Author: admin

I’ve looked at clouds from both sides now
From up and down and still somehow
It’s cloud illusions I recall
I really don’t know clouds at all.

-Joni Mitchell

I had a wonderful conversation with a colleague this week about machine learning. Not about the specific algorithms and mathematics, but about the philosophy that makes ML tick — the general approach that makes it work as well as it does.

“Maching learning”, as some of you know, is an approach to heuristic algorithms (sometimes known by the sexier term “artificial intelligence”). When a problem is too difficult for a computer to solve by straight ahead computation, sometimes we resort to sneakier methods — approaches that try to look for shortcuts to a solution, and usually (but not always) find them.

What’s generally called “cloud computing” — looking at lots of examples of “things like this” by sifting through large amounts of data, and then using those examples to make better guesses about new things — makes heavy use of such shortcuts. For example, if you want your machine learning algorithm to recognize faces, you can “train” it by showing it lots of examples of photos “in the cloud” that somebody has already labeled as pictures of faces.

The conversation I had this week was about something a little more subtle: The fact that machine learning usually works because it uses information about big things to figure out something about small things, but also information about small things to figure out something about big things.

For example, early techniques for recognizing faces usually started by looking at a low resolution version of a picture and saying “hey, here’s a fuzzy blob that might be a face.” Then it looked at a higher resolution version of the same picture to check for things like eyes, nose and mouth in the proper place.

This didn’t work very well, because in a low res picture there are lots of fuzzy blobs that might be a face, but when you look more closely, most of them turn out not to be faces. Machine learning ups the game by going in both directions at once.

Not only does it look for faces, and check whether there are eyes and noses and mouths inside, but it simultaneously looks for smaller features like eyes, noses and mouths, and checks to see whether they are inside bigger features that look like faces.

The big power-up here is that we’re checking both “big to small” and “small to big”, looking in particular for connections that work in both directions.

It seems pretty simple when you put it like that. Yet this simple change in thinking has had a huge impact on our ability to use computers to recognize things.

In pictures it’s small
But up close it is really
Quite insanely large

I gave a talk today, and everything went really well until I ran out of time. It didn’t end up being a bad talk. I just ended up speaking a lot faster in the last five minutes than in the first twenty minutes, and rushing through things in the end.

There is no single answer to the question: “How long should I talk for?” Sometimes people want me to squeeze everything into half an hour, including questions, and other times I’m told: “We have the room for three hours. Use as much time as you want.”

So I’m thinking of turning my talk into an accordion talk — a presentation that can shrink and grow to fit the available time. Since I write all my own presentation software, this shouldn’t be too difficult.

The fundamental idea, I’m thinking, is to tag each slide according to how long a talk it would be part of. The most important “tent pole” slides would be in every talk — even the short 15 minute talks.

But other slides would be tagged by “40 minutes” or “60 minutes” — meaning that I should skip over that slide if my talk is supposed to come in at less than 40 or 60 minutes, respectively.

Yes there is still some fuzziness in this. The technique heavily relies on my knowing how long it takes for me to present various given slides.

But I think it would be far better than the sort of guessing that I do now.

I loved all of the thoughtful comments on my Artist / Scientist post. So many interesting terms and definitions! I confess, the first time I ever saw the word “polymath” I thought it meant somebody who is good at algebra, geometry and calculus. In that spirit, herewith a short lexicon based on a misinterpretation of some very nice words. Feel free to add more!

Ambidextrous: Able to use any sugar.

Circumspect: Examine from all angles.

Consequential: In reverse order.

Extraneous: Outside a locomotive.

Impeccable: Beak-proof.

Kinescope: Tool for visualizing your family tree.

Oxymoron: What you become when your brain goes without air for too long.

Parasympathetic: Approving of very lightweight aircraft.

Penultimate: Bic or Mont Blanc, platinum edition.

Ponderous: Thoughtful.

Postulate: Create blisters.

Proliferate: Argue against abortion.

Stagnation: Country of bachelors.

Today I heard yet another talk at a research symposium in which the speaker, when asked about the difficulty of doing research involving both aesthetic invention and technological invention, responded by saying that artists and scientists need to collaborate.

I hate this answer. It suggests that there are two different species of being: “the artist” and “the scientist”.

Fortunately, another speaker later in the day pointed out that the artist and the scientist can be the same person. And I completely agree.

