amitoz_azad

I am not in academia - but the meetings you have described reminded me of the meetings I had during my PhD years. I was supposed to attend 1 or 2 meetings every year in different cities in France (it was joint work between different teams from different french labs). Your first realization (where nothing gets solved, no one to make final decision) was also my experience while attending those meetings.

Reading your post - it seems to me that you are not happy with kind of collegues/peers you are working with ("someone has to cover for someone else, people with larger ego can get other to do part of their job").

Whether you stay in academia or go to industry - you are likely to encounter the same problem depending upon the quality of peers around you.

If you are passionate about research, try moving towards top-tier european universities/labs (ecole polytechnique, tum). I feel that the research is more directed and you will have better peers.

This year I published at IJCAI-2022 (acceptance rate 15%). My article was selected for a long oral presentation (170/4535). I am the sole author of my article. (https://www.ijcai.org/proceedings/2022/0387.pdf)

The information in the blog could be used by anyone, but I have written this piece particularly for the current and the prospective Ph.D. students who are aiming for their first breakthrough.

Inpainting is a process of image restoration, the idea is to fill the damage, deteriorated or missing part of an image to present a complete image. It can also be used to remove unwanted parts in an image.

Deep learning based approaches use GANs to inpaint. This requires significant amount of training. The proposed tool quickly inpaints by solving a PDE on graphs.

This is amazing! As a student I am always looking for tools to make good presentations fast. Wonderful !

However, I am wondering is there any way to add equations through latex of mathjx? Write now I feel that only way to include equations is via images or hand draw them.

Hey guys :) - I made my first web app - Basically you choose your practicing language - Then you are connected with a random learner practicing the same language and then you both discuss a question in your practicing language for 2 minutes. - The whole process looks like this: https://ibb.co/3cNzgNM

- This project has application in pointcloud denoising. - It implements chambolle-pock primal-dual algorithm. - The use of torch_geometric allows faster processing on gpu.

-Eikonal equation for segmentation. -PyOpencl backend. -It has application on graph data-structure. -Like images, pointclouds, text, audio etc.

The word Yoga comes from the root "yog" in Sanskrit. It literally means "sum", "union". Yoga is to be in union with nature. It happens when one is living harmoniously.

I second this !

I had no money to buy Windows OS.

I would say "Yes!", based on my experience they do work.

This post is so much full of wisdom.

That's dark

A whole man is a holy man !

So it all boils down to $$$? Any other possible reason?

Why do you think standford didn't make video lectures of this course "machine learning on graphs" public but decided to make the just the slides and assignments public?

My PyTorch hackathon submission. Basically I am showing that how PDEs on graphs can be solved with Message Passing class of torch_geometric and created a tool torch_pdegraph to facilitate this. I work in the field of PDEs on graphs and I realized that Message passagin equation (the way of forward convolution) on graphs (https://arxiv.org/pdf/1704.01212.pdf) can also be used to solve some PDEs on graphs.

"a lot of people that criticize debuggers..." But what really making me think is that big giants like Torvalds, Rob Pike, Guido van Rossum. Rober C Martin, John Graham they don't like using debuggers (https://lemire.me/blog/2016/06/21/i-do-not-use-a-debugger).

Yes, the jupyter notebook thing, rightly pointed out by you, it does give a quick feedback. One can also start writing code from the beginning in it and push a working cell into a .py file. This way one is very sure from the beginning that how the code will work. The downside of it, which I think, it feels to me that code is written in a very linear fashion.

Coming to debugger in python (pdb), in your blog you criticize using debugger to step line by line. But this need not be the case, one can jump to debugger prompt and set a break point from the prompt itself where one feels something is wrong and continue to that break point and repeat this process. So one need not to jump through each line. In such a case it works equivalently to carefully put print statements, i.e judiciously put print statements = judiciously put breakpoints, what do you think?

I tend to use ipython kernel for debugging in python, it is not a debugger but does the job for me to find what is wrong while working on a scientific project. What are your thoughts on it?