XOutlier
Download .zipAdd to Chrome

Yi Ma

@YiMaTweets128k followers

Chair Professor of AI, Hong Kong University Pursuing a Mathematical Theory of Intelligence:

Open on X
Top outlier
54×
Best organic post
Peak views
968k
Single post
Bookmarks
25k
Across 101 organic posts
Last scan
Oct 16
Community cache
Posts 101 posts
#DatePost↓ OutlierViewsLikesRepostsBookmarks
1Nov 23, '23Based on our latest study: , compression seems to be all there is in current AI systems, including GPT-4. The remaining question is: Can compression alone lead to general intellige54×968k984158634
2Oct 10, '22For all young folks who are interested in intelligence, I highly recommend to study five subjects and understand how they are related: compressive sensing (sparse coding), informat25.2×03.8k5032.3k
3Jul 1, '23Mathematics is the art of giving the same name to different things(Henri Poincaré). Machine learning is the art of giving different names to the same thing.17.7×317k4715341
4Jul 17, '22I always tell my students: if you only read paper published in the past five years, the probability that you will have any ground-breaking idea in your lifetime is nearly zero. The14.5×03.7k426380
5Aug 22, '25Our latest book on the mathematical principles of deep learning and intelligence has been released publicly at: It also comes with a customized Chatbot that helps readers study and13.2×237k1.4k2741.7k
6Dec 16, '23Today, I am deeply shocked and saddened by the news that my long-time dear friend, professor Xiaoou Tang of CUHK, advisor to Dr. Kaiming He and my predecessor at MSRA, peacefully p12.1×216k6725133
7Sep 13, '249.11 is still larger than 9.8, despite can memorize solutions to PhD level questions. Again, memorizing is not understanding and knowledge is not intelligence.9.2×165k93585156
8Nov 10, '25Based on this book, I am developing and teaching an entirely new graduate course on Principles of Deep Learning (and Intelligence) at HKU this semester. The slides and videos for t9.2×164k684116961
9Oct 8, '25Our new open book is being actively updated, thanks to the great effort by the team. The following thread provides a nice overview of the book and its latest updates. Please always8.6×155k54285386
10Jul 15, '25I believe all professors in the field of AI and machine learning at top universities need to face a soul-searching question: What can you still teach your top (graduate) students a8.3×148k1.2k108484
11Nov 23, '23If you want to fully understand deep networks, Transformers in particular, please check out our latest paper: I hope this comprehensive study will convince you that there is absolu8.2×147k1k192994
12Jun 18, '23I am getting increasingly fascinated by both simplicity and complexity of the following equation that captures the flow of information gain. It literately summaries everything I kn7.5×134k58569492
13Sep 23, '24I saw some people didn’t believe my previous post.7.5×134k3372870
14Jun 5, '23I hope this work “White-box Transforms via Sparse Rate Reduction” will help both theoreticians and practitioners fully understand deep networks, particularly Transformers. Understa7.5×134k35975261
15May 3, '22Mine is too free for download at: but your time of reading books is not free, even more costly if got misled… So be sure you choose the right books to read.7.2×01.1k243554
16Jun 26, '24Intelligence is the ability to imcrement knowledge, hence knowledge is the integral of intelligence.7.0×125k1.2k179390
17Nov 18, '25After 30 years of research and teaching, I found that I cannot teach any subject unless I have understood it fully. Hence I normally had to write a book before I teach any new cour6.5×117k2.2k198404
18Jan 11, '23Heading to Hong Kong tonight — a new chapter in my life.6.4×115k48332
19May 9, '23The best part of research is realizing you have understood something that no one else in the world has. That feeling is absolutely surreal, and the most addictive.6.3×113k7589459
20Mar 30, '22Finally, my new book on "High-Dim Data Analysis with Low-Dim Models: Principles, Computation, and Applications" is available in stock on Amazon: for anyone who needs a rigorous mat6.1×01.3k155351
21Mar 31, '22Based on the new book, we have developed a new graduate course on data science: Berkeley EECS 208 "Computational Principles for High-dimensional Data Analysis": , with a complete s6.0×0995207433
22Jul 12, '22My personal view on Intelligence -- the most difficult jigsaw puzzle I have ever tried to understand my whole life. The paper is not an invitation for debate. Please no rushed comm5.8×01k144437
23Apr 27, '22I am developing and teaching a new course on 3D Vision at Berkeley this semester: CS294: Geometry and Learning for 3D Vision. It will be finished in about two weeks. I will make al5.5×01.4k104224
24Sep 28, '25In my previous high-dim data analysis book: , I have deliberately highlighted that the most efficient class of optimization algorithms that we ever know (including power iteration 5.3×95k70276723
25Nov 24, '23As machine learning has almost become an art of giving different names to the same thing in past decade, just want to remind everyone what Henry Poincaré said: Mathematics is the a5.1×91k82291118
