-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy path90155533_thread.json
57 lines (57 loc) · 18 KB
/
90155533_thread.json
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
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
{
"id":"90155533",
"name":"Shlomo Engelson Argamon",
"screen_name":"ShlomoArgamon",
"tweets":[
{"id":"1206614627720794112","timestamp":"1576514319","retweet_count":"6","favorite_count":"47","in_reply_to_status_id":"null","in_reply_to_user_id":"null","in_reply_to_screen_name":"null","text":"Taking @vgr’s challenge:\n\n1 like = 1 opinion (actually, fact :) on #MachineLearning and the nature of knowledge."},
{"id":"1206614951466541062","timestamp":"1576514396","retweet_count":"0","favorite_count":"0","in_reply_to_status_id":"1206614627720794112","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"@vgr (not to mention the hype machine...)"},
{"id":"1206616447692222465","timestamp":"1576514753","retweet_count":"2","favorite_count":"9","in_reply_to_status_id":"1206614627720794112","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"@vgr All machine learning algorithms are biased, the only question is do we know what the biases are and do we care?"},
{"id":"1206616449399308289","timestamp":"1576514754","retweet_count":"3","favorite_count":"13","in_reply_to_status_id":"1206616447692222465","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"All statistical machine learning algorithms are essentially devices for interpolating between data points already seen. They can’t generalize to novel situations."},
{"id":"1206616640995155968","timestamp":"1576514799","retweet_count":"4","favorite_count":"11","in_reply_to_status_id":"1206616449399308289","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"85% of the progress in machine learning over the past 15 years is due to increases in hardware performance and availability."},
{"id":"1206616791839055874","timestamp":"1576514835","retweet_count":"2","favorite_count":"8","in_reply_to_status_id":"1206616640995155968","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"The true science of machine learning is the science of figuring out which biases are needed to learn for which tasks."},
{"id":"1206617062988222464","timestamp":"1576514900","retweet_count":"1","favorite_count":"10","in_reply_to_status_id":"1206616791839055874","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"Nobody knows how to build systems with common sense. We don’t even have the equivalent of our chemical theory about this"},
{"id":"1206617063864836101","timestamp":"1576514900","retweet_count":"2","favorite_count":"7","in_reply_to_status_id":"1206617062988222464","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"Nobody knows how to build systems with common sense. We don’t even have the equivalent of alchemical theory about this yet."},
{"id":"1206617887169032192","timestamp":"1576515096","retweet_count":"1","favorite_count":"7","in_reply_to_status_id":"1206617063864836101","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"Knowledge = justified true belief. \n\nBut justification requires communication, truth can be a matter of degree, and belief is a matter of the causal structure of an agent, not sentences in its head."},
{"id":"1206619310971600901","timestamp":"1576515436","retweet_count":"3","favorite_count":"4","in_reply_to_status_id":"1206617887169032192","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"#MachineLearning on one foot: Use an appropriate representation and bias to update your priors from the data. All else is commentary. Now go and learn."},
{"id":"1206619807484915712","timestamp":"1576515554","retweet_count":"1","favorite_count":"1","in_reply_to_status_id":"1206619310971600901","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"https://t.co/Xnp1uxwyUH"},
{"id":"1206620106685648896","timestamp":"1576515626","retweet_count":"1","favorite_count":"6","in_reply_to_status_id":"1206619807484915712","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"12. All good theoretical frameworks for machine learning are equivalent to Bayes, or approximately so."},
{"id":"1206620439390433281","timestamp":"1576515705","retweet_count":"2","favorite_count":"3","in_reply_to_status_id":"1206620106685648896","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"13. Connectionist methods will need to either incorporate or simulate symbolic processing to get beyond perception/response tasks."},
{"id":"1206628836265398279","timestamp":"1576517707","retweet_count":"0","favorite_count":"8","in_reply_to_status_id":"1206620439390433281","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"14. We will need fundamentally new conceptual breakthroughs to get beyond current \"interpolative\" #ArtificialIntelligence. These will be very different from anything currently conceived of, and will not achieve #SOTA on any known tasks for some time."},
{"id":"1206628999927156736","timestamp":"1576517746","retweet_count":"1","favorite_count":"5","in_reply_to_status_id":"1206628836265398279","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"15. SOTA- and publication-chasing are bad for science and bad for scientists."},
{"id":"1206629406657126401","timestamp":"1576517843","retweet_count":"0","favorite_count":"2","in_reply_to_status_id":"1206628999927156736","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"16. Knowledge (such as it is) inheres in the system as a whole, not any particular representations or algorithms."},
{"id":"1206629500974518273","timestamp":"1576517865","retweet_count":"0","favorite_count":"3","in_reply_to_status_id":"1206629406657126401","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"17. A \"brain in a vat\" knows nothing, as it is causally connected to nothing."},
