I am an undergrad trying to break into AI and machine learning research. Hopefully this will be useful. Does anyone know any other resources that will be helpful? So far I've just been reading AI: A Modern Approach, by Russel and Norvig.
It's hard to recommend something if we don't really know your level of knowledge so far.
I wholeheartedly recommend the fast.ai [0] course. It provides a lot of instantly applicable code, coupled with very good explanations which you can try out on novel problems later. It's focused on "learning by doing", and not "learning by reading" which fits my style really well.
That said, it doesn't dissuade the watcher from reading later, it's just not recommended to start out with.
I think this class/site is mentioned on every relevant HN for very good reason: it's actually that good.
I dislike learning from video (upping playback speed helps), I dislike the coding style of the library and the notebooks (nonlinear notebook execution especially), and I still think this is the best available class on anything deep-learning related, and it's only getting better. The top-down, practice-before-theory approach is excellent, but they still get into the theory, often in a much more intuitive and better motivated way than you get elsewhere. Also tons of little breadcrumbs dropped throughout lessons and in the forum to dig deeper for those inclined to.
If you go this route, make sure to follow the suggestion of re-implementing each lesson, from scratch, without referring back to the original notebook. It's a little too easy to not do that and miss out on the lessons you learn from struggling through the actual code.
So far I have taken one introductory class on AI in general, but it did not cover machine learning. I took one class on machine learning, but I only grasped the basics of several algorithms from the class. These classes, and the textbook I've partially completed, are the extent of my knowledge.
Take as much probability and linear algebra as you can conveniently do – as much for the intuition as for the symbol-manipulation mechanics – and don't underrate the importance of domain expertise in any problem you get interested in!
it seems weird to count "enthusiastic, uncompensated endorsement from satisfied customers" as "marketing"; maybe it is technically marketing, but it carries none of the negative connotation your comment seems to imply
I've done Andrew Ng's Coursera specialization (deeplearning.ai) and course.fast.ai, and I would 100% recommend starting with fast.ai. (Seeing results more quickly is motivating. It's also free.) When you know that you enjoy the topic, feel free to learn more rigorous ML theory from other sources.
I would highly recommend going through François chollet's Deep Learning with Python[1] book. The technical concepts are explained very well and since you have gone through the Modern AI book you won't have an issue understanding them. It's a very hands-on book and by the time you finish it you will be able to use Deep Neural Nets to solve many problems.
I would also recommend going through the scikit-learn documentation. Some of the tutorials/examples there are pretty good.
At the end of the day, it all comes down to your personal learning style. For me the thing that worked was to go through the above mentioned steps and then find a problem I was interested in and try to solve it using my newly found skills. That way you will discover new tools and methods.
Finally, the Deep Learning [2] book is also very good but I would not recommend it to a beginner. It's better to use it when you have a basic understanding of Machine Learning and you want to gain a deeper understanding of the concepts.
I would pair [1] with Hands On Machine Learning with Scikit-Learn and Tensorflow by Aurélien Géron (I own both). It gives an excellent overview of machine learning including non-deep stuff (plus the ins and outs of scikit-learn and tensorflow).
First one is a bit older school, but takes you through all the fundamentals and actually explains a lot of the math involved. It also gets you thinking a lot more about how to solve problems from a Linear Algebra standpoint and the types of problems machine learning is good for tackling.
Second one is a much more modern day set of courses specifically focused on Deep Learning techniques and problem solving.
I thought both were great. First one is free as well...
worth noting this is reinforcement-learning specific; fascinating field, getting a lot of press the last few years, but best considered an important and distinct field that just happens to intersect with ML/DL. I'd suggest understanding DL on simpler problems (well-understood CV/NLP problems) before wading into using it in reinforcement learning.
The bulk of academic work is chasing grants, not "curiosity driven research". If universities want professors not to leave in droves to tech companies that will consistently give them funding and cut away the bullshit that eats up the majority of a professor's time, they should try competing, rather than bemoaning that professors are no longer following the sacred path of academic asceticism.
While I would like to stop chasing grants as much as the next tenure track assistant professor, I don't think it's fair to characterize "the bulk" of academic work as chasing grants.
I do not think this is what the piece is arguing. Rather, the concern is that by creating these 80/20 splits, the core values of the university are compromised. There's nothing a priori wrong with industrial research, it's this attempted hybrid that's problematic. Hence the title, "you cannot serve two masters".
