Fun viz: if you do a pca of Cifar10 with random crop and flip augmentations, you get this
February 4, 2026 at 4:51 PM
Fun viz: if you do a pca of Cifar10 with random crop and flip augmentations, you get this
There was a bug, it's now fixed (still not working properly).
If you want to generate this kind of images, it's actually fairly simple: train an autoregressive MLP on a cifar10 (with positional encoding).
If you want to generate this kind of images, it's actually fairly simple: train an autoregressive MLP on a cifar10 (with positional encoding).
March 15, 2025 at 11:07 PM
There was a bug, it's now fixed (still not working properly).
If you want to generate this kind of images, it's actually fairly simple: train an autoregressive MLP on a cifar10 (with positional encoding).
If you want to generate this kind of images, it's actually fairly simple: train an autoregressive MLP on a cifar10 (with positional encoding).
CIFAR10 tends to be a *little* less silly than MNIST. It can still be informative if you can scale your method down enough. I guess for MNIST that’s kinda true too, but you need to be able to keep the same structure while you scale your total params down to like 25.
August 25, 2026 at 1:26 PM
CIFAR10 tends to be a *little* less silly than MNIST. It can still be informative if you can scale your method down enough. I guess for MNIST that’s kinda true too, but you need to be able to keep the same structure while you scale your total params down to like 25.
"My classifier always gets the right class for CIFAR10 in less than 10 guesses" 🙃
December 13, 2024 at 7:33 AM
"My classifier always gets the right class for CIFAR10 in less than 10 guesses" 🙃
I'd love to have a symposium on the current state-of-the-art in Julia + things like a hackathon, a training competition (e.g., reproducing the CIFAR speedruns), etc.
Anyone else interested in this?
#julia #julialang
Anyone else interested in this?
#julia #julialang
GitHub - KellerJordan/cifar10-airbench: CIFAR-10 speedruns: 94% in 2.6 seconds and 96% in 27 seconds
CIFAR-10 speedruns: 94% in 2.6 seconds and 96% in 27 seconds - KellerJordan/cifar10-airbench
github.com
October 7, 2025 at 4:59 PM
I'd love to have a symposium on the current state-of-the-art in Julia + things like a hackathon, a training competition (e.g., reproducing the CIFAR speedruns), etc.
Anyone else interested in this?
#julia #julialang
Anyone else interested in this?
#julia #julialang
In experiments across MLPs and ResNets on CIFAR10 and ViTs on ImageNet1K, we show that 𝝁P² indeed jointly transfers optimal learning rate and perturbation radius across model scales and can improve training stability and generalization.
🧵 8/10
December 10, 2024 at 7:08 AM
In experiments across MLPs and ResNets on CIFAR10 and ViTs on ImageNet1K, we show that 𝝁P² indeed jointly transfers optimal learning rate and perturbation radius across model scales and can improve training stability and generalization.
🧵 8/10
Driven by these findings, we develop variants of local-SSL (CLAPP++). They reach the performance of BP baselines on CIFAR10, STL-10, Tiny-ImageNet, while also setting new SOTA of local learning rules on these dataset and ImageNet. Bonus: 40-60% less GPU VRAM and shorter wall clock time than BP.
June 16, 2026 at 1:42 PM
Driven by these findings, we develop variants of local-SSL (CLAPP++). They reach the performance of BP baselines on CIFAR10, STL-10, Tiny-ImageNet, while also setting new SOTA of local learning rules on these dataset and ImageNet. Bonus: 40-60% less GPU VRAM and shorter wall clock time than BP.
#AROS💍 leverages neural ODEs and Lyapunov stability theory to craft an embedding method to smartly detect OOD samples. Strikingly, we can improve performance on popular adversarial detection benchmarks such as CIFAR10 vs CIFAR100 by over 40% 👏
🔥🚀 we are excited to keep pushing this line of work 💪
🔥🚀 we are excited to keep pushing this line of work 💪
November 27, 2024 at 9:12 PM
#AROS💍 leverages neural ODEs and Lyapunov stability theory to craft an embedding method to smartly detect OOD samples. Strikingly, we can improve performance on popular adversarial detection benchmarks such as CIFAR10 vs CIFAR100 by over 40% 👏
🔥🚀 we are excited to keep pushing this line of work 💪
🔥🚀 we are excited to keep pushing this line of work 💪
Looks cool. Couple of stupid questions - 1) does it work on ImageNet scale? I see only imagenette, and the benefit there already vanishes. What about speed/scalability as well?
