5 ms·
A Guide to Deep Learning
- hnarayanan 10y agoThis looks really good, and an interesting wag to describe the landscape. Is it just my phone or are the links completely broken on phones?
- roye 10y agoI saw stars indicate difficulty; do colors also have some meaning?
- bhaumik 10y agoThey contain resources from the same source. e.g. all the "green" content comes from http://www.deeplearningbook.org/ http://www.deeplearningbook.org/.
- miguelrochefort 10y agoI'll bite. We see these being posted every week. Why?
- AndrewKemendo 10y agoNo one has yet won the "These are the best resources for learning ML" aggregator and a lot of people are interested in and building them. So people keep posting them.
- hrant25 10y agoMany people (especially students from universities) keep asking me and my coworkers at YerevaNN about good educational resources. They have different levels of math background, some want to study theory and watch visualizations, others want to play with the code before reading formulas.. At some point we understood it's better to spend some time and build a guide that will cover most of these questions (and, as always, we spent a lot more time on this than we expected)
- eanzenberg 10y agoI think OP is asking, why spend time writing another guide when there's plenty that suffice, including full books and papers on the topic?
- hrant25 10y agoI was looking for something that would answer common questions for some time, but couldn't find anything. Pointing students to research papers unfortunately doesn't work in most of the cases. Deep learning book is too hard for some and they want to see a comparison of different sources.
- blueyes 10y agothis is by far the most visually attractive list of deep learning resources i've seen. and they hit a lot of the main points.
- kowdermeister 10y agoIf links are broken for you, then turn of your adblocker, because he is measuring clicks with Google analytics and he's JS is broken thanks to the missing ga function. #issue reported
- bamura 10y agoGood consolidation @adamnemecek...
- nsxwolf 10y agoEveryone's into deep learning, but what would I actually do with it? With some other field, like computer graphics, one can fairly quickly get a 3D cube spinning on their screen and know it has some relation to the special effects in the Star Wars movie they just saw. No one makes it obvious what the hobbyist can expect to do with deep learning or how it relates to the broader world.
- malux85 10y agoEveryones into Graphics programming, but what would I actually do with it? The uses are myriad, you can build Star Wars Movies, you can build computer games. You can do CAD and 3D modelling and printing, you can visualize fluid simulations But each is an entire field. The same with Deep Learning - You can do image classification for medical diagnosis. Self Driving Cars. Realtime Translation. OCR. You can model chemical reactions by doing latent space exploration. You can model gene-gene interactions. You can build a image recogniser to tell if your cat is on the couch. You can build an AI that plays GO, or any other system or game where the rewards are time-delayed and sparse. You can make it play Atari games. You can stitch satellite images together. You can reconstruct parts of photos that are missing. You can colour black and white images. You can de-noise wind sounds from microphones. You can search for comets. You can monitor deforestation. You can count cars in car parks. You can search vast ocean areas for survivors. You can build security drones that fly around at night and look for anomalies. You can turn a webpage of unstructured text into structured query-able forms (see named entity recognition). You can create visual art (google: neural style transfer) These are just a small drop of some of the cool things that are going on - and while it might be difficult for you to advance state of the art without a GPU cluster (just as it would be difficult to advance state of the art in Graphics programming) the hobbiest can certainly start in any of these fields, just pick one that inspires you.] EDIT: I run a Deep Learning startup - if you want some pointers or help getting started, or would like advice don't hesitate to email me - it's in my profile :)
- gopher_protocol 10y agoOne of my partner's tasks as a legal assistant is to go through mountains of OCRed PDFs and classify them and extract pieces of data so that lawyers and paralegals can go through them more easily. Do you imagine deep learning would be an appropriate means of automating that, or is it overkill?
- AndrewKemendo 10y agoThe primary thing missing from all of these guides is that you need to have two things for ML: 1. A purpose for utilizing it 2. A data set to train/act on Without that, all you get are a bunch of shovels and picks, but no idea of what kind of wood/bricks you need or a plan for the house.
- estsauver 10y agoI will mention that you can get surprisingly far with just a little bit of labeled training data now. Transfer learning is a pretty powerful technique.
- 1024core 10y agoYou still need to train a model on what you're transferring from. It's not like you can manufacture trained models out of thin air.
- AlexCoventry 10y agoI think the idea is that, for example, you start with some pre-trained visual object-identification network, remove the final logistic-regression layer, and replace with a layer which you train your problem on.
- AndrewKemendo 10y agoUndoubtedly. However there are three sets of data you need to a functional system: Training Data and Validation data sets (a lot of sources exist already) and the data set you want to get insights from (Novel data sets usually). These of course work together in reinforcement scenarios. So for example if you want to build a tool that tells you what kind of objects are in any given image you can use ImageNet to Train and Validate your CNN. At that point you have a classifier. But for an actual useful application you need a user to submit new images to actually give some kind of valuable output. It's that second data set that is super hard to get - because it's novel.
- hrant25 10y ago
- md2be 10y agoI didn't see k-means nearest neighbor in the list. Also, wouldn't a mathematical statistics be a prerequisite?
- Eridrus 10y agoAs far as I can tell, Deep learning in practice is pretty far ahead of the theory, so people are largely following their intuition around rather than being guided by theory.