26Aug 19, '22The Diffusion Process that new generative methods (DALLE) rely on is known as Laplace's method (or Langevin dynamics). You can find a systematic intro in Ch. 9 of my new book on Hi4.9×0935107366
27Sep 7, '25Amazing illustrations! In our new book, Chapter 5 is dedicated to (compressive) autoencoder. I believe most people do not know the history well. The idea of autoencoding is to ensu4.9×87k79776592
28Jul 17, '22All key ingredients of AI: Computers (Turing), Neural Networks (McCulloch), Feedback Control (Wiener), Game Theory (von Neumann), Information Theory (Shannon) were all introduced i4.8×01k111241
29Oct 21, '25From what I know, we barely scratch the surface regarding understanding intelligence… next ten years will be a golden era of rediscovering intelligence for academia. At least my st4.8×86k55330232
30Jan 29, '23Back to Hong Kong from mainland China. Finally can access google and twitter normally. Normalcy for some is luxury for others.4.8×86k3031110
31Mar 24, '25Visited a new AI startup Hyperbolic in San Francisco by a math prodigy from Berkeley, Zhang Yue…4.6×82k5821378
32Nov 15, '25Has Elon read our new book? To be precise, *current AI* technology is ALL compression and correlation -- only implementation is still far from optimal and natural. But I tend to be4.5×81k47644235
33Oct 3, '25As an open book, we welcome all people to contribute, to revise, and to critique. Meanwhile, as we are teaching this subject, we will also open source all teaching materials (lectu4.5×80k919133679
34Jul 18, '25After studying the mathematics and computation of Sparsity for nearly 20 years, I have just realized that it is much more important than I ever realized before. It truly serves as 4.3×78k48731296
35Mar 10, '24When I was younger, I always worshiped people who could find the right solution to a difficult problem. But now I've realized that it is so much harder and so much more important t4.3×77k4884278
36Jun 24, '24The main reason why I have decided to join HKU is that it presents an opportunity to reform the education of data science, computer science, and AI. I believe curricula in these ar4.2×75k3051049
37Sep 5, '24This is our latest work that tries to reveal what diffusion/denoising model is really doing in terms of learning a data distribution: Subspace Clustering and Subspace Denoising. Wh4.1×73k796120673
38Nov 27, '22Probably everyone know what is happening in China these days. When I started my tweeter account, I decided to use it only for academia purposes, not to discuss anything political (3.7×01.1k9724
39Oct 13, '25The lecture slides and videos for the first six weeks of my new course are now posted on the open book website:3.6×65k65298760
40Jul 12, '23The more I study intelligence and I understand it, the more am l baffled why information theory has not played a central and leading role. Intelligence is all about discovering and3.6×64k3142291
41Sep 29, '23Feel strongly that the career of many young researchers around me has been kidnapped. They are blackmailed to pay ransoms periodically for their career, in the form of papers deliv3.5×64k3152949
42Nov 12, '22I always challenge myself every 5 years. Starting in Jan. 1, I will serve as the director of a new Institute of Data Science of Hong Kong University. The institute is very well fun3.5×01.1k5659
43Nov 18, '25This course is nearly done, with 11 lectures (3 hrs each) offered so far. I will give the last lecture next week, discussing some open topics about intelligence. The course has bee3.4×61k45640440
44Jul 30, '22Increasingly, I found many machine learning papers and work do not know the history of their field nor know results in other related fields -- do not even pay the very basic effort3.3×08808175
45Nov 20, '23The main problem with OpenAI is everything it does is not open!3.2×58k303206
46Jul 11, '22At a university, for research in AI, we cannot compete with big tech companies in scale or in speed, but we can for sure compete whose idea/approach is more correct.3.2×08908048
47May 1, '22Some students once asked me how to measure success in research, # of papers or citations? No! Earn respect from people you respect, never envy from people you envy… the one you sho3.2×08919351
48Oct 2, '24Learning, compression, denoising, reducing entropy, and gaining information are all nearly synonymous.3.2×57k61767331
49Oct 12I am offering a new course on Principles of Deep Representation Learning at HKU based on our new open-sourced book: The first six lectures (in English) have been recorded and made 3.1×56k825124867
50Oct 5, '25I am developing and teaching a new course on Principles of Deep Learning and Intelligence at Hong Kong University this semester based on our new open book. The slides of lectures f3.1×56k959161914
51Sep 7, '24I met so many people (students, researchers, and even investors) say they are so interested in understanding intelligence. Yet, they never bother to prepare themselves with the bas3.1×56k33245114
