cosmosa

In Weaknesses of RNN architecture I’ll motivate the need for the transformer architecture. In Transformer architecture I’ll go more in depth into its architecture. Finally, in Experiments and results I’ll show how the transformer performance compares to the RNN performance, and as a sub-research question, how the original encoder-decoder transformer architecture compares to GPT-style decoder-only architecture in Encoder-decoder vs decoder-only

The online degree earned is the same as one in person. However I found the lectures to be much more "staged" online than the free-flowing in person lectures and discussions. So I would argue that the online lectures are of lower quality.

I would be grateful you have a job; most software engineering jobs pay relatively well. Be grateful for this pay.

Find something outside of work that excites you if work doesn't excite you. Work isn't everything. If the actual thing you do at work doesn't excite you, try making friends at work who do excite you.

If writing simple CRUD apps is boring, it sounds like you actually have some aptitude afterall. If it's boring, try to find the sweet spot of challenge with capability. It sounds like you are capable of more than what you are doing now, so gravitate in that direction and move from there.

pycharm -- 'nuff said

Good article. Shows the risks/difficulty in applying machine learning models in the wild. Also shows the more sober side of applying ML models, where a domain expert says "Yeah, I know, I didn't need an ML model to tell me that". ML models will make mistakes based on bias, but hopefully in the long run they will outperform humans.

It's memorized in the same way that natural language syntax is memorized. It's just learned.

This is excellent. I had been wanting to see something like this for a long time. It was difficult for me to imagine the orbit of earth around the sun, and nothing I found showed it well. Thanks for making this!

It is not really optional, it is always printed I think, and just thrown out by the cashier if you say you don't want it. If you do self checkout in my experience it's always printed. I think maybe gas stations are the only place where it asks you if you want a receipt.

What a bunch of odd features.

what is a scissor mechanism?

Sounds interesting but unable to read the full paper

I liked their presentation of the data. Cool interactivity and scrolling transition with notes beside the plots.

I've seen many different formats used.

1. Focus on open ended questions with very little programming

2. Give very difficult algorithm questions 3. Take home assignment

4. Knowledge based questions

5. Brain teasers

By far I think the most useful is take home assignment as it is the most fun, relaxed, and realistic way to measure someone's ability to code.

I don't think algorithm questions are that realistic because usually they are unrealistically difficult.

Mixture of take home and knowledge based questions with maybe some algorithm questions of realistic difficulty is best.

Khan academy material is quite good for multivariate calculus. It's taught by the guy behind 3blue1brown. His teaching style is very intuitive.

In my opinion moocs have a long way to go to be comparable to an in person experience. First of all the execution of the pre-recorded lectures is not nearly as good as in person. Writing on the board vs seeing slides and hearing a voice vs seeing someone talk in person is very different. The pacing of the pre-recorded lectures is very mechanical as I'm sure it's tough teaching a camera, whereas in person it is pretty natural. The ability to ask questions in class really helps too. Not meeting with other students/tas/professors in person really has a bad outcome on learning.

It's still possible to try to recommend as close a match as possible. If no matches, then recommend based on actor, director, producer, etc. Currently Netflix just throws in the towel and recommends some completely unrelated content.

It may seem like the selection is terrible on the streaming service, but I think a lot of content is just hard to find.

Great!

I'm in the same boat as you. I started studying advanced machine learning/AI concepts, and understand the intuition just fine, but I feel like i'm grasping around in the dark without a good foundation in statistics. Currently reading this online book and watching lectures: https://www.stat.berkeley.edu/~stark/SticiGui/Text/toc.htm. The material is stated quite simply without advanced math, but it actually helps a lot to understand concepts better.

Looks good, I will definitely use this as a reference!

I stopped reading when it said AI is a sub-field of machine learning. It’s actually the other way around.

I am currently 60% through the GT omscs program. I’d say it’s good for your money. So far I have been unimpressed with the quality of the recorded lectures both compared to other recorded lectures and on campus lectures. Group projects are a pain. The focuses especially in ML are weak, you really have to do a lot of work on your own in order to be competent in a specialization. Because of the lack of courses they offer I’ve ended up taking 3 / 6 courses which I didn’t need to take or enjoy taking. I’d say it’s best if you want to get a well rounded overview of some modern trends in cs rather than focusing on any specialization.