Babblings

About Statistics

Regarding the relation between ML and statistics, we should probably distinguish empiricism from statistics. ML is empiricism, which states merely the reliance on experiences, while could make use of them in more sophiscated manners.

About Black Boxes

From solid laws to probablities in physics, and from probablities to neural networks in AI, the further, the more nonsense, the better the effect. I'm now holding a reasonable superstition that the deepest secrets of the universe are hidden in probablities and causalities. My hope is reignited. There's a marvellous job on Hilbert's sixth problem.

About Education

There has started an education reformation in China, and I do have a lot of things to say. There are two facts about US education system: evaluation and admission criteria are subject to colleges, and colleges do care about demands of the society, both of which have been ... a bit running athwart in China.

About Spatial Vision

I think Berkeley basically stated that concepts such as "distance" were formed with joint senses. (By the way, if you still have doubts about the necessity to speak concisely, go read Berkeley's essays and you'll probably get it all. The last thing I did that was nearly as frustrating as looking for such a sentence to quote was debugging my CUDA codes.)

About P/NP

Is deep learning working like a non-deterministic Turing machine?

About Understanding

I have a feeling that this can also be associated with language theory.

Alright I was considering could math be a language so that Kant's opinion on math could be fake, until I knew that Wittgenstein had stated this. Math is a language. Reasoning can be possibly restricted in linguistics.

Algebra

François Viète invented algebra which greatly boosted the development of math and science. I'm curious about if this can be applied to deep learning to make it more seemly reasonable and tractable. (or doubting if algebra system really brings "reasons" :P)

About End-to-End Learning

I'm feeling it myself and also heard of it from a professor in piano -- if you want to play fast, rather than think about next movements, you should think about the melody of next phrase. Biological neural network do favor end-to-end occationally.

Now I realize that this can be possibly described as forming abstractions for the movements.

About Language Model

This can be a bit harsh but do be a problem if we are training on existing distribution, while speaking of this can also pull out a series of fundamental issues of machine learning which can be another name for statistics :P

It's indeed super impressive for current language models to capture grammar and possibly concepts so well. However, grammar is a finite set of rules, while what're even more charming can be the infinite possibilities behind reasoning.

The Living Trade

Basically we are all trading for living necessities with particular parts of our body, most of which will withstand wear or tear for use. However, brain can be considered as one of the rare exceptions that will even normally be polished after use. This leads to two benefits of living on brain -- sustainability of the trade and its continued growth.

Obviously I didn't see that every work requires a brain, but it still makes sense on the perspective of wear versus polishment. They are like bases being below and above one in an exponential function. Theoritically if ones start with wearing themselves for living at the very begining, they will never get the chance to excape. However, things will get more interesting if AI gets involved, for both those're embodied and those not.

About Theory

Theories have no power. Theories are only recording how we can make use of the power of nature, just like magic spells.

Well I'm starting to wonder if caloric theory can be true...or how I can enable computer to analyze this for me.

Nature don't do arithmetic. It's human that does.

About Capital

Now I get the difference between socialism and capitalism. Basically, making significant social impacts will normally align with the growth of capital in capitalism, which will not be true in socialism. Socialism is designed to prevent the capital from having vital impacts on society.

About Neural Network

Now I get why they were saying that neural network didn't work. Because it really doesn't work.

Now I get it better. It's all about dark magi... sorry, initialization. Just like clustering.

How Greatness is Daunting

Leaders are born in sobriety and thus pain. That's how greatness are "thrust" upon them according to Shakespeare.

Social Objectives

Every field has its objective, such as levels in video games, power in politics, and assets in business. In academia, that would be discoveries and creations leading to more possibilities for the future, which ideally drive researchers with their curiosity, pride, belief, and responsibilities.

Now I get why I'm saying this. What I really wanted to say was that I found working in all these fields are subjectively the same, and the only difference would be the objectives. In addition, objectively, the objectives are not pari passu. Some are having dominant forces over others, which people would normally ignore when they are working.

About Supervised Learning

Not until now do I realize that the terms "supervised" and "unsupervised" are actually referring to the supervision's reliance on human labor instead of the model itself. Models would be supervised by their objectives in both cases. Hence, it can be properer to name them as "human-supervised" and "non-human-supervised". (just kidding)

My Math Ability

Previous to my junior year, I had to form an image for everything before I allowed myself to memorize any conclusions. I think that can explain why Einstein's being not interested in math at first, as well as my failure in my math final exams in sophomore year. I'll never try such stupid stuff as forming graphic understanding for every single topic again. There shall be more strategies to store and retrieve concepts. In fact, even physics topics nowadays have gone far beyond intuitive empiricism, and Einstein did spend years getting used to Remannian geometry.

From Philosophy to Science

Bacon is considered as the founder of "experimental philosophy", which is exactly what "science" means, I suppose. A lot of people are confused by the role of philosophy nowadays, and not aware of their being philosophers themselves, I suppose. Science is no more than a methodology for philosophy, and that's where the name of the degree "doctor of philosophy" comes, I suppose.

About Divination

This is exactly how divination is working, which has also been confirmed in quantum physics, I suppose.

I've heard about some interesting recent math studies and trying to figure out what they are talking about.

Being Einstein

Actually there's a saying in China that people with stronger noses are more stubborn.

About Knowledge

Knowledge can be scaffolding for the outside, as well as cages for the inside.

About Growing

Lotuses grow in mud, and that's how they gain their beauty.

An Excuse

The more I learn, the more precise I find it is.

Now I get a better explanation for this. Truly smart people are learning more to make stuff in their mind less.