Stefanie Jegelka

Visiting Professor@Massachusetts Institute of Technology·美国·OpenReview

研究方向

learnable algorithms · neural parameter symmetries · invariance · equivariance · self-supervised representation learning · optimal transport · Graph neural networks · discrete optimization in ML · submodularity · determinantal point processes

Stefanie Jegelka

Any-Subgroup Equivariant Networks via Symmetry Breaking

When More is Less: Understanding Chain-of-Thought Length in LLMs

Sequential Parallel Duality in Prefix Scannable Models

CSRv2: Unlocking Ultra-Sparse Embeddings

SSL4RL: Revisiting Self-supervised Learning as Intrinsic Reward for Visual-Language Reasoning

Scaling Attention via Feature Sparsity

Learning Equivariant Models by Discovering Symmetries with Learnable Augmentations

Symmetries in Weight Space Learning: To Retain or Remove?

Better Together: Leveraging Unpaired Multimodal Data for Stronger Unimodal Models

Bootstrapping Language and Numerical Feedback for Reinforcement Learning in LLMs

Route Experts by Sequence, Not by Token

Fairness Aware Reward Optimization

Learning Linear Attention in Polynomial Time

Higher-Order Graphon Neural Networks: Approximation and Cut Distance

$G1$: Teaching LLMs to Reason on Graphs with Reinforcement Learning

Learning Diffusion Models with Flexible Representation Guidance

Beyond Matryoshka: Revisiting Sparse Coding for Adaptive Representation

Generalization, Expressivity, and Universality of Graph Neural Networks on Attributed Graphs

Learning with Exact Invariances in Polynomial Time

What is Wrong with Perplexity for Long-context Language Modeling?

On the Emergence of Position Bias in Transformers

Learning with Exact Invariances in Polynomial Time

Geometric Algorithms for Neural Combinatorial Optimization with Constraints

An Information Criterion for Controlled Disentanglement of Multimodal Data

Generalization Bounds for Canonicalization: A Comparative Study with Group Averaging

Beyond Interpretability: The Gains of Feature Monosemanticity on Model Robustness

On the Role of Depth and Looping for In-Context Learning with Task Diversity

Learning on LoRAs: GL-Equivariant Processing of Low-Rank Weight Spaces for Large Finetuned Models

Learning Efficient Positional Encodings with Graph Neural Networks

Are Synthetic Classifiers Really as Good as Real Classifiers?

Computing Optimal Regularizers for Online Linear Optimization