Top Tweets for #hyperzzw
@sainingxie @thoma_gu @boyangzheng_ an obvious sophistry. #HyperZZW is the first to propose downsampling causes the information loss in the latent space, so it directly compress the image in the pixel space. You & your Meta colleagues plagiarized it many times. You think no we don’t have BG & do these things…
#DyTpaper: Normalization layer includes normalize/standardization and affine (scale & shift) operations.
#Github issue: Affine is a separate layer.
How do you think?
@ylecun @boredyannlecun
In addition, you and your colleagues have multiple paper ideas similar to #HyperZZW.
The DyT authors think that the #affine parameters are a #separate layer from #normalization.
I think it is wrong.
Below is my response, although they closed the issue.
https://t.co/6a42XWmMvH
This is the same fucking idea as #HyperZZW!
No latent space
No downsampling layer
Less information loss
😅😅😅
At WAIC 2025, Hinton once again emphasized the importance of fast weights. I think I have a deeper understanding of this concept than you. The future belongs to #HyperZZW, although it is now in adversity. In addition, this concept alone cannot change the dilemma of AI.
Geoffrey Hinton:
We don’t use fast weight sounds suspicious!
@iclr_conf 🔍 Formula comparison evidence:
#HC and #SDD papers' core formula vs. our 2024 #HyperZZW work → completely consistent
@_albertgu 1) My #HyperZZW proposal was formed earlier than your #Mamba. (October 2022).
2) I tried #detokenization a long time ago.
3) The future architecture will be based on #HyperZZW instead of #Mamba.
https://t.co/CIDUbVISg5
https://t.co/Tw3rBrVSqC
https://t.co/z3VPYLxGxT
We no longer need to divide the image into #patches to calculate attention scores between them.
This also tells us that converting images into tokens in ViT is unreasonable.
As we know, the widely used #norm includes three parts: affine params, momentum, standardization. This paper removes the latter two.
I think the motivation of #KaimingHe’s recent papers is very similar to my #HyperZZW work. I once shared my research with him…
#HyperZZW: However, the additive connections between layers in residual learning, while addressing the performance degeneration, pose challenges
for the exploration of interpretable architectures [33, 45] and the adoption of greedy layerwise training [27].
#Modularization
I doubt that the recent #HyperConnections was inspired by my #HyperZZW and published in ICLR 2025.
Link: https://t.co/fxLIxtebeF
In addition, AC mentioned one work (FractalNet) in the meta review, and this work was discussed in #HKM's latest Fractal Generative Models.
My #Terminator network will ...
1. As I stated before, the gradient vanishing problem of neural networks has been exaggerated over the past few decades.
2. Why Highway Net performs worse than ResNet?
3. A new variant of FWP is my #HyperZZW operator!
4.
We just released our interview with the father of Generative AI - @SchmidhuberAI!
The G, P, and T in "ChatGPT" (GPT means "Generative Pre-Trained Transformer") go back to Juergen's work of 1990-91 when he published what's now called "Unnormalised Linear Transformers," "Self-Supervised Pre-Training" for deep learning with long texts, and "Generative Adversarial Networks" for Artificial Curiosity. Remarkably, principles of both Transformers and LSTMs date back to 1991, the only palindromic year of the 20th century!
Transformers are easier to parallelise, but LSTMs can solve problems which are unsolvable by Transformers. In this first part of our two part show, we discuss the history and the future of the field, with a focus on abstract planning, reasoning, and "learning to think."
It's just dropped on MLST!
Last Seen Hashtags on Sotwe
طيز
Seen from Germany
femboy
Seen from United States
ซอยมังกร
Seen from Thailand
femboy türk
Seen from France
Homeless
Seen from Australia
นายแบบชักว่าว
Seen from Thailand
minichat()********************************************* *filter:native_video()
Seen from Poland
gettr
Seen from United States
chinese #nsfw #asian
Seen from Indonesia
รับงานสายไหม
Seen from Thailand
Most Popular Users
Elon Musk 
@elonmusk
241.5M followers
Barack Obama
@barackobama
118.8M followers
Cristiano Ronaldo
@cristiano
114.5M followers
Donald J. Trump
@realdonaldtrump
111.9M followers
Narendra Modi
@narendramodi
107.1M followers
Rihanna
@rihanna
98.7M followers
NASA
@nasa
92.3M followers
Justin Bieber
@justinbieber
91.8M followers
KATY PERRY
@katyperry
90M followers
Taylor Swift
@taylorswift13
84M followers
Lady Gaga
@ladygaga
75.5M followers
Virat Kohli
@imvkohli
73.5M followers
Kim Kardashian
@kimkardashian
70.9M followers
YouTube
@youtube
68.8M followers
Neymar Jr
@neymarjr
66.5M followers
Bill Gates
@billgates
65.3M followers
Selena Gomez
@selenagomez
63.1M followers
The Ellen Show
@theellenshow
62.2M followers
CNN
@cnn
61.7M followers
X
@x
60.7M followers