SELECT & COMPRESS
Learning from less
Coreset selection and dataset distillation: finding or creating compact datasets that preserve what matters for learning.
A CURIOUS MIND. A DATA-CENTRIC LENS.
Hello, I’m
Arundhati.
PhD researcher at DFKI & RPTU
Kaiserslautern-Landau, Germany
I study how we can help models learn more from less data. My research brings together data selection, dataset distillation, and continual learning, with a growing focus on vision–language models.
Data-centric learning✿Continual learning✿Dataset distillation✿Vision + language
01 / RECENT NEWS
Little updates.
New beginnings.
02 / RESEARCH
Small datasets.
Big questions.
What does a model really need to learn?
I explore the role of training data: what we keep, what we compress, and how we keep learning.
REMEMBER & ADAPT
Learning over time
Continual learning: exploring how data selection can support learning from new experiences while retaining previous knowledge.
CONNECT & UNDERSTAND
Learning across modalities
A growing focus on vision–language models, and how data-efficient learning can connect images and text.
03 / PUBLICATIONS
Ideas, in print.
Papers on data-efficient learning,
diffusion models, and multimodal systems.
Just Leaf It: Accelerating Diffusion Classifiers with Hierarchical Class Pruning
A hierarchical approach to diffusion classification that prunes irrelevant label groups, reducing class evaluations and speeding up inference.
Presented at ICANN in Padua, Italy. Arundhati S. Shanbhag and Brian B. Moser contributed equally.
HyperCore: Coreset Selection Under Noise via Hypersphere Models
Adaptive coreset selection using class-specific hyperspheres to identify useful training samples in noisy datasets.
PRISM: Diversifying Dataset Distillation by Decoupling Architectural Priors
Dataset distillation with different teacher architectures for supervision and regularization, encouraging greater diversity in synthetic training data.
SubZeroCore: A Submodular Approach with Zero Training for Coreset Selection
A training-free approach to choosing representative data subsets by balancing coverage and local density.
Diffusion Models, Image Super-Resolution, and Everything: A Survey
A survey of diffusion-based image super-resolution, covering foundations, methods, open challenges, and future research directions.
Audio-Visual Environmental Cues in Human-Robot Interaction
Exploring audio-visual environmental information for human–robot interaction.
04 / THE PERSON BEHIND THE PAPERS
Rooted in research.
Curious about life.
My path has taken me through recommendation systems, diffusion models, and multimodal robotics.
Outside research, I enjoy cooking, reading, gardening, crocheting, and travelling to experience different cultures.
GOOD IDEAS START WITH A CONVERSATION
Let’s grow something together.
Questions about my research, shared interests, or a potential collaboration?
I’d love to hear from you.