Convolutional Neural Networks for Visual Recognition
this course will be taught again in Spring 2017!14 days ago
There is a graduate level probabilistic graphical model course on coursera. Starts in a couple weeks. I plan on taking that. I believe Stanford has a recorded graduate course on convolutional neural nets. And, I think there is also a high quality reinforcement learning course out there. Caltech Learning from Data course on Edx is also excellent, but is just wrapping up if you missed it. Not sure when it will be offered again, but video are online if you are motivated.
13 days ago
I'd strongly recommend the Caltech Learning from Data course. The whole course is posted online on the Caltech website and you can do it on your own time via that resource. Or you could wait till the next run on edX, but that may be a while from now.
I found MIT's Analytics Edge to be a great intro to ML concepts, and the internal Kaggle competition that is part of the course is tons of fun and a great way to push your learning.
The Andrew Ng ML course on Coursera is also a great introduction, though I felt it did too much hand holding on the assignments.
Beyond that, just pick up a book and start reading. Tibshirani and Hastie's An Introduction to Statistical Learning is one great text that is freely available as a PDF on their website. There are of course other texts like Bishop's Pattern Recognition and Machine Learning (PRML) .
If you are interested in Neural networks/Deep Learning, you want to look into Andrej Karpathy's course at Stanford. It isn't a MOOC but you can find Youtube playlists of all lectures and find all HW and assignments on the website. I would also highly recommend Yoshua Bengio / Ian Goodfellow's new text on Deep Learning that you can read for free online:
12 days ago
To the recommendations already posted, I'd like to add that the Tibshirani and Hastie's An Introduction to Statistical Learning is also available as a self paced course on lagunita.stanford.
Also, besides Daphne Koller's 3-course specialization on Probabilistic Graphical Models, on Coursera you can also find Geoffrey Hinton's Neural Networks for Machine Learning.
They're all in my to-do list, along with the already mentioned Yaser S. Abu-Mostafa's Learning from Data. I just need to find the time :)
-
And Geoffrey Hinton's Neural Networks for Machine Learning starts tomorrow...
posted 12 days ago by Mark_B2 Community TA
-
-
@Mark_B2 (or anyone else) - Have you taken the Geoffrey Hinton NN course? If so, what did you think? I'm hoping it is fairly advanced and deep versus more of a survey.
posted 11 days ago by patcoady
-
-
@patcoady - I haven't taken it but it is fairly in-depth though from what I've heard, the content can be a bit dry at times.
posted 11 days ago by HamsterHuey
-
-
I work with neural networks for a living and am learning things from the Hinton course. There's not a ton of math in it but it's very deep in terms of developing intuition. I've already learned some new ways of thinking about the techniques I am familiar with.
posted 11 days ago by dfannius
-
-
@patcoady I've learn 3 weeks so far and enjoy it. I second the impression of @dfannius "There's not a ton of math in it but it's very deep in terms of developing intuition".
posted 10 days ago by Mark_B2 Community TA
-
-
12 days ago
On a lighter note, if you are interested in not so much theory but practical usage, there are courses on edX provided by, OMG! microsoft that goes over use of scikitlearn package in python. There are also many Udacity free courses minus degree you can take which uses Python. I am assuming either Python or R is your choice of ML platform. Granted Maltlab and Mathematica are making strides in this area but I for one am sticking with open-source at the moment. Due to time, I have been slacking off on getting my hands on some of the latest ML packages but you can get access to GPU based ML packages, such as theano and tensorflow, for both Python and R pretty easily if you are willing to go through the installation hassles as well. Xgboost is also viable option if you are into competition as well. H2O is also a decent platform to do ML on. There are so many, I can't even keep up! :)
11 days ago
CMU 10708, Stanford cs231n, Coursera course "Neural Networks for Machine Learning", CMU 10701 or Stanford cs229 as a solid foundation for Machine Learning, and of course the two books "Pattern Recognition and Machine Learning" and "The Elements of Statistical Learning, 2nd Edition", I also found that the tutorial from deeplearning.net is very helpful.
Actually this is my plan for the coming year. I have only taken the famous Coursera Machine Learning course. But now seems to me that that course is only for amateurs. I can use sklearn fluently, but I feel like that I have to go through some serious materials seriously to be a real practitioner.
