn by a particular book's spine or title or cover. Sometimes it's the unexpected finds that are the sweetest, like when I found a weathered copy of The Third Industrial Revolution in Powell's, a book that's been out of print for decades.
The migration to electronic readers takes away a lot of that. Nothing has to go out of print when it's digital, and there goes the thrill of finding something that's hard to find. So too goes the ease of serendipity.
When computer scientists at Google’s mysterious X lab built a neural network of 16,000 computer processors with one billion connections and let it browse YouTube, it did what many web users might do — it began to look for cats.
The “brain” simulation was exposed to 10 million randomly selected YouTube video thumbnails over the course of three days and, after being presented with a list of 20,000 different items, it began to recognize pictures of cats using a “deep learning” algorithm. This was despite being fed no information on distinguishing features that might help identify one.
Picking up on the most commonly occurring images featured on YouTube, the system achieved 81.7 percent accuracy in detecting human faces, 76.7 percent accuracy when identifying human body parts and 74.8 percent accuracy when identifying cats.
“Contrary to what appears to be a widely-held intuition, our experimental results reveal that it is possible to train a face detector without having to label images as containing a face or not,” the team says in its paper, Building high-level features using large scale unsupervised learning, which it will present at the International Conference on Machine Learning in Edinburgh, 26 June-1 July.
“The network is sensitive to high-level concepts such as cat faces and human bodies. Starting with these learned features, we trained it to obtain 15.8 percent accuracy in recognizing 20,000 object categories, a leap of 70 percent relative improvement over the previous state-of-the-art [networks].”
The findings — which could be useful in the development of speech and image recognition software, including translation services — are remarkably similar to the “grandmother cell” theory that says certain human neurons are programmed to identify objects considered significant. The “grandmother” neuron is a hypothetical neuron that activates every time it experiences a significant sound or sight. The concept would explain how we learn to discriminate between and identify objects and words. It is the process of learning through repetition.
“We never told it during the training, ‘This is a cat,’” Jeff Dean, the Google fellow who led the study, told the New York Times. “It basically invented the concept of a cat.”
“The idea is that instead of having teams of researchers trying to find out how to find edges, you instead throw a ton of data at the algorithm and you let the data speak and have the software automatically learn from the data,” added Andrew Ng, a computer scientist at Stanford University involved in the project. Ng has been developing algorithms for learning audio and visual data for several years at Stanford.
As I am reading, our cat, who happens to be called Leibnizcomes expectantly into the house from the garden. But what, if anything, does he understand by his own name? He also recognizes other designators, such as "cat" (in English and Dutch).
The classical view is that animals learn their names through classical conditioning, viz., what they learn is to *respond* to the name, not recognize themselves as such. Positive reinforcements such as cuddles and treats teaches the animal come to the owner whenever they hear their name. Similarly, the animal learns that if it is in a situation that the owner did not like in the past (e.g., trying to steal food left on the kitchen stove or opening the dustbin), it runs away as soon as it hears its name since the past conjoining of name + bad situation was negatively enforced.
The problem with this view is that recent work has shown that animal (in particular dog and parrot - not much work on cats because they are notably uncooperative in experimental settings) language learning skills are far more sophisticated. Dogs, for instance, can fast map new words for unfamiliar objects. They do this by reasoning by exclusion: if asked "fetch the dinosaur" and presented with a heap of objects, one of which does not correspond to a word the animal knows, the dog will take the dinosaur and remember this word for months to come.
Also, animals have sophisticated conceptual understanding - more sophisticated than classical behaviorism + conditioning has it. Moreover, animals such as chimps, dolphins and even sea lions have shown capacities to learn to map symbols to concepts. Bottlenose dolphins in the wild have signature whistles to denote each other.
So if a dog can learn the word "ball" by fast mapping a linguistic expression to a concept, why would the dog not similarly learn to fast map his name to himself? Typically (this is anecdotical), dogs learn their names really quickly, and at any rate our cat learned his name within a few days. We tried operant conditioning to teach him other simple things and that took months and months.
Two councillors leading the TTC say it’s time to move ahead on an accumulating wish list of transit projects: 175 kilometres, including six subway lines, 10 LRTs and five bus and streetcar routes across the city.
The latest proposal would dramatically expand Toronto’s transit network over the next 30 years. And while it comes with a $30 billion price tag, it is well within the city’s grasp, say the councillors at the helm of the Toronto Transit Commission.
TTC chair Karen Stintz (Eglinton-Lawrence) and vice-chair Glenn De Baeremaeker (Scarborough Centre) believe residents can be persuaded to make the kind of transformative investment that would save Toronto from descending into a prosperity-crushing, gridlocked future.
They are calling their proposal OneCity and are asking council to approve a staff study of the plan in July. Councillors would then have until October to take the plan to their constituents before considering approval.
“What this does is it clarifies for the city of Toronto what our network is, how it fits into the regional context and how we propose to fund it,” Stintz said. “The funding, if approved, is dedicated, dependable and debt-free.”
[. . .]
The first funds from the Stintz-De Baeremaeker plan would go toward converting the SRT route into a subway line, at a cost of about $2.3 billion. The project would have a head start from the $1.8 billion the province has already committed, Stintz said.
Although a subway would be routed somewhat differently from the SRT, it would have the advantage of not shutting down the SRT for four years — unlike the current plan, which calls for putting SRT riders on buses for that period, while the new LRT is built.
[. . .]
The second priority in the new OneCity vision would be an east waterfront LRT, at a cost of about $300 million. Waterfront Toronto has allocated $90 million toward transit on the lakefront east of Yonge St., and developers there have been bracing for a temporary transit solution such as bus rapid transit.
All the lines in the OneCity plan have been approved at one time or another, and in some cases the environmental assessments have been done for years, said De Baeremaker.
Stintz said she’s had preliminary discussions with the province and the mayor’s staff about the plan, but no commitment of support.
OneCity rebrands some potentially divisive projects such as a downtown relief subway line, which has been renamed the “Don Mills Express” line.
It also aligns with some regional transit projects, including the air-rail link, which the councillors say could be converted to public transit by adding three more stops. A second set of tracks to GO’s Stouffville line would allow for a Scarborough Express above-ground subway or train that delivers riders from Steeles Ave. to Union Station.
![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
branchandrootdancing_serpent
