Today I encountered this amazing website, which is a medium for presenting research in a new way. It is actually similar to the Medium website but just for the machine learning articles and the blog posts are peer-reviewed. When I checked the steering committee I found out why this is so amazing. Yoshua Bengio, Andrej Karpathy, Ian Goodfellow, … are members of the committee. Unfortunately, they didn’t continue the project but it still can help a lot.

So for instance I found this introduction about Gaussian Process very interesting. Check it out.


@SingularThinker

The Simplest Math Problem No One Can Solve - Collatz Conjecture

We are stupid, aren't we?

This simple problem has some sort of magic and as always Vertasium narrated it aweosmely.

Watch the video here.

#video #math
@SingualrThinker

Can you answer this question?

🔗 Link
@SingularThinker

#meme :)))))
@SingularThinker

I decided to share more with you in order to interact more with each other and enjoy the following discussion. Hopefully, this will work out well. I will write mostly in English, but feel free to write in Persian or English.

For the beginning, let's start with this tweet. What do you think? Do you agree with that?

P.S: Check the mentions.
🔗 The Bitter Lesson Blog by Rich Sutton link

#Tweet
@SingularThinker

Can LLMs Separate Instructions From Data? And What Do We Even Mean By That?

I saw this thread on X and get very excited.
I didn't go through the paper, but I liked the way of thinking and modeling the problem with information theory. It was inspirational. In addition, imo the main problem is very related to the Godel Incompleteness theorem and/or the halting problem. In a nutshell, the system could interpret a sentence both as an instruction and data, which is the underlying cause of many self-referential paradoxes. Interesting idea for doing research.

#language_modeling #tweet
@SingularThinker

💥Don't miss this.

Have you ever thought about what exactly is the meaning of matrix?
A box of numbers? Multiple column/row vectors? A way of representing linear transformation? and maybe all/none is true.

But Have you ever tried to visualize a matrix in your mind at most for 2*2 ones? Watch this 4 episode exciting series, SEE Matrix, on YouTube if you haven't already.

⚠️Spoiler Alert:

In the first chapter, it visualizes different types of matrices. In the second, it visualizes how one can decompose a symmetric matrix by using spectral decomposition. Ultimately, we'll see how any matrix can be decomposed into four interpretable transformations.

#math #video
@SingularThinker

LLMs attain better scores than humans in exams. Does it mean that they are more intelligent?

If you asked the above question to Francois Chollet, he would say no. But anyway, what is the meaning of intelligence? He explained his idea in this paper.

#language_modeling
@SingularThinker

Modern deep learning models often consist of many model parameters, which are more than the number of training data points and therefore over-parameterized. These over-parameterized models can be trained to overfit the training data (achieving close to 100% training accuracy). It is called benign overfitting when their predictions on unseen test data are so accurate. If they perform poorly on test data, it is known as harmful overfitting which is the classical viewpoint about overfitting.

This was part of the paper, Benign Overfitting in Two-layer Convolutional Neural Networks, which Yuan Cao presented in the One World Seminar Series on the Mathematics of Machine Learning. Complement to this the one world seminars on probability, Information, Signals, and Data (MINDS), methods for arbitrary data sources (MADS), and imaging and inverse problems (IMAGINE).

So if you are interested in any of these directions check their webpages.

#neural_networks
@SingularThinker

Did you know that you can have access to Github Copilot for free?

Github provided a GitHub Student Developer Pack which gives you access some real-world tools and education. You just need to verify you are student(university or school). Check this out.

#tools
@SingularThinker