If you want to understand DeepSeek's algorithm, really. That is, at a do-it-yourself/mechanical/I-could-work-at-DeepSeek--level, this is the video to get you there.
He creates a smaller, accessible problem and uses their algorithm. I haven't seen another explanation like it. Well done Dr Nica!
These days, I've been working on recommender systems. As far as machine learning applications go, they are some of the highest impact things you can focus on.
This work has been with the message optimization company Aampe. A good message is often one that mentions a thing people are interested in.
It's been quite interesting and a lot of fun. So much so that I think I'll do a deep dive video on recommender systems one day.
To support some client work, I've been doing a contextual bandit literature review. I've come across several good papers, but if I had to pick one to recommend, it would be..
It turns supervised learning problems (where there's tons of data) into CB problems, and compares performance of several CB algorithms.
Here are the reasons I like it:
โข The algorithms are those implemented in the overpowered, opensource software Vowpal Wabbit, used for CB and large scale online learning. So what you see in the paper is what you can use immediately.
โข It gives practical advice for getting CBs working in the real world. A lot of CB literature is focused on theory, so practical engineering advice is hard to come by. Another good paper for this is arxiv.org/abs/1606.03966
Work has me quite busy lately, so I haven't had much time for making videos. But I do have a big one planned for later this year.
In the meantime, I'm starting an email list, since it's easier to communicate what I'm interested in in written form. The articles will be like would you'd find on my website: truetheta.io/concepts/
My next video will be on reinforcement learning that works commercially.
I've collected a few good use cases, but I'm interested in more.
If you have direct experience with RL working in industry, I'd like to chat with you. If you're interested, send me an email (see the channel's About).
Just came across an excellent video byโช@MihaiNicaMathโฌ . I've seen several CLT proofs before, but this one is now my favorite. It uses moment convergence, which (to me) is a pretty light weight way to arrive at the CLT. Very well done!
So grateful you all enjoy this channel as much as I do - Thank you!!
These videos are going to get better - easier to watch, new intense topics, textbook quality coverage! Excited to make it all happen next year! You'll see!
Just wanted to say.. thanks everyone for all the ideas/feedback on my channel update video.
Iโm just so pumped thereโs such an enthusiastic and educated audience that appreciates these deep dives.
And the comments show thereโs real interest in making this weird thing work. Those ideas got me feeling ambitious to get them folded in and produce the education youโre hoping for (e.g. with balanced sound!). Itโs all a work in progress, but with enough time, I believe I can master videos that deliver mind bending concepts efficiently and powerfully.
Also, if you wrote a fat informative comment and I have responded, I will! There were a ton of comments this time. I do a handful each day.
Mutual Information
Mutual Information
Very honest, very balanced, probably correct, and refreshing amid all the AI selling.
4 months ago | [YT] | 9
View 1 reply
Mutual Information
If you want to understand DeepSeek's algorithm, really. That is, at a do-it-yourself/mechanical/I-could-work-at-DeepSeek--level, this is the video to get you there.
He creates a smaller, accessible problem and uses their algorithm. I haven't seen another explanation like it. Well done Dr Nica!
10 months ago | [YT] | 74
View 5 replies
Mutual Information
These days, I've been working on recommender systems. As far as machine learning applications go, they are some of the highest impact things you can focus on.
This work has been with the message optimization company Aampe. A good message is often one that mentions a thing people are interested in.
It's been quite interesting and a lot of fun. So much so that I think I'll do a deep dive video on recommender systems one day.
In the meantime, we had a public discussion on the topic: https://www.youtube.com/watch?v=gSEdk...
1 year ago | [YT] | 19
View 2 replies
Mutual Information
To support some client work, I've been doing a contextual bandit literature review. I've come across several good papers, but if I had to pick one to recommend, it would be..
๐๐ก๐ ๐๐จ๐ง๐ญ๐๐ฑ๐ญ๐ฎ๐๐ฅ ๐๐๐ง๐๐ข๐ญ ๐๐๐ค๐-๐จ๐๐. arxiv.org/abs/1802.04064
It turns supervised learning problems (where there's tons of data) into CB problems, and compares performance of several CB algorithms.
Here are the reasons I like it:
โข The algorithms are those implemented in the overpowered, opensource software Vowpal Wabbit, used for CB and large scale online learning. So what you see in the paper is what you can use immediately.
โข It's a good explanation of CB and how supervised learning relates as a subroutine. In fact, it inspired a recent post of mine: truetheta.io/concepts/reinforcement-learning/conteโฆ
โข It gives practical advice for getting CBs working in the real world. A lot of CB literature is focused on theory, so practical engineering advice is hard to come by. Another good paper for this is arxiv.org/abs/1606.03966
1 year ago | [YT] | 48
View 5 replies
Mutual Information
Hey everyone,
Work has me quite busy lately, so I haven't had much time for making videos. But I do have a big one planned for later this year.
In the meantime, I'm starting an email list, since it's easier to communicate what I'm interested in in written form. The articles will be like would you'd find on my website: truetheta.io/concepts/
If you're interested, sign up here: mailchi.mp/truetheta/true-theta-email-list
1 year ago | [YT] | 62
View 0 replies
Mutual Information
My next video will be on reinforcement learning that works commercially.
I've collected a few good use cases, but I'm interested in more.
If you have direct experience with RL working in industry, I'd like to chat with you. If you're interested, send me an email (see the channel's About).
Thanks everyone!
1 year ago (edited) | [YT] | 91
View 9 replies
Mutual Information
Just came across an excellent video byโช@MihaiNicaMathโฌ . I've seen several CLT proofs before, but this one is now my favorite. It uses moment convergence, which (to me) is a pretty light weight way to arrive at the CLT. Very well done!
2 years ago | [YT] | 29
View 1 reply
Mutual Information
10,000! Holy shit!
So grateful you all enjoy this channel as much as I do - Thank you!!
These videos are going to get better - easier to watch, new intense topics, textbook quality coverage! Excited to make it all happen next year! You'll see!
3 years ago | [YT] | 82
View 8 replies
Mutual Information
Just wanted to say.. thanks everyone for all the ideas/feedback on my channel update video.
Iโm just so pumped thereโs such an enthusiastic and educated audience that appreciates these deep dives.
And the comments show thereโs real interest in making this weird thing work. Those ideas got me feeling ambitious to get them folded in and produce the education youโre hoping for (e.g. with balanced sound!). Itโs all a work in progress, but with enough time, I believe I can master videos that deliver mind bending concepts efficiently and powerfully.
Also, if you wrote a fat informative comment and I have responded, I will! There were a ton of comments this time. I do a handful each day.
3 years ago | [YT] | 58
View 7 replies
Mutual Information
One of my favorite panels (before I knew to chill out with so many words on screen):
3 years ago | [YT] | 29
View 9 replies
Load more