Peyman M. Kiasari
@kiasari.bsky.social
ML researcher and engineer | currently focusing on computer vision, especially DS-CNNs
M. Lindberg can be a hallucination of T. Lindeberg, he is also Swedish.
September 26, 2025 at 11:07 AM
M. Lindberg can be a hallucination of T. Lindeberg, he is also Swedish.
Thank you!
Thats a good question and we have to clarify this in camera ready version.
In Table 1, in "Acc with 8 (greedy search)", we have calculated the proportions of each kind of filter in each layer; we use that in Table 2.
Thats a good question and we have to clarify this in camera ready version.
In Table 1, in "Acc with 8 (greedy search)", we have calculated the proportions of each kind of filter in each layer; we use that in Table 2.
September 23, 2025 at 8:36 AM
Thank you!
Thats a good question and we have to clarify this in camera ready version.
In Table 1, in "Acc with 8 (greedy search)", we have calculated the proportions of each kind of filter in each layer; we use that in Table 2.
Thats a good question and we have to clarify this in camera ready version.
In Table 1, in "Acc with 8 (greedy search)", we have calculated the proportions of each kind of filter in each layer; we use that in Table 2.
Our paper has been accepted to #NeurIPS2025 as poster 🥳
Looking forward to presenting our poster.
Looking forward to presenting our poster.
September 22, 2025 at 12:28 AM
Our paper has been accepted to #NeurIPS2025 as poster 🥳
Looking forward to presenting our poster.
Looking forward to presenting our poster.
Of our six challenges, I expect future AI to solve 'shortest path' first, but GPT-5 still has a long way to go.
August 10, 2025 at 8:59 PM
Of our six challenges, I expect future AI to solve 'shortest path' first, but GPT-5 still has a long way to go.
Is #GPT5 the dreamed AGI? Not even close!
It still can't solve our easiest task, which humans score 100%: What is the shortest path between the two square nodes?
This is from our challenge the Visual Graph Arena (vga.csail.mit.edu)
It still can't solve our easiest task, which humans score 100%: What is the shortest path between the two square nodes?
This is from our challenge the Visual Graph Arena (vga.csail.mit.edu)
August 10, 2025 at 8:59 PM
Is #GPT5 the dreamed AGI? Not even close!
It still can't solve our easiest task, which humans score 100%: What is the shortest path between the two square nodes?
This is from our challenge the Visual Graph Arena (vga.csail.mit.edu)
It still can't solve our easiest task, which humans score 100%: What is the shortest path between the two square nodes?
This is from our challenge the Visual Graph Arena (vga.csail.mit.edu)
Every time I write a paper, I ask the best LLM I have to explain it to me in detail.
Never got it right. Always made up delusional stuff.
Never got it right. Always made up delusional stuff.
Kinda wild that the LLM reviews in reviews.cspaper.org correctly predicts the really mistaken reviewer comments
reviews.cspaper.org
July 30, 2025 at 10:23 PM
Every time I write a paper, I ask the best LLM I have to explain it to me in detail.
Never got it right. Always made up delusional stuff.
Never got it right. Always made up delusional stuff.
We’re presenting our work today at #ICML2025, Poster Session 4 West, W-214!
If you are interested in computer vision reasoning and multimodal LLMs come visit us!
If you are interested in computer vision reasoning and multimodal LLMs come visit us!
July 16, 2025 at 5:11 PM
We’re presenting our work today at #ICML2025, Poster Session 4 West, W-214!
If you are interested in computer vision reasoning and multimodal LLMs come visit us!
If you are interested in computer vision reasoning and multimodal LLMs come visit us!
Our paper "Visual Graph Arena: Evaluating AI's Visual Conceptualization" has been accepted at #ICML2025! 🎉🥳🎉
We introduce a dataset testing AI systems' ability to conceptualize graph representations.
Available at: vga.csail.mit.edu
More info + Camera ready version coming soon!
We introduce a dataset testing AI systems' ability to conceptualize graph representations.
Available at: vga.csail.mit.edu
More info + Camera ready version coming soon!
May 1, 2025 at 3:24 PM
Our paper "Visual Graph Arena: Evaluating AI's Visual Conceptualization" has been accepted at #ICML2025! 🎉🥳🎉
We introduce a dataset testing AI systems' ability to conceptualize graph representations.
Available at: vga.csail.mit.edu
More info + Camera ready version coming soon!
We introduce a dataset testing AI systems' ability to conceptualize graph representations.
Available at: vga.csail.mit.edu
More info + Camera ready version coming soon!
Thanks for your reply Paul. I appreciate the clarification.
I'll be looking forward to seeing more of your work in the future.
I'll be looking forward to seeing more of your work in the future.
February 17, 2025 at 9:48 PM
Thanks for your reply Paul. I appreciate the clarification.
