Archana Swaminathan (@TweetsArchana) profile picture

PhD student at @umdcs | ML, CV and the likes

Joined August 2020
Photo by Archana Swaminathan (@TweetsArchana)
Archana Swaminathan retweeted
Replying to @VirginAtlantic
@VirginAtlantic It's been 4 days since my BLR→BOM→LHR→IAD trip and there's still no sign of my bags. BOM→LHR leg was delayed and I missed my connection. Was rebooked onto @united, arriving 19 hours late. But my bags never made it to IAD! @united says they never scanned them.
1
5
7
241
We introduce LEIA, a novel approach for representing 3D articulated objects by learning view-invariant latent embeddings for different articulation states. Visit our poster #299 at @eccvconf #ECCV2024 in Milan on Oct 3rd (Thursday), 10:30am. Project Page: archana1998.github.io/leia/
4
5
7
1,888
Work done with @learn2phoenix , @kamalgupta09, Shishira, Vatsal and my awesome advisor @abhi2610.
1
176
We see consistent improvement over previous SoTA that was limited to single-part articulation. Additionally, we also show the latent space can be disentangled for representing different object part motion!
1
170
In conjunction with a hypernetwork that modulates the NeRF weights, LEIA can linearly interpolate between the learned embedding space to generate unseen in-between articulation states. It's robust to both single and multi-part articulation!
1
109
LEIA is a method of generating unseen articulation states from just observing start and end state of articulation. We have a learnable latent dictionary for state-specific embeddings, and then enforce a manifold loss for structure in latent space.
1
121
Archana Swaminathan retweeted
Unconditional #diffusion pretraining learns discriminative representation; hence are great candidates for Unified Models for #GenAI as well as #recognition. Please join us at our #DiffSSL poster (#101) at #ECCV on 1 Oct 1030AM. @eccvconf – at Milano, Lombardia
"Do text-free diffusion models learn discriminative visual representations?" tl;dr: Yes, diffusion models learn highly useful, discriminative visual representations even w/o text! #DiffSSL website:mgwillia.github.io/diffssl arxiv:arxiv.org/abs/2311.17921 @zhoutianyi @abhi2610🧵
3
1
5
1,694
Archana Swaminathan retweeted
I will be speaking about my work in INRs (namithap10.github.io/xinc/) at the INRV workshop at #CVPR2024. Come see me in Summit 335-336 at 5PM on Tuesday!
1
6
14
2,079
Archana Swaminathan retweeted
🍎 Imagine chopping an apple: each slice, a new shape, a new stats. Now, imagine teaching a computer to recognize each slice & even predict the next! 🤖 How can AI understand the subtle changes from a whole apple to a sliced one, and apply this knowledge universally? 🤔
Researchers @umiacs are teaching computers how to recognize and imagine fruits and vegetables in various forms—even as they’re being peeled, sliced or chopped into pieces. Their work to advance generative AI will be presented at #ICCV2023. Learn more: go.umd.edu/3PXbrst
1
8
2
19
2,616
Archana Swaminathan retweeted
Compositional Generalization is going to be the next big milestone in #GenAI and we need better datasets and benchmarks to make progress. Checkout our work Chop-N-Learn at #ICCV2023! chopnlearn.github.io
Researchers @umiacs are teaching computers how to recognize and imagine fruits and vegetables in various forms—even as they’re being peeled, sliced or chopped into pieces. Their work to advance generative AI will be presented at #ICCV2023. Learn more: go.umd.edu/3PXbrst
2
6
1
17
4,301
Archana Swaminathan retweeted
Dubbing feels unnatural due to unsynced lips & audio. Presenting "Diff2Lip: Audio Conditioned Diffusion Models for Lip-Synchronization" tl;dr: #Diff2Lip:arbirary speech+face video→hq lip-sync website:soumik-kanad.github.io/diff2… arxiv:arxiv.org/abs/2308.09716 @_sakshams_ @abhi2610🧵
1
3
2
13
2,460
Archana Swaminathan retweeted
Have you ever wondered if generative models like #diffusion models inherently can solve discriminative tasks? Presenting our work: "Diffusion Models Beat GANs on Image Classification" arxiv: arxiv.org/abs/2307.08702 website: mgwillia.github.io/diff-ssl hf: huggingface.co/papers/2307.0…🧵
Diffusion Models Beat GANs on Image Classification paper page: huggingface.co/papers/2307.0… While many unsupervised learning models focus on one family of tasks, either generative or discriminative, we explore the possibility of a unified representation learner: a model which uses a single pre-training stage to address both families of tasks simultaneously. We identify diffusion models as a prime candidate. Diffusion models have risen to prominence as a state-of-the-art method for image generation, denoising, inpainting, super-resolution, manipulation, etc. Such models involve training a U-Net to iteratively predict and remove noise, and the resulting model can synthesize high fidelity, diverse, novel images. The U-Net architecture, as a convolution-based architecture, generates a diverse set of feature representations in the form of intermediate feature maps. We present our findings that these embeddings are useful beyond the noise prediction task, as they contain discriminative information and can also be leveraged for classification. We explore optimal methods for extracting and using these embeddings for classification tasks, demonstrating promising results on the ImageNet classification task. We find that with careful feature selection and pooling, diffusion models outperform comparable generative-discriminative methods such as BigBiGAN for classification tasks. We investigate diffusion models in the transfer learning regime, examining their performance on several fine-grained visual classification datasets. We compare these embeddings to those generated by competing architectures and pre-trainings for classification tasks.
1
6
1
23
3,438
Paper Alert!! 📜🧨 #ACL2023NLP #NLProc FORK: A Bite-Sized Test Set for Probing Culinary Cultural Biases in Commonsense Reasoning Models. Work done with @rachelrudinger at @ClipUmd Details in 🧵(1/n) :
2
10
1
30
8,177
i change my Apple Watch face everyday to match the colour of my outfit and it feels nice 💅
5
two years in deep learning research and I still prefer to use print statements to debug instead of pdb 🥲
1
1
4
three dosas take 15 min to make but by the time you’ve sat down to eat, 2 are done and you finish the third in 30 seconds
10
it’s Friday and my code isn’t working. Rip weekend
1
3
update: it’s 2:45pm and my code is working y’all, time to sleep
5
had an incredibly well-rested night and downed a big cup of coffee to start my work with, now I’m jumpy and anxious. Lol
1
7
Archana Swaminathan retweeted
The internet is breaking today from viral claims that the new Pfizer data dump of 80k pages revealed real vaccine efficacy of the Pfizer/BioNTech vaccine in the phase 3 trial was in fact 12%, not 95%. In this 🧵I investigate these claims.
🚨🚨 Pfizer data released today. 80,000 pages. Pfizer knew vaccine harmed the fetus in pregnant women, and that the vaccine was not 95% effective, Pfizer data shows it having a 12% efficacy rate.🚨🚨
224
999
446
2,591