My method for generating startup ideas is notice the things that I hate. For example, I hate shaving, and would kill for a device that I can stick my face into and become clean-shaven one minute later. I also hate choosing the shower temperature -- I want the shower learn the temperatures I like. I also hate not knowing the definition of terms in government websites. I can't stand passwords. Etc. This feeling is my indicator of a good potential startup idea.
Many comments expressed concern about the alleged inappropriateness
of the title. Even the no-free
lunch theorem has been invoked, and words like SVM mentioned.
However: The original title, "Neural Networks officially best at object
recognition", is much more appropriate than the current title, because
it is by far the hardest vision contest. It is nearly two
orders of magnituder larger and harder than other contests,
which is why the winner of this contest is best at object recognition.
The original title is much more accurate and should be restored.
Second, the gap between the first and the second entry is so obviously
huge (25% error vs 15% error), that it cannot be bridged with simple "feature
engineering". Neural networks win precisely because they look at the
data, and choose the best possible features. The best human feature
engineers could not come close to a relentless data-hungry algorithm.
Third, there was mention of the no-free lunch theorem and of how one
cannot tell which methods are better. That
theorem says that learning is impossible on data that has
no structure, which is true but irrelevant. What's
relevant that on the "specific" problem of object recognition
as represented by this 1-million large dataset, neural networks
are the best method.
Finally, if somebody makes SVMs deep, they will become more like neural
networks and do better. Which is the point.
This is the beginning of the neural networks revolution in computer vision.
Actually, it does, since the difference in performance between entry #1 and entry #2 is so huge (25% error vs 15% error!), and since this is by far the hardest computer vision challenge yet!
Sorry for disagree, but it seems more related to the fact that they are using deep convolutional learning rather than the neural network itself. If you use an ANN with the same set of features side by side with a SVM you will see very equivalent results.
I will be more agree with a title like "Deep Convolutional learning overperformed traditional techniques in Object Recognition"
No... but I'd bet that if you use the high dimensional features resulted from the deep convolutional learning process as an input of an SVM the difference would not be that significant.
Well yeah, but then you're basically putting the meat of the NN algorithm into the SVM. I'd call the resulting algorithm a neural network with an SVM frosting. You might as well train naive bayes directly on the final nth layer of the NN instead of SVM on the (n-1)th layer, would be an almost equally weak argument for the thesis that NNs are not superior to the other algorithms on this task, since basically all the power is coming from the NN.
I stand corrected on SpaceX... yes it seems Musk founded it.
But on Tesla Motors the fact of the matter is that it was started by engineers (Eberheard and Tarpenning) and control was taken away by an investor (Musk)
Some facts:
1. The first wikipedia page review on Tesla Motors from Jue 12, 2006 [1]:
"The firm was started in 2003 by engineers Martin Eberhard and Marc Tarpenning....
Tesla has also managed to secure initial funding from prominent investors, such as PayPal co-founder Elon Musk, and Google co-founders Sergey Brin and Larry Page."
RedHerring Article July 8 2006[2]:
"Tesla Motors said earlier this month that it has raised a $40-million Series C round of financing led by VantagePoint Venture Partners and Elon Musk"
Finally from Eberheard himself [3]: "Mr. Musk was one of the leads I followed."
From what I remember from The Paypal Wars (highly recommended):
There was a culture clash when X got bought by Paypal. Elon was in charge of X and became the CEO of the combined company (as Peter Thiel took a leave of abscence) and wanted to switch from Unix to Windows. Max Levchin, the main tech guy at Paypal hated that idea. Elon initially got his way, and Max got marginalized. He then started looking into fraud cases and discovered that fraud, although low as a percentage of revenue, was growing rapidly. Max then persuaded people that the #1 priority should be to fix the fraud issue before it would kill them, and that they didn't have the time to worry about the technology stack. Elon Musk got kicked out, Peter Thiel came back to Paypal as CEO and Max managed to squash the fraud problem. Their competitors either failed to get traction or got crushed by fraud. With Paypal as the last one standing they won and got bought by Ebay.
These findings should not be misinterpreted, since it is indeed impossible to reach any goals. A person in their late 20s is extremely unlikely to become an olympic swimmer or a squash player because of physical limitations. And some people will relentlessly try and never succeed.
But believing that you can change is useful because it makes it easier to persist in my efforts to change. Because if I believed that change is impossible, I'd give up on the spot.
I found that I simply cannot believe a statement like "I can get much smarter" or "I can get much better at X", but I found that I can easily (fully, honestly, without reservations) believe that "I can get a bit more smarter", or "my intelligence is sufficient for mastering this material, so I need to push harder", or "I can get at least a bit better socially." These beliefs motivate me and make it easy for me to do the work even when it looks like progress is nonexistent. This is the meaning of believing that you can change.
Some people are extremely competent but are risk averse, so they don't take the risk that's necessary for success. Others are risk tolerant but are not very capable, so they are unlikely to succeed despite taking the risk. The trick is to be in both groups simultaneously. So if you know in your heart of hearts that you are competent, then it is worth taking the risk.
Robots are currently unable to be autonomous in any meaningful way. They still have a limited ability to perceive because computer vision is not good yet. They lack the ability to act sensibly in novel environments. Building such robots is the holy grail of AI, and it'll take a while before they are built.
A degree in mathematics requires a very large amount of effort and discipline, especially given your other obligations. Is this effort best spent on a mathematics degree, or maybe you could spend this effort differently and get what you want faster?
While point a) is a good reason to get a mathematics degree, points b) and c) are not. For point b), machine learning and statistics are much more appropriate than mathematics, and for point c), it is worth knowing that machine learning requires a fairly small subset of the mathematics you'd learn in a math degree. For example, a math degree covers many areas of mathematics (such as a heavy focus on proofs, abstract algebra, complex analysis and topology) that have no bearing on statistics and on practical machine learning. Conversely, a math degree also does not focus on statistics and probability, which are essential for data analysis.
Thus were I in your shoes, I would only study the math that is necessary to understand statistics and machine learning, and would start taking a machine learning course. The only math you need is multivariate calculus, linear algebra, and probability.
