Rahul Vigneswaran

Professional overthinker. Avid sleeper. Applied Scientist Intern @ Amazon. Masters Student @ IIT Hyderabad.

[Rahul Vigneswaran @somethingmyname]

@cse_iith Paper Link: https://t.co/NMtFkdHjjM

[CSE IITH @cse_iith]

𝗣𝗮𝗽𝗲𝗿 𝗔𝗰𝗰𝗲𝗽𝘁𝗮𝗻𝗰𝗲 𝗔𝗻𝗻𝗼𝘂𝗻𝗰𝗲𝗺𝗲𝗻𝘁 🎉

Paper titled "Transitioning Heads Conundrum: The Hidden Bottleneck in Long-Tailed Class-Incremental Learning" has been accepted at TMLR 2026 (Transactions on Machine Learning Research).

Authors: Rahul Vigneswaran, Hari Chandana Kuchibhotla, Vineeth N Balasubramanian

👏 Congratulations to all the authors!

🔍 Key Highlight:
This work introduces DEREK (DEcoupling Representations for Early Knowledge Distillation), a method addressing a previously overlooked challenge in Long-Tailed Class-Incremental Learning (LTCIL): the Transitioning Heads Conundrum.

In LTCIL, head classes that are well-represented in earlier tasks become tail classes in subsequent tasks due to memory constraints, leading to accelerated catastrophic forgetting. DEREK mitigates this by decoupling head and tail learning via specialized expert networks and applying Early Knowledge Distillation before data constraints take effect, preserving rich representations.

Across 2 LTCIL benchmarks, 12 experimental settings, and 24 baselines, DEREK consistently establishes new state-of-the-art performance.

#MachineLearning #ContinualLearning #LongTailedLearning #KnowledgeDistillation #TMLR2026 #IITHyderabad

[CSE IITH @cse_iith]

We are pleased to share that the paper "HARE: Human-in-the-Loop Algorithmic Recourse" by Sai Srinivas Kancheti, Rahul Vigneswaran, Bamdev Mishra, and Prof. Vineeth N. Balasubramanian has been accepted at TMLR (Transactions on Machine Learning Research) 2025.

🔹 Paper Link: https://t.co/iBIR0DQcDJ
🔹 Code Repository: https://t.co/lubWG0zJ2D

This work introduces HARE (Human-in-the-loop Algorithmic Recourse), a method that actively incorporates human feedback to generate personalized, actionable, and realistic recourse strategies—moving beyond conventional counterfactual explanations.

Authors:
Kancheti Sai Srinivas (CSE, IIT Hyderabad)
Rahul Vigneswaran K (CSE, IIT Hyderabad)
Bamdev Mishra (Microsoft India)
Prof. Vineeth N Balasubramanian (CSE, IIT Hyderabad)

We congratulate the authors on this achievement and look forward to the impact of their work in the field of Fair AI, Human-in-the-loop Learning, and Actionable Machine Learning.

#IITHyderabad #CSEIITH #TMLR2025 #MachineLearning #AI #AlgorithmicRecourse #FairAI #HumanInTheLoop

[Rahul Vigneswaran @somethingmyname]

This is a collab with Sai Srinivas, Bamdev Mishra, Vineeth N Balasubramaniam

Code: https://t.co/wCPD1yczMo
Paper: https://t.co/PttJI6FHU7

5/5

@IITHyderabad @cse_iith @TmlrOrg @TmlrPub

[Rahul Vigneswaran @somethingmyname]

🚨 My paper got accepted into TMLR 25! 🚨

Are you a victim of your loan being rejected by an algorithm? Did the countermeasures provided by your bank seem unreasonable?

Ask your bank to adopt HARE! 🐇✨

Paper: https://t.co/PttJI6FHU7

1/5
@TmlrOrg @TmlrPub @cse_iith

[Rahul Vigneswaran @somethingmyname]

This has been in the works for quite some time, and I’m thrilled to finally share it! If you care about Fairness in AI, Human-in-the-loop learning, Actionable ML, this work is for you!