{"id":"1206665055519084547","timestamp":"1576526342","retweet_count":"0","favorite_count":"4","in_reply_to_status_id":"1206629500974518273","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"18. Imagine what we could accomplish using GOFAI with the knowledge-base equivalent of the computing power we can now devote to backpropagation learning! Incredible!"},
{"id":"1206682616306585610","timestamp":"1576530529","retweet_count":"2","favorite_count":"7","in_reply_to_status_id":"1206665055519084547","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"19.\n#Statistics is about models.\n#DataMining is about patterns.\n#MachineLearning is about prediction.\n#DataScience is about the data.\n\nCan you figure out what the elephant really is?"},
{"id":"1206683278138318849","timestamp":"1576530687","retweet_count":"1","favorite_count":"7","in_reply_to_status_id":"1206682616306585610","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"20. Understanding is a triad:\n\n Interpretation\n /\\\n /___\\\n Action Explanation"},
{"id":"1206683776249745413","timestamp":"1576530806","retweet_count":"6","favorite_count":"8","in_reply_to_status_id":"1206683278138318849","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"21. Replace the phrases \"artificial intelligence\" and \"machine learning\" in any news article by the phrase \"computer program\", and remove the phrase \"learns like a person/baby/brain\", and see if the achievement seems as cool or impressive."},
{"id":"1206683905283231744","timestamp":"1576530836","retweet_count":"0","favorite_count":"4","in_reply_to_status_id":"1206683776249745413","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"22. Then read the original research article to see what was actually accomplished."},
{"id":"1206688473224224770","timestamp":"1576531925","retweet_count":"1","favorite_count":"4","in_reply_to_status_id":"1206683905283231744","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"23. Language learning is not just machine learning applied to sequences. Nor is automated genomics. Nor is time series analysis. Ad infinitem."},
{"id":"1206688744721592329","timestamp":"1576531990","retweet_count":"0","favorite_count":"6","in_reply_to_status_id":"1206688473224224770","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"24. You can get 80-plus percent of the possible accuracy by applying advanced machine learning to a problem without knowing anything about the domain. You can also do that using logistic regression."},
{"id":"1206689081431875586","timestamp":"1576532070","retweet_count":"1","favorite_count":"9","in_reply_to_status_id":"1206688744721592329","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"25. Does deep learning solve problem X? First, compare it to logistic regression and naïve Bayes. Then you might get a clue.\n\n(If the authors didn’t, be suspicious, very suspicious.)"},
{"id":"1206699398970925060","timestamp":"1576534530","retweet_count":"3","favorite_count":"7","in_reply_to_status_id":"1206689081431875586","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"26. As a field, #MachineLearning is stuck in a local minimum. Not all learning is function approximation, and I warrant most of the really interesting kinds of learning are not. Let’s go back to exploring the full space of learning tasks and methods."},
{"id":"1206701860947709953","timestamp":"1576535117","retweet_count":"0","favorite_count":"3","in_reply_to_status_id":"1206699398970925060","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"27. #MachineLearning is the science of finding the right bias for the problem. So… KNOW YOUR DOMAIN!"},
{"id":"1206715003488940032","timestamp":"1576538251","retweet_count":"0","favorite_count":"5","in_reply_to_status_id":"1206701860947709953","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"28. If your fancy #MachineLearning system gets 99.5% accuracy (wow!!), either:\n\na. You have a bug in your evaluation procedure, or\n\nb. Logistic regression will get 99.3% accuracy."},
{"id":"1206715878726914048","timestamp":"1576538459","retweet_count":"0","favorite_count":"7","in_reply_to_status_id":"1206715003488940032","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"29. #DataScience is sensemaking. \n\nThus, if you create a black box model with perfect prediction accuracy, you have FAILED."},
{"id":"1206716581553868800","timestamp":"1576538627","retweet_count":"0","favorite_count":"6","in_reply_to_status_id":"1206715878726914048","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"30. For any #MachineLearning or #ArtificialIntelligence application, the “system” includes the people that use it, and the organizational matrix they are embedded in. Any analysis that does not account for that is woefully inadequate."},
{"id":"1206941772842184709","timestamp":"1576592317","retweet_count":"0","favorite_count":"5","in_reply_to_status_id":"1206716581553868800","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"31. \"Neural networks\" are not brain-like. Unless you take your glasses off and squint. After a couple of beers."},
{"id":"1206942779760422916","timestamp":"1576592557","retweet_count":"0","favorite_count":"4","in_reply_to_status_id":"1206941772842184709","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"32. The problems of \"bias\" and \"fairness\" in #MachineLearning are mainly problems of specifying implicit assumptions, and have no purely technical solutions. The real issue is a version of the old \"is/ought\" conundrum."},
