Flying Scholars – CS retweeted
Replying to @flowersslop
I think they meant "log progress": double the dataset size, get a (small) constant increase in performance.
Flying Scholars – CS retweeted
> me waiting for one of these companies to release a simple search engine on anna's archive, instead of slop-washing it through "AI"
Flying Scholars – CS retweeted
Question for a mathematician: if the fourier transform of a Dirac at 0 is the sum of all 0-phased fourier components, what is the fourier transform of a signal containing 2 diracs, one at 0 and one at 1?
Flying Scholars – CS retweeted
7/7 papers accepted at #IAMTHEBEST. Not one worth mentioning the content. Congratulations in the replies will be met with condescension. GFY
Flying Scholars – CS retweeted
"The Neuron as a Direct Data-Driven Controller"
🔗 arxiv.org/abs/2401.01489
🎯 Thinking of biological neurons as trying to stabilize their own local environment could explain many of their properties: spike-timing dependent plasticity, contrast adaptation, temporal integration..
Flying Scholars – CS retweeted
Calling bullshit.
This is big. Researchers were able to generate images directly from brain EEG signals.
It's the best performing “thoughts-to-image" model yet to be released.
Here's how they did it:
▸ EEG signals are collected from the brain and a masked signal pre-training is used to create an effective EEG encoder.
▸ The encoder is fine-tuned with limited EEG-image pairs using Stable Diffusion.
▸ The EEG, text, and image spaces are aligned using CLIP encoders to create a unified space.
▸The EEG embedding representation is optimized for image generation using a loss function.
▸ High-quality images are generated from the optimized EEG representation using Stable Diffusion.
Flying Scholars – CS retweeted
Nice review of the MLP-Mixer.
I disagree with "...very simple deep learning architecture for vision". No, it is well engineered and complex architecture.
I agree with: the most of the review.
We reviewed
"MLP-Mixer: An all-MLP Architecture for Vision"
by Ilya Tolstikhin, Neil Houlsby, Alexander Kolesnikov, Lucas Beyer et al.
arxiv.org/abs/2105.01601v4
Review by ~anonymous
rentry.co/rev_fs_0003
Flying Scholars – CS retweeted
"I believe that including a term to maximize mutual information between segment representations and the generated sequence of phonemes may result in better control over the textual information in the transcription..."
Full commentary by @OswaldoLudwig👇
Some thoughts on the work Unsupervised Speech Recognition.
researchgate.net/publication…
Flying Scholars – CS retweeted
"Technical goals in ML like accuracy are usually considered neutral for society. This study shows how they might in fact further support the concentration of power by Big Tech. However, the argumentation is one-sided and blind to the societal benefits of these same goals..."
We reviewed "The Values Encoded in Machine Learning Research"!
by Abeba Birhane, Pratyusha Kalluri, Dallas Card, William Agnew, Ravit Dotan, Michelle Bao
arxiv.org/abs/2106.15590v1
Review by ~Anonymous
rentry.co/rev_fs_0004
Flying Scholars – CS retweeted
We reviewed
"MLP-Mixer: An all-MLP Architecture for Vision"
by Ilya Tolstikhin, Neil Houlsby, Alexander Kolesnikov, Lucas Beyer et al.
arxiv.org/abs/2105.01601v4
Review by ~anonymous
rentry.co/rev_fs_0003

