52May 18, '25I have decided to teach a new course in this Fall at Hong Kong University: Principles of Deep Representation Learning based on a new manuscript that my colleagues and students are 3.0×54k37624238
53Jan 6, '23Heard this work from Yu-Xiang Wang of UCSB at SLowDNN. By far the best work known revealing the true role of deep networks in regression, by integrating different pieces of the puz3.0×54k33055131
54Dec 8, '224,5 years ago, I half-jokingly told my students, for good research directions, one should look to the opposite of what everyone is doing: if everyone is using back prop, you should2.9×077246106
55May 5, '25I always believe (and have advocated in the past decade or so) that the theory side of EECS, especially information theory, system/control theory, and signal processing should have2.9×52k35532171
56May 27, '22Classic fields like signal processing, information theory, and control systems should realize by now all their powerful principles and classical models and methods are just one non2.9×067359130
57Sep 25, '25Nothing is better than staying up almost all night helping write ICLR papers while being stranded in a hotel in Rome (due to the Typhoon in Hong Kong). No Roman holidays apparently2.9×52k4651363
58Jul 10, '22Just finished a paper on a topic that has consumed my past five years and it has taken entire three months to write on a daily basis with two colleagues. I have never spent so much2.9×08612087
59Aug 21, '22If folks in control theory, info. theory, game theory, and optimization still don't work together to study intelligence, they will just have to watch ML and AI to reinvent everythi2.8×06977385
60Aug 8, '25I start to believe that there is some subtle difference between compression (common for all intelligence) and abstraction (unique for artificial intelligence of human). They are de2.7×49k43754188
61Feb 18, '25From Google's Switched Transformer, to LoRA, to CRATE, and now to Deepseek's NSA, we see sparse or low-rank structures repeatedly play important roles in processing high-dimensiona2.5×46k37849376
62Aug 23, '22Had a conversation with some students today about what is truly important about research: it is much better to have a wrong solution to a right problem than to have a correct solut2.5×06638740
63Mar 5, '25So, given the recognition of RL by the Turing award, if you seriously want to study the nature of intelligence, please study the history of control and cybernetics. I have given a 2.4×42k36230367
64May 5, '22Finished my Geometry&Learning for 3D Vision course, with a full set of slides: Ended with Lec10 connecting 3D Vision to low-dimensional (sparse, low-rank) models; and Lec11 fro2.4×042487144
65May 10, '22To repair the increasingly fragmented world of knowledge, support all effort for writing books, not 8-pager conference papers...2.3×049364130
66Oct 9, '25When I started at Berkeley in early 2018, I decided to understand deep learning and intelligence, and thought that was once-in-a-decade opportunity. During the pandemic when I star2.2×40k37424119
67Jun 14, '22I was shocked to know that Dr. Jian Sun, my former colleague of the MSRA Visual Computing Group, has passed away. We will miss him dearly. May his soul rest in peace.2.2×0581608
68Sep 17, '25Also I am teaching a new course to nearly 200 graduate students this semester based on the new open-source textbook at the University of Hong Kong. We plan to post all the lecture 2.2×40k41269342
69Oct 17, '25If you’re seriously interested in learning principles of intelligence, please join our tutorial on learning low-dim distributions this Sunday the 19th at ICCV in Hawaii!2.1×38k44558284
70Sep 24, '22The so-called top conferences are becoming the slot machines of academia. Please don't make young researchers addictive to them. We all know the process is highly random, but they 2.1×06134131
71Sep 13, '25Is there any difference between compression and abstraction? Is there any difference between memorizing and understanding. If there is (or suggested so by the new GAUSS report), wh2.0×36k31942195
72Aug 12, '22Information theory, control theory, game theory, optimization/deep networks, machine learning, and statistical learning, all work on the same set of ideas of intelligence, just dif2.0×04755579
73Apr 22, '22Andrew Ng on new IEEE Spectrum cover: "In AI, Small is the New Big." Very baffled: isn't that what I have been advocating against all odds in the past ten years, while Andrew and o2.0×04923475
74Apr 16, '24Talking to many junior faculty members and students in AI lately. Many seem to be somewhat lost with all the seemingly fast progresses made by the industry. My suggestion to them i2.0×35k3114373
75Sep 1, '25Gave my first lecture today at HKU for my new course on Deep Representation Learning using my new book. Spent a fair amount of time talking about the history of Cybernetics movemen1.9×34k32812140
76Apr 26, '22I asked Stephen Boyd to recommended his book editor so that we can get the same deal to release preproduction version of our book online, free for everyone: (on Github, dropbox, CS1.9×033265121
77Sep 17, '22If you set your research goal as to publish in the next conference, then it is highly probable you will never amount to anything significant.1.8×05134234
