Sayeed Shafayet Chowdhury

@pmddomingos @elonmusk @bayeslord There is hardly anything to bet, we don't even know how to know if we ever reach AGI

@elonmusk @bayeslord Wanna bet that AGI will not arrive in 2026?

lol. cute. singularity is going to shred this flavor of belief into a trillion little pieces

Excited to share that I’ll be joining IU Indianapolis (IU Indy, R1) as a Tenure-Track Assistant Professor in Computer Science (Luddy) in Fall 2026 🎓 I’m recruiting fully funded PhD students to work on: ⚡ Neuromorphic & efficient ML (Spiking Neural Networks, ANN–SNN hybrids, event-based vision) Looking for candidates with CS/ECE (or related) background, strong deep learning skills (PyTorch preferred), and excitement about building a new lab from the ground up. 🔗 Details & how to apply: tinyurl.com/Phd-applicatio… Please retweet/share or tag students who might be interested 🙏 #PhDPositions #PhDOpportunity #PhD #CSPhD #MachineLearning #DeepLearning #ComputerVision #AIforHealth #NeuromorphicAI #SpikingNeuralNetworks #AcademicTwitter #AcademicChatter #IndianaUniversity

The world used to be whoever is right has a louder voice. Now it seems that it has become whoever has a louder voice is right...

When I was a PhD student, many times I heard an unwritten consensus that goes somewhat like: "You write 3 first-authored papers on loosely related topics, then staple them together to make your dissertation." After staying in the industry for a while and trying to ship research that works, I find the consensus above toxic and can instill wrong mindsets for junior researchers. First, the first-authorship definition is toxic. It incentivizes junior researchers to create projects with sufficiently small scopes so that they can first-author. These projects are very different from what they encounter in the real world, where useful products typically require multiple people, if not multiple teams, to collaborate effectively. At the closure of such projects, credit assignment is usually tricky enough that credits are assigned to groups rather than individuals. Going after first authorship completely neglects the collaboration aspect. Case in point: how many authors are there on GPT4, Gemini, LLAMA-3? If those authors were PhD students, how many of them can use some innovations that they made in these papers for their dissertations? Second, the "3 papers on loosely-related topics" is even more ridiculous. It creates the "paper counting" mindset, which in turn incentivize junior researchers to think short-term, like "OK, the NeurIPS deadline just passed. What can I do in 4 months to meet the ICLR deadline?" This way of thinking tells these researchers to avoid pursuing long-term directions. Another terrible consequence of the "paper counting" mindset is that it litters all publication venues with trash submissions and creates unnecessary pressure for reviewing. This pressure, in turn, leads to trash reviews. I am not sure if this is still the PhD standard today. If it is, I hope it changes. PhD students in maths and physics graduate with one paper, sometimes zero paper, all the time. We need to relearn doing slow science: maintain a long-term goal. Ideally, the long-term goal should go with short-term milestones, but these milestones should never compromise the goal itself.

@ArtDeza In the same thread, please check the comment of Yong Zheng-Xin, in his attempt, GPT-4V was correct. So, the results on this particular example are not representative of whether GPT-4V 'can' or 'can't' solve it. But, I agree with the theme of your post, our focus needs calibration

"Computer Vision is Solved. Let's focus on LLM's, raise 50M and throw a bunch of data and GPUs while we melt the glaciers. dATa i5 aLL YoU NeEd, dOnT gEt a pHD. n00bs". Great thread showing this is not the case! 😁

Carlini got fed up breaking every adversarial defense that gets published (rightfully so), and is now using GPT4 to automate it. This is a legit statement paper, with transparency on methods, that should be read as a criticism of the related literature arxiv.org/abs/2307.15008

@spandan_madan :(

Productive day. Completed all 3 TODOs. #ICCV2023

@stanmaltman 2/2 faster, but the bus serves a more general purpose than just my reaching a to b. Perhaps if you don’t want wasteful multiplication, LLMs are not the option, there are far more efficient ubers (multiplication models) available for that

@SayeedShafayet It's not faster than multiplication because it's orders of magnitude more ops. And it matters because it only approximates when a lot of conditions are just right. Also worth considering, the models as a result are full with wasteful networks.

GPT-4 does trillions of calculations per second, but doesn't know how to add or multiply.

@stanmaltman 1/2 well, it's wasteful since it's a generalized model, so multiplication results is neither its soal goal nor its primary objective. I think it like a public bus, if I want to go from a to b, the bus passes through n extra stops taking a route, if I took a uber, I would reach

@stanmaltman As long as the result is close enough, what difference does it make 'how' it produces the result? we can consider it like a LUT, in some cases, it maybe even faster than actual multiplication. Approximate computing does so, I assume (though there actual * is done)

@GeoffreytheC @jackccrawford @pmddomingos It's not approximating, it's recalling. It's technically impossible for it to do multiplication, because it stores numbers as fragments of text. It's recalling similar expressions and selecting probable tokens.

Can we normalize citing co-first authored papers as Author1, Author2 et al. (Year)? Science has changed, with many big projects requiring multiple co-leaders (with diverse knowledge+skills). It doesn't seem like too much to ask to acknowledge both/all co-leaders in this way.

@ICCVConference can we upload pre-trained models to anonymous google drive and share the link in supp? (the files will not be modified after deadline obviously)

#ICCV2023 authors will not be penalized for submitting code via a link to an anonymous GitHub. The code should be on a branch that will not be modified after the supplementary material submission deadline.

@gobindasaha07 @LifeAtPurdue @wacv_official Congratulations!

Our paper "Saliency Guided Experience Packing for Replay in Continual Learning" was selected as an Award Finalist paper at WACV 2023 :):). Paper link: arxiv.org/abs/2109.04954 @LifeAtPurdue #wacv2023 @wacv_official

@Armandi45522247 maybe here- youtube.com/watch?v=VMj-3S…

@mike64_t @anri_m_lombard @karpathy Hi,where can I find the lecture?

I can't explain how amazing @karpathy's lectures are. Andrej's lectures are detailed enough that I could not only follow along, but write my own tensor processing + autograd engine in Java+Kotlin & C++ from scratch. And best of all, it's 2x faster than PyTorch! SIMD for the win!

@jbhuang0604 @HadiZayer @TingHsuan1129 @YaoChihLee I am also a Phd student working on energy-efficient CV algos. I am actively looking for internships in summer 2023.

Three 1st-year Ph.D. students in my lab are looking for summer internships! 🤩 They are smart, driven, and extremely hardworking! Plz feel free to reach out to them! @HadiZayer: hadizayer.github.io @TingHsuan1129: tinghliao.github.io @YaoChihLee: yaochih.github.io

@zamir_ar where should I apply if I'm a phd student?

I will hire again from the Summer@EPFL program this year. Several great projects came out of S@E interns in the past, eg CLIPasso (SIGGRAPH22 best paper), Omnidata (ICCV21). Apply if our interests align. (this is for BS/MS interns. PhD visitors have another program)

@eccvconf how are we going to recognize the shuttles?

#ECCV2022 is providing shuttle service to the main conference venue. Shuttle bus info: ortra.com/events/eccv/Sh…

@kmoretticompsci Thanks a lot Katherine, hope to see you there

@SayeedShafayet Hi Sayeed, exhibits are open Tuesday Oct 25- Thursday Oct 27. Springer is in booth 3 @ Expo Tel Aviv.

Getting ready for #ECCV2022 📚 stop by the Springer booth next week in Tel Aviv!