Scholar
Prasanna Mayilvahanan
Google Scholar ID: 3xq1YcYAAAAJ
Max Planck Institute for Intelligent Systems & University of Tübingen
Citations & Impact
Citations
148
H-index
5
i10-index
4
Publications
6
Co-authors
0
Publications
Resume (English only)
Academic Achievements
- MATH-Beyond: A Benchmark for RL to Expand Beyond the Base Model (arXiv 2025): Introduces MATH-B to distinguish whether post-training methods sharpen existing reasoning or discover novel solution paths
- LLMs on the Line: Data Determines Loss-to-Loss Scaling Laws (ICML 2025): Shows identical downstream performance across setups when trained on same data with same loss
- In Search of Forgotten Domain Generalization (ICLR 2025, Spotlight): Reveals CLIP’s high performance on style shifts is due to presence of such images in training data
- Does CLIP's Generalization Performance Mainly Stem from High Train-Test Similarity? (ICLR 2024): Demonstrates CLIP’s OOD generalization is not primarily due to train-test duplicates
- Compositional Generalization from First Principles (NeurIPS 2023): Introduces a theoretical framework for analyzing compositional generalization in neural networks
- Representation Learning for the Clustering of Multi-Omics Data (IEEE/ACM TCBB 2022): Proposes a neural network method for multi-omics data integration and clustering
Co-authors