Also CIFAR10/100 results are really terrible. I have got >90% in 2016, and yours are around 70%
Also CIFAR10/100 results are really terrible. I have got >90% in 2016, and yours are around 70%
December 1, 2025 at 6:57 PM
Looks cool. Couple of stupid questions - 1) does it work on ImageNet scale? I see only imagenette, and the benefit there already vanishes. What about speed/scalability as well?
Also CIFAR10/100 results are really terrible. I have got >90% in 2016, and yours are around 70%
Also CIFAR10/100 results are really terrible. I have got >90% in 2016, and yours are around 70%
There are a lot of different datasets in computer vision, with various dimensions, number of classes, text annotations or different kind of structural information (e.g., bounding boxes). Some popular general datasets are MNIST, CIFAR10/100, Pascal VOC, ImageNet and COCO.
December 3, 2024 at 6:00 AM
There are a lot of different datasets in computer vision, with various dimensions, number of classes, text annotations or different kind of structural information (e.g., bounding boxes). Some popular general datasets are MNIST, CIFAR10/100, Pascal VOC, ImageNet and COCO.
∇QDARTS: Quantization as an Elastic Dimension to Differentiable NAS
Payman Behnam, Uday Kamal, Sanjana Vijay Ganesh et al.
Action editor: Naigang Wang
https://openreview.net/forum?id=ubrOSWyTS8
#imagenet #cifar10 #optimized
Payman Behnam, Uday Kamal, Sanjana Vijay Ganesh et al.
Action editor: Naigang Wang
https://openreview.net/forum?id=ubrOSWyTS8
#imagenet #cifar10 #optimized
June 5, 2025 at 12:07 AM
∇QDARTS: Quantization as an Elastic Dimension to Differentiable NAS
Payman Behnam, Uday Kamal, Sanjana Vijay Ganesh et al.
Action editor: Naigang Wang
https://openreview.net/forum?id=ubrOSWyTS8
#imagenet #cifar10 #optimized
Payman Behnam, Uday Kamal, Sanjana Vijay Ganesh et al.
Action editor: Naigang Wang
https://openreview.net/forum?id=ubrOSWyTS8
#imagenet #cifar10 #optimized
Revisiting the unsupervised learning of weights with competing hidden units and giving the project some (deserved) love. See what happens when you train 100 hidden units on CIFAR10 (live, on CPU).
🐙 github.com/dcasbol/biol...
🐙 github.com/dcasbol/biol...
December 28, 2024 at 10:29 PM
Revisiting the unsupervised learning of weights with competing hidden units and giving the project some (deserved) love. See what happens when you train 100 hidden units on CIFAR10 (live, on CPU).
🐙 github.com/dcasbol/biol...
🐙 github.com/dcasbol/biol...
such as ResNet-101 and Vision Transformer (VIT). On various computer vision benchmarks including CIFAR10, CIFAR100, and ImageNet1k. Our approach compresses model parameters by approximately 50x and reduces model size by 75, while achieving accuracy [5/6 of https://arxiv.org/abs/2505.17856v1]
May 26, 2025 at 6:18 AM
such as ResNet-101 and Vision Transformer (VIT). On various computer vision benchmarks including CIFAR10, CIFAR100, and ImageNet1k. Our approach compresses model parameters by approximately 50x and reduces model size by 75, while achieving accuracy [5/6 of https://arxiv.org/abs/2505.17856v1]
Researchers introduced tighter PAC‑Bayesian risk certificates for contrastive learning; CIFAR‑10 tests show the bounds closely match observed empirical errors. Read more: https://getnews.me/new-tight-pac-bayesian-certificates-boost-contrastive-learning-theory/ #pacbayesian #cifar10
September 20, 2025 at 3:37 PM
Researchers introduced tighter PAC‑Bayesian risk certificates for contrastive learning; CIFAR‑10 tests show the bounds closely match observed empirical errors. Read more: https://getnews.me/new-tight-pac-bayesian-certificates-boost-contrastive-learning-theory/ #pacbayesian #cifar10
I know it’s only cifar10 but sota fid without any mini batch OT was already < 2 with 35nfe back in 2022.
Are you sure this makes any difference in the competitive setting? Seems like choosing hyper params makes more of a difference
arxiv.org/abs/2206.00364
Are you sure this makes any difference in the competitive setting? Seems like choosing hyper params makes more of a difference
arxiv.org/abs/2206.00364
June 16, 2025 at 11:18 AM
I know it’s only cifar10 but sota fid without any mini batch OT was already < 2 with 35nfe back in 2022.