4/5

[Nicola Strisciuglio @NicStrisc]

Happy with the interest received by our work "What do neural networks learn in image classification? A frequency shortcut perspective" https://t.co/V3hmJvEU2t at @ICCVConference this morning. Looking forward to the next steps in this project @UTwente 📸with #ShunxinWang

[Stanislav Fort @stanislavfort]

🚨 New paper 🚨
A single adversarial attack can influence the classification of 100s of images simultaneously towards 100s of attacker-specified classes & I use this to learn about their geometry:

Multi-attacks: Many images + the same adversarial attack → many target labels

[Glen Berseth @GlenBerseth]

@UMontrealDIRO (my department) has 3 openings in machine learning. Strong candidates may also join @Mila_Quebec. More detailed are available here.
https://t.co/69TRiylOLE

[AK @_akhaliq]

DreamDistribution: Prompt Distribution Learning for Text-to-Image Diffusion Models

paper page: https://t.co/4MIQgkUYPS

DreamDistribution finds a prompt distribution of reference images, then it can be used to generates new 2D/3D instances, capable of text-guided editing and more

[AK @_akhaliq]

Parrot Captions Teach CLIP to Spot Text

paper page: https://t.co/G8LQGgqJ34

Despite CLIP being the foundation model in numerous vision-language applications, the CLIP suffers from a severe text spotting bias. Such bias causes CLIP models to `Parrot' the visual text embedded within images while disregarding the authentic visual semantics. We uncover that in the most popular image-text dataset LAION-2B, the captions also densely parrot (spell) the text embedded in images. Our analysis shows that around 50\% of images are embedded with visual text content, and 90\% of their captions more or less parrot the visual text. Based on such observation, we thoroughly inspect the different release d versions of CLIP models and verify that the visual text is the dominant factor in measuring the LAION-style image-text similarity for these models. To examine whether these parrot captions shape the text spotting bias, we train a series of CLIP models with LAION subsets curated by different parrot-caption-oriented criteria. We show that training with parrot captions easily shapes such bias but harms the expected visual-language representation learning in CLIP models. This suggests that it is urgent to revisit either the design of CLIP-like models or the existing image-text dataset curation pipeline built on CLIP score filtering.

[Litu Rout @litu_rout_]

Stable Diffusion based inverse problem solvers are slow. 🐌

Additionally, photoshop and other text-guided image editing solutions fall short when the input is noisy. 🤷‍♂️

Introducing STSL 🎯, the first efficient second-order Tweedie sampler to address these problems! – 🧵⬇️

[Tengda Han @TengdaHan]

Excited to advertise the "Self-Supervised Learning: Theory and Practice" workshop @NeurIPSConf 2023!
📍Room: 217 -- 219
⏰Saturday, Dec 16
🔊Speakers: @y_m_asano @dlarlus Alyosha Efros, Abhinav Gupta @chelseabfinn @xinyun_chen_
📄73 accepted papers, with 4 oral presentations

[Jia-Bin Huang @jbhuang0604]

How to get unstuck?

"It doesn't work." 🤷‍♀️🤷‍♂️

In most research projects, ~99% of the time your experiments DO NOT work. What should we do to get ourselves unstuck? 🤔

Sharing some tips I found useful. 🧵

[Rahul Vigneswaran @somethingmyname]

@onucharlesc @hugo_larochelle @MakarandTapaswi

[Vedant Shah @veds_12]

Unlearning via Sparse Representations

Work with @f_traeuble, Ashish Malik, @hugo_larochelle, @MichaelMozer1, @prfsanjeevarora, Yoshua Bengio & @anirudhg9119

https://t.co/6edmvYc8LB

TLDR: We propose a new approach enabling nearly compute-free class unlearning. 🧵

[Sagar Vaze @Sagar_Vaze]

"No Representation Rules Them All in Category Discovery"

Joint work with Andrea Vedaldi and Andrew Zisserman at @Oxford_VGG.

Project Page: https://t.co/ThHv3gUUxs
Code: https://t.co/VVolK9IVms
Paper: https://t.co/g4ppAlhXq7

[Rosanne Liu @savvyRL]

4 more days till #ICLR2024 Tiny Papers deadline! We have so far 30 submission, and 100 reviewers and 10 ACs recruited 😀

Authors: make sure to submit before Dec 8! It's just 2 pages ("tiny" papers), so hopefully even if you just learned it today, you'll still be able to make it!