Quoting the piece, "Part of the point of being a big company is to control your environment by crushing, containing, or co-opting inconvenient innovations." I think the author is arguing that attitude is fundamentally at odds with the values of the academy.
If BigCompany expects its research center to crush inconvenient innovations, they're not really running a research center, they're just calling it that because they like the titles.
i read "inconvenient" as "competing". if your r&d can lead to a few choice patents (or you "just" buy them) you can sometimes hamstring competition or get little graft by leeching off competitor's developments.
Top conferences accept papers with a large expected impact, not necessarily papers the authors were really curious about. It's deeper than merely what time is spent on, it's also where accolades are given. The facebook thing shows real problems to the researchers and provides good data, good starts to high impact papers.
No, Google is the dominate search engine which is part of an oligopoly which while still not as economically efficient as a perfectly competitive market, they have nowhere near the power of a monopoly.
If Google replaced all of their search results with ads, people would easily switch to bing as bing isn't that much worse.
Facebook on the other hand definitely is "pretty much" a monopoly (but not completely) for certain subsets of the social media market due to network effects which is why they shovel ads down your throat on their platform and they are making tons of money.
That is fair. However, I still argue that it is not nearly as strong as it is easy for me to switch from google search to bing as most of my search results won't really change as they are basic searches but it is incredibly painful to switch social media networks as you have to rebuild your individual content and network connections for the new sight.
I actually have set Bing as my default search on my phone
... and let me tell you Bing is not nearly as good as Google when it comes to bringing me the right search results first.
But I got tired of Google asking me to “prove that I’m not a robot” by tapping on roads and street signs with every new search. I use incognito mode, and since they can’t track me, they either are punishing me or just automatically assume I’m not human.
Google replaced all there search results with ads a looong time ago.
I blogged the first time I saw Google results had all ads on the visible part of the results page. Now often its all ads for the whole page. And you will not find any page of results that is not defined by Google's definition of what "organic" search should be. E.g amp, https, having an adwords account, hosted by them, tuned your page to their adwords requirements. etc etc.
Can you take a screenshot because I have never seen a page of all ads? I just did a few random searches for typical highly advertisable things and I got 1-4 ads but you could still see organic search results. Granted the "asbestos lawyer" search did have half of the original screen area full of ads but scrolling down showed most of the actual page with organic search results.
Free speech in America isn't just valued as a legal right before the government, but as a general principle. Most people think that free speech in most situations (not just before the government, and with a few restrictions) benefit everyone.
Freedom of association is also an important general principle. For instance, I don't think many place any value whatsoever on the freedom of other people to speak in their own home. I certainly wouldn't let someone stay in my house while saying the kinds of things I regularly see on Facebook, and that's my right. I think this is quite an important property of our system, and I think most agree.
As a private commercial enterprise, Facebook itself is, for the people who own it, much like my home; they are free to set whatever policy they like regarding what goes on there, as long as they aren't discriminating against certain specifically protected classes of people. It's shortly tricky, it's not unreasonable to argue that a platform the size of Facebook should be treated more like a public space than a private one. But I personally think the bar for overriding freedom of association should be extremely high, and I don't think Facebook is over it. Those who have been disallowed to associate with Facebook remain free to associate with other very similar platforms.
> In 2015, researchers evaluating how Facebook Zero shapes information and communication technology use in the developing world found that 11% of Indonesians who said they used Facebook also said they did not use the Internet. 65% of Nigerians, 61% of Indonesians, and 58% of Indians agree with the statement that "Facebook is the Internet".
Yeah I think that's awful that they do that. I was specifically talking about in the US, where it is easy to use a competitor. (The comment I replied to was also talking specifically about the US: "Free speech in America...")
The best subreddits (askhistorians, etc.) will instantly delete low quality posts.
Reddit as a whole banning bad subreddits (fatpeoplehate, coontown) made reddit a better place for everyone, and the trash moved to voat which is barely active now.
Most big sites will now ban you for for threatening or very hateful speech.
Free speech is not the issue. No one is taking away free speech by banning you on Twitter.
Your free speech is taken away when something you say lands you in prison. Getting banned from a shitty private website doesn't violate your free speech rights in any way, full stop.
It's a principle most people don't understand. You have the right to say what you want, in public, and not be put in prison. You don't have the right to an audience or to accessing private infrastructure. There's a whole public Internet that you can do whatever you want on. Say whatever you want. Build your own Twitter, spout whatever nonsense you want. No one will censor you!
