AI / ML Resources
 Share
The version of the browser you are using is no longer supported. Please upgrade to a supported browser.Dismiss

 
View only
 
 
ABCDEFGHIJKLMNOPQRSTUVWXYZAAAB
1
Here's a list of useful resources with some information on each one. Obviously it's not exhaustive, and will continue to be updated as I learn more. It's really skewed towards convolutional networks / computer vision right now; that should change over time. I don't make any claims toward these works or their accuracy; I've just found these good for my personal learning! Key explaining priority colors on the right. Hope you enjoy.
2
PriorityTitleTypeCitationContributor(s)Notes
3
Neural Networks for Machine LearningCourserahttps://www.coursera.org/learn/neural-networksGeoffrey Hinton, University of Toronto / GoogleA truly awesome course taught by one of the all-time greats in ML. If you're just starting and want to quickly learn how to build your own neural networks for machine learning, I'd start here. You don't really need to watch the videos because there are pdfs of all the video slides in the "lecture resources" for each week. It's not too useful to do the Octave programming assignments either. But you should complete the quizzes and the final exam to make sure that you understand the content. It can certainly be tricky. Don't pay for the certificate - just complete the course for free.Key
4
Machine LearningCourserahttps://www.coursera.org/learn/machine-learningAndrew Ng, Coursera / Stanford / Baidu AI / Google BrainThe classic ML course, also highly recommended. I haven't got that far into it myself, but the instruction is great, and just about everyone who's wanted to learn about machine learning has encountered this resource. Read the pdfs of the lecture slides, do the quizzes. Don't do the programming assignments, don't watch the videos, and don't pay for a certificate.Super important or necessary to go over!!
5
TensorFlowAPIhttps://www.tensorflow.org/The TensorFlow team at GoogleI've really only used TensorFlow so far as my ML API. It's a good API although a bit obtuse in documentation. Don't be surprised if you end up looking through their open source code, and end up more confused than if you hadn't bothered. Still an awesome tool.Really interesting to read, but not highest priority.
6
A Beginner's Guide to Understanding Convolutional Neural NetworksBloghttps://adeshpande3.github.io/adeshpande3.github.io/A-Beginner's-Guide-To-Understanding-Convolutional-Neural-Networks/Adit Deshpande, UCLA CS '19A CS undergrad at UCLA wrote a really good explanation of convolutional neural nets, which are basically the standard tool in computer vision / image recognition.Interesting, but it's up to you if want to bother reading it.
7
The Data Science BlogBloghttps://ujjwalkarn.me/blog/Ujjwal Karn, ML at FacebookA blog on data science, including work in image recognition and NLP (natural language processing). I've looked at their info on convolutional networks.
8
What to Think about Machines that ThinkBookhttps://www.amazon.com/What-Think-About-Machines-That/dp/006242565XEdited by John Brockman, Edge.orgAlmost 200 short essays by leading scientists, researchers, philosophers, etc. on the development of intelligent machines. A great book for someone looking to understand AI / ML from a philosophical / moral / societal perspective, not just from a technical standpoint. Has some interesting opinions.
9
Steps toward Artificial IntelligencePaperhttp://worrydream.com/refs/Minsky%20-%20Steps%20Toward%20Artificial%20Intelligence.pdfMarvin Minsky, MITA paper from the 60s about the kind of heuristic problem solving we'd expect an intelligent machine to be capable of. Kind of long; haven't read it completely myself.
10
The Coming SwarmTV segmenthttp://www.cbsnews.com/news/60-minutes-autonomous-drones-set-to-revolutionize-military-technology/60 Minutes, CBSDiscusses autonomous ML systems in the military (self-controlled drones). Explained in terms a layperson can understand.
11
Artificial Intelligence, Automation, and the EconomyPaperhttps://obamawhitehouse.archives.gov/sites/whitehouse.gov/files/documents/Artificial-Intelligence-Automation-Economy.PDFThe Obama White HouseDiscusses impact of automation / ML in the economy. Particularly interesting from a policymaking perspective, as it discusses industries ripe for automation (ex. trucking) and plans for retraining displaced workers. Takes the "augmentation" position (AI will augment rather than remove our jobs), so doesn't voice much concern about AI disrupting highly-educated professions.
12
MobileNets: Efficient Convolutional Neural Networks for Mobile Vision ApplicationsPaperhttps://arxiv.org/pdf/1704.04861.pdfThe MobileNets team at GoogleMobileNets use depthwise and pointwise separable filters to make depthwise separable convolutions. Demonstrates the tremendous speed / memory improvements gained from using these rather than standard convolutional filters. Can trade off accuracy for model efficiency, with least efficient models still an order of magnitude less intensive than current state-of-the-art and accuracy nearly the same.