{"id":"1206960913762443266","timestamp":"1576596880","retweet_count":"0","favorite_count":"4","in_reply_to_status_id":"1206942779760422916","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"33. Most “impressive” accomplishments of #MachineLearning for #NaturalLanguageProcessing are advanced applications of the Eliza Effect. (I’m looking at you, GPT-2...)"},
{"id":"1206961117815332865","timestamp":"1576596929","retweet_count":"1","favorite_count":"6","in_reply_to_status_id":"1206960913762443266","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"34. Don’t be afraid of the algorithms getting too smart, be afraid of people giving them too much power before they do."},
{"id":"1206976379629248513","timestamp":"1576600568","retweet_count":"0","favorite_count":"4","in_reply_to_status_id":"1206961117815332865","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"35. You know something if you can do something with it. That might involve action or communication, but beware: It is very easy to convince insufficiently critical observers that you know something, even inadvertently. Even yourself."},
{"id":"1206976481152421889","timestamp":"1576600592","retweet_count":"2","favorite_count":"5","in_reply_to_status_id":"1206976379629248513","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"The first principle is that you must not fool yourself – and you are the easiest person to fool. \n\n--Richard Feynman"},
{"id":"1206978355310014465","timestamp":"1576601039","retweet_count":"1","favorite_count":"0","in_reply_to_status_id":"1206976379629248513","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"36. Your #MachineLearning algorithm works - congratulations! Excellent accuracy on out-of-sample data - wonderful!\n\nBut do you know if it learned what you wanted it to learn? Does it recognize stop signs, or large-enough-red-regions-with-certain-specific-other-colors-nearby? Hm."},
{"id":"1206987692791255040","timestamp":"1576603265","retweet_count":"0","favorite_count":"0","in_reply_to_status_id":"1206978355310014465","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"37. What statistical assumptions does your method make? More importantly, what assumptions does your evaluation procedure make? And do they match reality? (Answer: No.)\n\nThe proof of the model is in the eating."},
{"id":"1206988835894235137","timestamp":"1576603537","retweet_count":"0","favorite_count":"1","in_reply_to_status_id":"1206987692791255040","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"38. Statistical #MachineLearning is algorithmic demagoguery - it is winner-take-all for (often subtle) patterns with a slight majority (plurality). \n\nThat is its power, and its danger."},
{"id":"1206991401977860096","timestamp":"1576604149","retweet_count":"0","favorite_count":"1","in_reply_to_status_id":"1206988835894235137","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"39. As a general rule on #MachineLearning, representations matter more than algorithms."},
{"id":"1207098574108004353","timestamp":"1576629701","retweet_count":"0","favorite_count":"2","in_reply_to_status_id":"1206991401977860096","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"40. The proper function to optimize in #MachineLearning is task dependent utility, even though this is nearly never done."},
{"id":"1207098908859650051","timestamp":"1576629781","retweet_count":"0","favorite_count":"0","in_reply_to_status_id":"1207098574108004353","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"41. The results of a single study, however rigorous, never generalize on their own. Validity depends upon statistical assumptions, and until you replicate, you don’t know whether those assumptions match reality."},
{"id":"1207099330039099396","timestamp":"1576629881","retweet_count":"0","favorite_count":"0","in_reply_to_status_id":"1207098908859650051","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"42. Forty-two."},
{"id":"1207107089774059522","timestamp":"1576631731","retweet_count":"0","favorite_count":"0","in_reply_to_status_id":"1207099330039099396","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"43.\n #ArtificialIntelligence !=\n #MachineLearning !=\n #DataScience !=\n #Statistics !=\n #Truth"},
{"id":"1207148488275976192","timestamp":"1576641601","retweet_count":"5","favorite_count":"3","in_reply_to_status_id":"1207107089774059522","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"44. Unless your #MachineLearning team includes both #ML and domain expertise, don't trust the results. \n\nIOW, don't just use out-of-the-box machine learning \"solutions\". They aren't."},
{"id":"1207195564611395584","timestamp":"1576652825","retweet_count":"1","favorite_count":"1","in_reply_to_status_id":"1206715878726914048","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"https://t.co/uyudE7FHvd"},
{"id":"1206699574435430409","timestamp":"1576534572","retweet_count":"0","favorite_count":"0","in_reply_to_status_id":"1206689081431875586","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"Bonus. \n\nSTOP CHASING SOTA!\n\n</rant>"},
{"id":"1206715275023978497","timestamp":"1576538315","retweet_count":"0","favorite_count":"0","in_reply_to_status_id":"1206683776249745413","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"https://t.co/wLWP8szulJ"},
{"id":"1206619408401129475","timestamp":"1576515459","retweet_count":"0","favorite_count":"0","in_reply_to_status_id":"1206617887169032192","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"Yes, Bayesianism is correct."},
{"id":"1206619566425726976","timestamp":"1576515497","retweet_count":"0","favorite_count":"1","in_reply_to_status_id":"1206619408401129475","in_reply_to_user_id":"90155533","in_reply_to_screen_name":"ShlomoArgamon","text":"So is minimum description length."}
]
}