I'll be looking forward to seeing more of your work in the future.
I'll be looking forward to seeing more of your work in the future.
Ironically, my post engagements are higher on 🦋 than on Twitter.
February 8, 2025 at 12:49 AM
Ironically, my post engagements are higher on 🦋 than on Twitter.
Camera-ready version is out! (arxiv.org/abs/2412.16751)
TL;DR: Deep CNN filters may be generalized, not specialized as previously believed.
Major update(Fig4): We froze layers from end-to-start instead of start-to-end. The result ironically suggests early layers are specialized!
TL;DR: Deep CNN filters may be generalized, not specialized as previously believed.
Major update(Fig4): We froze layers from end-to-start instead of start-to-end. The result ironically suggests early layers are specialized!
February 8, 2025 at 12:31 AM
Camera-ready version is out! (arxiv.org/abs/2412.16751)
TL;DR: Deep CNN filters may be generalized, not specialized as previously believed.
Major update(Fig4): We froze layers from end-to-start instead of start-to-end. The result ironically suggests early layers are specialized!
TL;DR: Deep CNN filters may be generalized, not specialized as previously believed.
Major update(Fig4): We froze layers from end-to-start instead of start-to-end. The result ironically suggests early layers are specialized!
Hi Paul,
I finally managed to look at Eq4.
I believe it doesn't represent DS-CNNs. Each kernel is convolved into a separate feature map, and you can't factor them out. (I marked the part I don't think represents DS-CNN in red)
Overall, DS-CNNs are not LCing kernels. They are LCing featuremaps.
I finally managed to look at Eq4.
I believe it doesn't represent DS-CNNs. Each kernel is convolved into a separate feature map, and you can't factor them out. (I marked the part I don't think represents DS-CNN in red)
Overall, DS-CNNs are not LCing kernels. They are LCing featuremaps.
February 5, 2025 at 5:23 PM
Hi Paul,
I finally managed to look at Eq4.
I believe it doesn't represent DS-CNNs. Each kernel is convolved into a separate feature map, and you can't factor them out. (I marked the part I don't think represents DS-CNN in red)
Overall, DS-CNNs are not LCing kernels. They are LCing featuremaps.
I finally managed to look at Eq4.
I believe it doesn't represent DS-CNNs. Each kernel is convolved into a separate feature map, and you can't factor them out. (I marked the part I don't think represents DS-CNN in red)
Overall, DS-CNNs are not LCing kernels. They are LCing featuremaps.
This is great work! and actually, we've cited you on (arxiv.org/abs/2401.14469)
Maybe we can cite it again on "Master key filters" again as another visual evidence 👌
Nice work, by the way.
Maybe we can cite it again on "Master key filters" again as another visual evidence 👌
Nice work, by the way.
Unveiling the Unseen: Identifiable Clusters in Trained Depthwise Convolutional Kernels
Recent advances in depthwise-separable convolutional neural networks (DS-CNNs) have led to novel architectures, that surpass the performance of classical CNNs, by a considerable scalability and accura...
arxiv.org
January 27, 2025 at 6:27 PM
This is great work! and actually, we've cited you on (arxiv.org/abs/2401.14469)
Maybe we can cite it again on "Master key filters" again as another visual evidence 👌
Nice work, by the way.
Maybe we can cite it again on "Master key filters" again as another visual evidence 👌
Nice work, by the way.
Thanks, I'll read it as soon as I get the chance to see that.
January 27, 2025 at 6:23 PM
Thanks, I'll read it as soon as I get the chance to see that.
Hi Paul, Thank you for joining!
Actually, two people referenced your work : D
Please correct me if I'm wrong. Are you sure that pointwise layers are LCing the "filters"? I'm having difficulties seeing that.
If we name filters as K and features as F, how can this result in LC of Ks?
Actually, two people referenced your work : D
Please correct me if I'm wrong. Are you sure that pointwise layers are LCing the "filters"? I'm having difficulties seeing that.
If we name filters as K and features as F, how can this result in LC of Ks?
January 25, 2025 at 1:20 AM
Hi Paul, Thank you for joining!
Actually, two people referenced your work : D
Please correct me if I'm wrong. Are you sure that pointwise layers are LCing the "filters"? I'm having difficulties seeing that.
If we name filters as K and features as F, how can this result in LC of Ks?
Actually, two people referenced your work : D
Please correct me if I'm wrong. Are you sure that pointwise layers are LCing the "filters"? I'm having difficulties seeing that.
If we name filters as K and features as F, how can this result in LC of Ks?
That's a good point. No they are not. Actually, In our paper (DS-CNNs models) each filter gets convolved into a separate feature map and generates a distinct new feature map - The new feature maps are linearly combined but not the filters.