13
Siamese Neural Networks for One-shot Image RecognitionPaperhttps://www.cs.cmu.edu/~rsalakhu/papers/oneshot1.pdfGregory Koch - Richard Zemel - Rusian Salakhutdinov, University of TorontoTalks about using Siamese neural nets to classify similarity between images. This is useful for one-shot image recognition, which should learn from a limited dataset by finding similarity between images (especially after some distortions) and be able to infer which "label" image each other image is most similar to.
14
Matt Harvey's articles on MobileNet in TensorFlowBloghttps://hackernoon.com/@harvitronixMatt Harvey, Coastline AutomationA good two part series: Creating insanely fast image classifiers with MobileNet in TensorFlow, and Building an insanely fast image classifier on Android with MobileNets in TensorFlow. Discusses using MobileNets on a mobile device. Doesn't really dive into the code. Another interesting article: Five video classification methods implemented in Keras and TensorFlow. Talks about some interesting relatively lightweight video classification models, including a combined CNN and RNN model which performs pretty well.
15
Towards Data Science: Machine LearningBloghttps://medium.com/towards-data-science/machine-learning/homeVarious ML bloggersI've read a good series written by Rutger Ruizendaal about building an image classification convolutional net, and setting it up on AWS. Also read a nice piece by Dat Tran on building your own real-time object detector. Dat applies it to identify racoons because he thinks they're cute and I guess some racoons live near his home. Think this is a good source with lots of interesting blog posts.
16
Josh Miller's GitHubGitHub accounthttps://github.com/jmiller656Josh Miller, RIT Engineering '19Josh really knows his stuff and has done some awesome / innovative / funny projects. Code uses TensorFlow API. Great examples of standalone projects in ML.
17
Colin Parsons' GitHubGitHub accounthttps://github.com/cparsons429Colin Parsons, WashU Math '20My GitHub account. Not nearly as impressive as Josh's.
18
Enhanced Deep Residual Networks for Single Image Super-ResolutionPaperhttps://arxiv.org/pdf/1707.02921.pdfDepartment of ECE, Seoul National UniversityA convolutional network architecture for image magnification. For implementation in TensorFlow, see Josh Miller's GitHub. Haven't read it completely myself.
19
Perceptrons: An Introduction to Computational GeometryBookhttps://www.amazon.com/Perceptrons-Introduction-Computational-Geometry-Expanded/dp/0262631113Marvin Minsky - Seymour Papert, MITOriginally published in the 60s. Demonstrates the limitations of perceptrons. Haven't read it myself; have only heard the summary. This early version of neural nets could only solve linearly separable problems - unable to learn even some concepts as simple as XOR. This catalyzed development of modern neural nets with nonlinear neurons.
20
Not Hot DogYouTubehttps://www.youtube.com/watch?v=ACmydtFDTGsSilicon Valley, HBOWarning: some profanity. It's pretty funny though - classic problem of world-changing technology being used to solve a totally unimportant issue.
21
How HBO's Silicon Valley built "Not Hotdog" with mobile TensorFlow, Keras & React NativeBloghttps://hackernoon.com/how-hbos-silicon-valley-built-not-hotdog-with-mobile-tensorflow-keras-react-native-ef03260747f3Tim Anglade, HBOA good explanation of how HBO actually built the hot dog / not hot dog app. Talks about the whole process - really interesting look at running a framework like that on a mobile device, as well as training the net and generating the dataset.
22
Generative Adversarial NetsPaperhttps://arxiv.org/pdf/1406.2661v1.pdfIan Goodfellow et. al, Google BrainGANs are a particularly powerful machine learning architecture, which basically train two competing networks. One learns how to generate data similar to input data (the generator), and the other learns how to discriminate between this machine-produced data and the real dataset (the discriminator). As these two networks compete with each other, the system converges toward a solution where the generator is forced to learn an extremly good representation of the data in order to fool the discriminator.
23
Harshvardhan Gupta's articlesBloghttps://hackernoon.com/@harshsayshiHarshvardhan GuptaSome really good articles on more "exotic" kinds of networks - Siamese networks, relational networks, CANs (creative adversarial networks), etc. Interesting technical explanations of some cool architectures applied to interesting problems.
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
Loading...
 
 
 
Sheet1