Are you sure this makes any difference in the competitive setting? Seems like choosing hyper params makes more of a difference
arxiv.org/abs/2206.00364
Are you sure this makes any difference in the competitive setting? Seems like choosing hyper params makes more of a difference
arxiv.org/abs/2206.00364
Ever wondered how many black-box queries it takes to peek inside a ViT's hidden layers? Researchers cracked the CIFAR-10 vision transformer with just 8,193 queries, revealing feed-forward dynamics and Hessian curvature. Dive in! #VisionTransformer #BlackBoxQueries #CIFAR10
🔗
🔗
September 1, 2026 at 5:58 AM
Ever wondered how many black-box queries it takes to peek inside a ViT's hidden layers? Researchers cracked the CIFAR-10 vision transformer with just 8,193 queries, revealing feed-forward dynamics and Hessian curvature. Dive in! #VisionTransformer #BlackBoxQueries #CIFAR10
🔗
🔗
[7/🧵] We show that PMAE outperforms MAEs in downstream image classification on CIFAR10, TinyImageNet and MedMNIST datasets.
Using a ViT-Tiny, we observe an average 38% improvement in linear probing performance compared to MAEs with the standard 75% masking ratio.
Using a ViT-Tiny, we observe an average 38% improvement in linear probing performance compared to MAEs with the standard 75% masking ratio.
March 19, 2025 at 8:44 PM
[7/🧵] We show that PMAE outperforms MAEs in downstream image classification on CIFAR10, TinyImageNet and MedMNIST datasets.
Using a ViT-Tiny, we observe an average 38% improvement in linear probing performance compared to MAEs with the standard 75% masking ratio.
Using a ViT-Tiny, we observe an average 38% improvement in linear probing performance compared to MAEs with the standard 75% masking ratio.
Added a whole new model for CIFAR10, since the other are shit
June 11, 2025 at 12:37 PM
Added a whole new model for CIFAR10, since the other are shit
Lookahead drifting model sequentially computes drifting terms for improved image generation performance on CIFAR10.
Lookahead drifting model sequentially computes drifting terms for improved image generation performance on CIFAR10.
arXiv cs.LG · ProbBrain News
probbrain.com
May 7, 2026 at 4:09 AM
Lookahead drifting model sequentially computes drifting terms for improved image generation performance on CIFAR10.
space exploration using MobileNetV1 with the CIFAR10 dataset. In the architecture, we introduce an additional non-convolutional unit, which merges the dequantization, batch normalization (BN), ReLU, and quantization between DWC and PWC into a simple [3/6 of https://arxiv.org/abs/2503.11707v1]
March 18, 2025 at 5:54 AM
space exploration using MobileNetV1 with the CIFAR10 dataset. In the architecture, we introduce an additional non-convolutional unit, which merges the dequantization, batch normalization (BN), ReLU, and quantization between DWC and PWC into a simple [3/6 of https://arxiv.org/abs/2503.11707v1]
This results in real-world speedups; there are two slightly slower cases because the models just didn't cycle. But those are on CIFAR10, as the dimension goes up we see better results because more of the mass exists on the edge of the ∞-norm ball, making cycling easier.
June 16, 2025 at 1:45 PM
This results in real-world speedups; there are two slightly slower cases because the models just didn't cycle. But those are on CIFAR10, as the dimension goes up we see better results because more of the mass exists on the edge of the ∞-norm ball, making cycling easier.
achieve better performance than ResNet. Experiments on the CIFAR10, CIFAR100, and SVHN datasets prove that this network can stably achieve considerable improvements over ResNet by simply making tiny corresponding changes to the original ResNet network [5/6 of https://arxiv.org/abs/2506.00992v1]
June 3, 2025 at 6:11 AM
achieve better performance than ResNet. Experiments on the CIFAR10, CIFAR100, and SVHN datasets prove that this network can stably achieve considerable improvements over ResNet by simply making tiny corresponding changes to the original ResNet network [5/6 of https://arxiv.org/abs/2506.00992v1]
on widely used domain-shift benchmarks, including CIFAR10-C and ImageNet-C, across various model architectures. With significant compression, it achieves accuracy improvements of 7.96%p on CIFAR10-C and 5.37%p on ImageNet-C over the full-precision TTA [6/7 of https://arxiv.org/abs/2505.20890v1]
May 28, 2025 at 6:06 AM
on widely used domain-shift benchmarks, including CIFAR10-C and ImageNet-C, across various model architectures. With significant compression, it achieves accuracy improvements of 7.96%p on CIFAR10-C and 5.37%p on ImageNet-C over the full-precision TTA [6/7 of https://arxiv.org/abs/2505.20890v1]