The recent Thinking

Holidays from home back to school, also counted for a period of time. Some changes have been made during this period, and there are many new ideas about life.

The first is that learning should not be forced. The last semester of their own ideas are a mess, but also to force themselves to be free to go to the book, the result is naturally not high efficiency where to go. Read the computer for more than a year, hear the jokes naturally will not be less, the total feeling inside the life of ridicule is not quite right, including a variety of overtime. His own idea is that both work and study should be just a part of life. From the perspective of the whole life to make their own efficiency is high, is really to do.

I also do not know what the right life should be, I can do is often from the perspective of the whole life to reflect on, to see where they are not. For example, the freshman time to see my brother's life is to go home at night to do all kinds of drag, shower before you have to look at the DotA game what, until late at night to rest, but also often complaining tired. I say to myself, I have to constantly reflect on life.

Now my daily work is, 7:3 A.M. natural wake up, wash up to do some basic sports; 40 minutes nap (earplugs are really sleep artifact AH), evening after dinner mood is good to stroll around the circle, night 12 o'clock turn off the light to sleep. Saturday Evening Ring School run a trip, Sunday afternoon as far as possible out of campus stroll. At other times, you don't have to go to school without a job. Basically stay in the library. The activities of these, mostly pushed off the good. Immediately feel, ah, neither heart Plug and can guarantee learning efficiency, very good. Do not know a period of time after the face of the four subjects of the big homework can still be so calm, if so, perhaps the rest of the work can be used.

As for the thought of the outside world, in this time saw the "drunken footsteps", "thinking, fast and slow", as well as self-study abstract algebra and audit "mathematical Logic", "Western philosophy guidance" after these classes, have a lot of sentiment. Just tell me the conclusion. The following are my understanding.

In the inference of Proposition p to proposition Q, we often rely on relativity rather than causality. If causality, then its reasoning process should be a rigorous proof process, but we rarely do such proof. By contrast, we are more inclined to "find the law" in such a way as to get the inference that inductive reasoning. It's like relying on neural network algorithms to train a formula that is somehow able to fit the test data.

What is the situation with such reasoning? We can only draw a hypothesis about such inferences. This hypothesis can only be confirmed by a variety of "facts" (The History of understanding probability theory is understood), so that the probability of success of this inference is nearly 1 and not equal to 1. can refer to "Hume question". And once a counter-example is cited, its inference cannot be established.

So the science of "falsification" can be understood: the success of the current theory is only a temporary success, once the counter-example will be faced with a huge crisis to have to propose various amendments. The meaning of philosophy is not to think blindly, but to verify how this "correctness" of science should be understood, such as the premise of certain theories.

But my personal thoughts are a little different. In fact, it is similar to Hmm. Of course, I did not do the relevant work, but from the book to understand the model, and it has no mention in the Book of special Understanding. That is, even if there is a counter-case to prove that this inference is incorrect, it is not that we are all wrong, but that we lack the prerequisite to understand its success. Continuing, assuming that the data we are exposed to is all, then we can see it as the probability that this inference succeeds and continue to apply it. Of course, science can not mess up, this is a direct correction to do, but we do not need to do the computer. The reason that I can ignore the combination of the preceding factors in the HMM model is that the idea has been to unify the probability of the occurrence of the first factor into a pre-condition of impression. Alas, I still do not speak very clearly, but I do not know specifically about this, okay.

But the new data will always have an effect on the parameters trained in the old probabilistic model, and we need to fix it. But how about the specific correction? Bayesian method. "Western Philosophy Guidance" class recommended bibliography for, scientific reasoning-the Bayesian approach. I haven't seen it yet, but thinking is common, and there's not much doubt about it.

Then there is the proof of the proposition. The main thing here is to talk about some of the things that come to mind in logic class. Axioms are recognized as inevitable, and theorems are inferred from them. This high school math textbook is all about. What about the definition? I always thought that the definition is like the axiom of existence, you do not admit it will not be good fun to go on.

But now I think I was wrong. Defining performance is just a hypothesis, which, with some axioms, you can draw many conclusions. However, should such a hypothesis be set up in itself? It's not. Not all of them make sense. So, if this hypothesis is true, what effect will it have, and this is what logic is about. I didn't realize it at first, so I always wondered when I looked at Goodman's strange "blue-green" definition. After class to roll the platform back to the PPT re-look, just figured inside of the fishy.

What does the situation mean? One of the corresponding cases is, if we use the method of supervising training in the process of machine learning, can we have enough reason to doubt that the supervision itself is problematic? Is it wrong to define all kinds of things in our daily life? These definitions affect our lives, and when we reflect on them, do we have to reflect on them and change our minds fundamentally from our understanding?

If you've seen the words/mirrors, maybe you can see what I'm talking about. When I saw the introduction of the book from somewhere else, I was really shocked--and that was the effect. Yes, the definition is everywhere, it is contained in our language, and the logic in the language begins to influence our understanding of all things from the most basic places.

For me (at least for now), the most basic definition of mathematics, 0 and 1, exists only to justify the relationship in logic. That is, mathematics is actually a form of logic proof.

All right, keep talking. Personally, I don't think that the simple training parameters of neural networks are really machine learning. My ideas are probably not quite the same as the concept of being told. The question I have doubts about is:

How can we break through our own understanding of the world, just like the revolution brought about by the group theory to solve the equation?

Starting from a specific example of the ancient German paradox, for the sequence i:1, 2, 3, 4, 5 ...  The law may be k, or (k-1) (k-2) (k-3) (k-4) (k-5) + K. So how do we derive the probability of a hypothesis to be true from the many competing hypotheses, based on the data?

q-d question: When a hypothesis H and its auxiliary hypothesis A are denied by some evidence, how does the effect of such corroboration be distributed between H and a? For example, your own ideas can be synthesized by many hypotheses, and if you find yourself wrong, how do you reflect on yourself?

There are many such questions. Yes, what I want to explore is, what if we use data to quantify human understanding of the world and the process of understanding it? We can reflect on ourselves and make changes to our ideas, what about machines? These, I think is the real machine learning, is what I really want to do in this field.

But before I toss the computer, I'm afraid I'll have to think about myself. When I was learning something, I found a strange phenomenon: I was not sensitive to the rigorous reasoning process in abstract algebra, and the understanding of each concept in the course of learning was quite fast, and I always thought of the model to explain it when I saw it. Perhaps this "talent" makes me learn the data structure, once I understand it will no longer worry about the implementation of the Code, and I clearly did not write code.

However, I have to overcome this obstacle ah. There are different proofs of reasoning, which can be directly separated from the various methods. What are the basic principles of reasoning, and how do they combine to derive some complex and interesting reasoning processes? This way of thinking from a more grassroots perspective may bring you endless benefits. However, the reality is that they have not been able to concentrate on thinking, you have to be called to toss the backstage. Frankly speaking, since the "Mathematical logic" class and I saw the abstract algebra, one of the impressions is: we because of examinations, ignoring the most fundamental logic, and thousands of years ago Euclid's "geometry originally", really is we despise too much.

Ah, after all, is after thinking, finally in two or three hours written out of the text. is a comprehensive collation of his recent thinking, but not too focused on presentation. Please forgive me for this.

Recent Thoughts