In fact, when it comes to research, it is far better if they are the same person. In industry the value proposition may be different, but in research, combining these two complementary forms of problem solving within a single brain is a huge win.

Yet I realize, thinking about it now, that we don’t have a good word for the person who is both an artist and a scientist. Maybe we should come up with such a word.

I’m open to suggestions!

Every few years I make sure to revisit the Musée Rodin. I love the great man’s sculptures, which are, as I’m sure you know, inspiring and magnificent.

But the real reason I have always gone is to see the work of Camille Claudel, his sometime student and sometime lover, and an artist whose work has breathtaking power and beauty, and an aesthetic subtlety that can be lacking in Rodin’s art.

Alas, during this visit I discovered that only the garden was open, not the interior of the Hôtel Biron, where the smaller works are traditionally on display, and which is currently under renovation. It is the Hôtel which houses the works of Claudel.

But then I found out, to my delight, that a selection of the works from the indoor collection are on display in a temporary structure that has been set up for just this purpose. Imagine my glee in discovering that I would once again be able to commune with my beloved Claudel sculptures.

Alas, whoever was charged with selecting works to display from the indoor collection had assumed that we clueless tourists would be interested only in Rodin alone. The work of Camille Claudel was nowhere to be found.

Fools.

Finally went to see Age of Ultron. Yes, it’s wall to wall action packed, with beautifully choreographed fight scenes, flip dialog between the battles, awesome visual effects, and split second editing to take your breath away.

But what surprised me is that it’s not, you know, an actual movie. Not in the sense that, say, Richard Donner’s Superman was an actual movie. Watching Ultron, I felt as though I had been dropped into the middle of something — and I guess in a sense I had been.

Yes, I know that at the level of an Avengers movie it’s less about gold statues, and more about gold. With a budget so enormous there is no margin for financial error. But for me the gold is starting to rust. I had somehow thought that Joss Whedon would find some way to surprise us.

Of course everybody sees a movie for different reasons, and you might like this one quite a bit. As a large screen action extravaganza it certainly gets the job done. So maybe it’s me — maybe I need to lower my expectations.

Or maybe by now Whedon is only doing the Avengers to buy himself freedom to work on smaller projects that he actually cares passionately about. And maybe that’s ok.

After the movie, my friend and I were comparing notes. I said “Well, that wasn’t Chaucer.”

Then I thought about it some more. “Unless,” I added, “you’ve seen A Knight’s Tale. In which case, that was Chaucer.”

Speeding across the European countryside on the TGV today, my trip was mostly uneventful. The day was clear and beautiful, and the scenery magnificent. I was even impressed by the lunches on display.

Europeans, possessed of a certain sense of style that we Yanks often lack, really know how to pack a picnic for a long train ride (the right wine is very important). All around me, people sipped their wine and enjoyed the rolling view outside, as they quietly conversed in French or German.

Then, around two hours into the ride, everybody started to notice a plaintive meowing. Conversation died down — the sound was unmistakable. Apparently a small kitten had gotten onto the train. Passengers started to look around, trying to figure out where it might be hiding.

Finally a man stood up, reached into the overhead luggage rack, and took down his backpack. At that moment we all realized the sound was coming from the bag. I was both astonished and dismayed. What sort of person makes a little kitten spend a three hour train ride inside a backpack?

All eyes were glued to the backpack as the man reached in with one large hand, digging around inside.

Finally, he pulled out a small black cellphone, which he handed to his young son, who was looking extremely sheepish and embarrassed. The boy switched off the alarm, then his father put the phone back into the bag, placed the bag back up in the overhead rack, and we continued on our way.

Today a colleague told me that she was impressed with something she had seen in the Oculus Crescent Bay demo — an interactive flying dragon. Not a real dragon of course. But because the virtual dragon was rendered in full surround VR, she said she felt a visceral sense of danger, even though she knew it was all make believe.

We then discussed the connection with the Lumiere brothers’ famous 1896 train sequence, and whether we might now be witnessing the last moment in history when anybody would be able to achieve that sort of effect in VR. In another few years it is likely that what she experienced may become as ordinary and expected as a shot of a train coming toward the camera.

She then told me that every time the dragon flew at her in the demo, she involuntarily dropped down, and looked for something to hide under — even though she knew intellectually that this made no sense.

Then I had an “aha!” moment. “Don’t you realize what you were experiencing?” I said.

“No,” she replied, “what was I experiencing?”

“That was an example,” I said, “of a dragon drop interface!”