January 24, 2025 at 12:42 AM
That's a good point. No they are not. Actually, In our paper (DS-CNNs models) each filter gets convolved into a separate feature map and generates a distinct new feature map - The new feature maps are linearly combined but not the filters.
IMHO, they are not actually using frozen filters. Let me explain myself:
A model learning (x,y,z) is mathematically equivalent to one using LC of "frozen" filters (1,0,0), (0,1,0), (0,0,1). They're doing the same optimization, just expressed differently. Same goes for LC of random filters.
A model learning (x,y,z) is mathematically equivalent to one using LC of "frozen" filters (1,0,0), (0,1,0), (0,0,1). They're doing the same optimization, just expressed differently. Same goes for LC of random filters.
January 24, 2025 at 12:22 AM
IMHO, they are not actually using frozen filters. Let me explain myself:
A model learning (x,y,z) is mathematically equivalent to one using LC of "frozen" filters (1,0,0), (0,1,0), (0,0,1). They're doing the same optimization, just expressed differently. Same goes for LC of random filters.
A model learning (x,y,z) is mathematically equivalent to one using LC of "frozen" filters (1,0,0), (0,1,0), (0,0,1). They're doing the same optimization, just expressed differently. Same goes for LC of random filters.
This is absolutely a great suggestion and we have to have this experiment too! Our GPUs are currently working for ICML, and after that I definitely do this experiment before AAAI last refinement deadline. I'll update you on this. (but in case you wonder about only frozen pointwise see Table 5)
January 24, 2025 at 12:12 AM
This is absolutely a great suggestion and we have to have this experiment too! Our GPUs are currently working for ICML, and after that I definitely do this experiment before AAAI last refinement deadline. I'll update you on this. (but in case you wonder about only frozen pointwise see Table 5)
I found out that they they create new filters through linear combinations of random filters, which isn't what we're doing. 🤔
And mathematically, LC of 49 random filters should span the entire 7x7 space, so it's not surprising that it works.
Open to discussion if I'm misunderstanding something!
January 23, 2025 at 11:23 PM
I found out that they they create new filters through linear combinations of random filters, which isn't what we're doing. 🤔
And mathematically, LC of 49 random filters should span the entire 7x7 space, so it's not surprising that it works.
Open to discussion if I'm misunderstanding something!
Fascinating that you mention this paper - our area chair noted this connection too! (Hat tip to the author @paulgavrikov.bsky.social)
After reading the paper, TBH, I couldn't see a deep connection. And I'm open to being wrong since you and AC both pointed this out. If I am wrong, please correct me.
After reading the paper, TBH, I couldn't see a deep connection. And I'm open to being wrong since you and AC both pointed this out. If I am wrong, please correct me.
January 23, 2025 at 11:23 PM
Fascinating that you mention this paper - our area chair noted this connection too! (Hat tip to the author @paulgavrikov.bsky.social)
After reading the paper, TBH, I couldn't see a deep connection. And I'm open to being wrong since you and AC both pointed this out. If I am wrong, please correct me.
After reading the paper, TBH, I couldn't see a deep connection. And I'm open to being wrong since you and AC both pointed this out. If I am wrong, please correct me.
I've explained those classical CNN parts in these two tweets:
bsky.app/profile/kias...
and
bsky.app/profile/kias...
bsky.app/profile/kias...
and
bsky.app/profile/kias...
5/13 We had to test with CNNs too. ResNet50 (a classical CNN) maintained 92.5% of its original accuracy when all layers were transferred! This number reaches to 93-94% for ResNet152.
But why? is this because of residual connections?
But why? is this because of residual connections?
January 23, 2025 at 11:21 PM
I've explained those classical CNN parts in these two tweets:
bsky.app/profile/kias...
and
bsky.app/profile/kias...
bsky.app/profile/kias...
and
bsky.app/profile/kias...
Hi Neil, Thank you for showing interest in our work!
We have experimented with both DS-CNNs and classical CNNs (ResNets in our paper, and you are right that our main focus was DS-CNNs). In DS-CNNs we only frozen depthwise filters but in classical CNNs all params are frozen just like Yosinski did.
We have experimented with both DS-CNNs and classical CNNs (ResNets in our paper, and you are right that our main focus was DS-CNNs). In DS-CNNs we only frozen depthwise filters but in classical CNNs all params are frozen just like Yosinski did.
January 23, 2025 at 11:21 PM
Hi Neil, Thank you for showing interest in our work!
We have experimented with both DS-CNNs and classical CNNs (ResNets in our paper, and you are right that our main focus was DS-CNNs). In DS-CNNs we only frozen depthwise filters but in classical CNNs all params are frozen just like Yosinski did.
We have experimented with both DS-CNNs and classical CNNs (ResNets in our paper, and you are right that our main focus was DS-CNNs). In DS-CNNs we only frozen depthwise filters but in classical CNNs all params are frozen just like Yosinski did.