5 days ago
I'd like to add a couple of self-study options in a little different direction. Many of the online courses cover details of algorithms that may or may not be useful or even things you remember. These two sources target some fundamental first-principle concepts that should stick in your memory and be useful in the future:
- Real analysis: Principles of Mathematical Analysis by Rudin (book+exercises)
- Statistical learning theory: MIT 9.520/6.860 (videos and psets online, fundamentals in probability and analysis will be helpful)
Also, consider joining this Facebook group to keep in touch and share problems and projects:
-
I can't agree more. Algorithms are like coins scattered all over the floor. In general every algorithm is one of its own kind. Learning algorithms is like picking up coins from the floor. The more you pick the more will slipped through you fingers.
Another way I can think of is plunging yourself into Kaggle.
Taking dozens of courses doesn't work. It doesn't work for me.
posted 4 days ago by Xiaohong-Deng
-
-
Hi @BradLee77, where did you find the psets for 9.520? I can't find it on Fall 2016 site
posted 4 days ago by chalupa
-
4 days ago
Machine Learning specialization by University of Washington on Coursera is pretty good as long as you don't mind the proprietary library they use. from 2nd course onwards they teach the algorithms relatively from scratch.
there's also a new specialization called Applied Data Science with Python but it seems to be practical (so far), not focused on algorithms.
4 days ago
Udacity's. Intro to AI is a good follow up course, covers a HUGE range of probability, graphical models, machine learning, and more. AI for Robotics seems like a nice follow up as well.
I have a few thoughts on this.
1.) If you haven't taken 6.041x, just do it. It is offered once a year -- beginning in Jan / Feb.
2.) You really need to have Linear Algebra down cold.
3.) A few people have mentioned Learning from Data. If you know linear algebra, and a bit of multi-variable calculus, then I strongly concur.
One of the things I enjoyed most about 6.008.1x was the introduction to and usage of Entropy / Information-Theoretic metrics. I'll highlight that the professor who does Learning from Data also teaches a course called Information and Complexity at Caltech. It is too much of a (herculean) time commitment to do a MOOC on this, but he said he will be soon posting his information theory course notes on his webpage, most likely in January. (So be sure to visit http://work.caltech.edu/ next year).
I look forward to reading them and adding to my information theory understanding. Note: that info-theoretic approaches come up in many areas of Machine Learning, perhaps most famously -- we've used them in PGMs and they are also used as a greedy heuristic for forming decision trees.
-
Agree! Probability theory at the level of 6.041x and linear algebra seems essential. I looked at some of the resources mentioned here, and without probability theory, multi-variable calculus and linear algebra, I don't think one can "effectively" learn these advanced topics.
posted 3 days ago by phn
-
3 days ago
Not sure if it was mentioned before, but in Januari there starts a course on AI here on edX by Columbia University. It's part of 4 course cycle. Could potentially be interesting.
-
I'm not. You can enroll in these courses as an auditor, as is the case for other edX courses, and "have complete access to all the course material, activities, tests, and forums." You're only missing out on the certificate.
posted 3 days ago by mrBB
-
-
-
Here's the material: http://ai.berkeley.edu/home.html
Should also point to http://ai.berkeley.edu/more_courses_other_schools.html list of other AI material huge listposted 3 days ago by Ryan_mS
-
-
Image Manipulation and Computational Photography
Computer Science Division
University of California Berkeley
CS 280 Spring 2016: Computer Vision
https://inst.eecs.berkeley.edu/~cs280/sp16/Combinatorial Algorithms and Data Structures
https://sites.google.com/site/ucbsaas/
https://people.eecs.berkeley.edu/~brecht/
扫一扫
1282
到【灌水乐园】发言
Neural Networks and Deep Learning
posted 10 days ago by Elaktrona
@Elaktrona - Thanks, yes, I do have that in my list of resources but forgot to include it. My mind was blown when I realized the Nielsen who wrote this online book on Neural Networks and Deep Learning is the same Nielsen from the famous "Nielsen and Chuang" book on Quantum Mechanics that I studied through during my undergraduate thesis on the same topic. It's humbling and inspiring to see some people achieve so much in so many different fields!
posted 10 days ago by HamsterHuey