iiinova — where your digital self keeps evolving

Silpa Vadakkeeveetil Sreelatha @silpavs.bsky.social · 02/07/2026

Work done with Dan Wang, @serge.belongie.com, Muhammad Awais, @anjandutta.bsky.social. If you're into interpretability of vision models, I would love to connect at the conference. See you in Seoul!

010

Silpa Vadakkeeveetil Sreelatha @silpavs.bsky.social · 02/07/2026

RAIGen is a label-free framework that looks inside diffusion models to identify rare but meaningful visual attributes using only the model's internals via Matryoshka SAEs and a novel minority metric. A new direction for auditing T2I models 🔗 vssilpa.github.io/RAIGen_webpa...

vssilpa.github.io

RAIGen: Rare Attribute Identification in Text-to-Image Generative Models

110

Silpa Vadakkeeveetil Sreelatha @silpavs.bsky.social · 02/07/2026

Presenting RAIGen at #ICML2026 in Seoul! Can we automatically surface rare, underrepresented attributes in text-to-image diffusion models, with no prior knowledge of what they are? Come meet me in poster session! 📅 Wed Jul 8, 10:30–12:15 KST 📍 Hall A, #3308

121

Reposted by Silpa Vadakkeeveetil Sreelatha

Adarsh Kappiyath @adarshk22.bsky.social · 31/01/2025

🚀 I’m thrilled to share that our paper, "Sebra: DeBiasing through Self-Guided Bias Ranking," has been accepted to ICLR 2025! 🎉

101

Reposted by Silpa Vadakkeeveetil Sreelatha

ELLIS @ellis.eu · 15/01/2025

Meet Srishti Yadav, an ELLIS #PhD Student at 🇩🇰 Uni Copenhagen & 🇳🇱 Uni Amsterdam. Passionate about #AI & society, she explores culturally aware & inclusive AI models. Read her advice for young scientists & learn why women's visibility in AI/ML is crucial. #WomenInELLIS

0213

Silpa Vadakkeeveetil Sreelatha @silpavs.bsky.social · 11/12/2024

In this paper, we aim for responsible (fair and safe) text-to-image generation in diffusion models. Join me to discuss how we can improve generative models for better fairness and safety! (5/5)

000

Silpa Vadakkeeveetil Sreelatha @silpavs.bsky.social · 11/12/2024

(2) Concept Denoising Score Matching for Responsible Text-to-Image Generation 🗓️ Sun, Dec 15 | 🕗 8:15 a.m. PST | 📍 Safe Generative AI Workshop Authors : Silpa Vadakkeeveetil Sreelatha, Sauradip Nag, @serge.belongie.com , Muhammad Awais, and Anjan Dutta (4/5)

111

Silpa Vadakkeeveetil Sreelatha @silpavs.bsky.social · 11/12/2024

We introduce DeNetDM, a debiasing method that improves robustness to spurious correlations. Our method requires no bias annotations or explicit data augmentation while performing on par with approaches that require either or both. More details can be found here : arxiv.org/abs/2403.19863 (3/5)

100

Silpa Vadakkeeveetil Sreelatha @silpavs.bsky.social · 11/12/2024

(1) DeNetDM: Debiasing by Network Depth Modulation 🗓️ Thu, Dec 12 | 🕚 11 a.m. – 2 p.m. PST | 📍 East Exhibit Hall A-C #4309 Authors : Silpa Vadakkeeveetil Sreelatha *, Adarsh Kappiyath*, Abhra Chaudhuri, Anjan Dutta — * equal contribution (2/5)

121

Silpa Vadakkeeveetil Sreelatha @silpavs.bsky.social · 11/12/2024

🚀 Excited to be at @neuripsconf.bsky.social in Vancouver! I'll be presenting my work on AI fairness & interpretability. Here's where you can find me (1/5)

110

Reposted by Silpa Vadakkeeveetil Sreelatha

Belongie Lab @belongielab.org · 03/12/2024

With @neuripsconf.bsky.social right around the corner, we’re excited to be presenting our work soon! Here’s an overview (1/5)

1166

Silpa Vadakkeeveetil Sreelatha @silpavs.bsky.social · 20/11/2024

Could you please add me? I am an ELLIS PhD student.

110