78Nov 6, '22I am teaching SVD to Berkeley first year students… this is one of the very few magical things that have simultaneously elegant algebraic, geometric, and statistical interpretations1.8×04572075
79Nov 27, '22For all my colleagues who study information science or data science, do you know how rich information that one piece of *white paper* can contain; and how massive information that 1.7×04766122
80Aug 1, '22Found a very interesting pattern in ML conference reviews: for a theoretical paper, the reviewers will ask for experiments instead; for an experimental paper, the reviewers will fo1.7×05103725
81Oct 10, '22I always believe the few subareas of Electrical Engineering: signal processing, control systems, information theory, optimization already have all the key elements of intelligent s1.7×03724385
82Dec 10, '22We wrote the new book both as a textbook for a new data science course and as a timely reference of fundamental mathematical and computational facts for ML/DS research. It systemat1.7×028115164
83Apr 29, '22I was asked in research where intuition of a fundamental idea may come from. Well, like solving a jigsaw puzzle, after you have seen enough pieces, you intuition starts to see the 1.7×04645231
84May 15, '22ICML meta reviews are just arbitrary! In rebuttal we did exactly what reviewers asked, but AC rejected saying: "it is not certain whether this new experiment performed in a short p1.6×04902017
85Aug 8, '22I always tell all my students: do not take outcome of *any* conferences seriously, no matter what others tell you. Focus on doing significant research and writing good papers. Trea1.6×04474132
86Jun 21, '24Our CVPR tutorial on the connection between low-dim models and deep networks was very well received. Now the slides of all lectures are available at the tutorial website: A good th1.5×28k29049232
87Jul 17, '22Only you read profusely about a topic (and history of all its related fields), you may realize how many people in the past have been way ahead of you in terms of ideas. I believe P1.5×04234433
88Nov 12, '22Instead of using blockchain for digital currency and burning billions, can we use it to design voting technology so that every cell phone can be a reliable voting booth? Certifiabl1.5×03843717
89Dec 7, '22My group has 1/100 of the personnel; and 1/1000 of the GPUs... Our only chance is not to follow what's popular, but to find truly new directions.1.5×04481732
90Apr 25, '22One month ago, my former students encouraged me to start a twitter account. One of the main reasons is that my Chinese "twitter" Weibo account might get suspended if I say anything1.5×0513710
91Jun 6, '24If you want to know the newest understanding about deep networks from the perspective of learning low-dim structures, please join our upcoming one-day tutorial at CVPR on June 18 i1.5×27k29058211
92May 15, '22Given my last year experience with ICML -- a paper that was accepted unanimously by four reviewers but eventually rejected by the AC without any concrete evidence, plus this year's1.5×04392021
93Aug 9, '22"Mathematics is the art of giving the same name to different things." -- Poincare. ML/DL seems to be the art of giving different names to the same thing. The scary part is that man1.2×03154027
94May 29, '22Often asked by students why l write books. Well, l don’t know about other people, but if l do not go through everything myself and get all details organized the way l think making 1.2×03711618
95Jul 17, '22Amazing, while I was typing a tweet... noticed that my followers have reached 20K... Only joined twitter less than four months ago, as my Weibo account (Chinese twitter) gets incre1.2×040475
96May 15, '22My biggest fear with sending papers to venues with arbitrary reviews and decisions: after a few encounters, my students will complete distrust my research taste and standards, and 1.1×03191816
97Jul 11, '22Can't someone write an interface between overleaf and arXiv? One button to submit, instead of spending a lot of time modifying the files to conform?1.0×0329920
98Dec 11, '22China will loose its competitive edge if scientists can all be silenced on scientific facts for political reasons — only one voice is allowed. That happened before in history, repe1.0×02981713
99Oct 10, '22It is interesting to see so many people/papers are repeating doing exactly the same (incremental) things to diffusion models as they did to GANs...1.0×02891025
100Nov 11, '22Will be exactly 50 years old tomorrow. But somehow got the feeling I just started to realize what my life is all about. Confucius says: "五十而知天命。" (At age of fifty, knowing your des0.9×0292810
101Dec 3, '22My Chinese Weibo account is suspended for fifteen days, without giving any reason except that "some content violated certain law". Well, as far as I know, forbidding citizens to sp0.9×029196

Scan another creator

@
@karpathy@levelsio@paulg@naval@marclou@illyism

Run this on any account.

XOutlier builds this table for every public profile you open on x.com, with live data and every post, not just the top 50.

Add to Chrome Download .zip