Eric Wong

Assistant Professor@University of Pennsylvania·OpenReview

研究方向

deep learning · adversarial examples · optimal transport · wasserstein distance · sinkhorn iteration · convex optimization · newton method · admm · perturbation sets · perturbation learning · variational autoencoders · generative modeling · explainability · faithfulness · feature attribution

Eric Wong

SuperActivators: Transformers Concentrate Concept Signals in Just a Handful of Tokens

Missingness Bias Calibration in Feature Attribution Explanations

Benchmarking Mitigations For Covert Misuse

BrowserArena: Evaluating LLM Agents on Real-World Web Navigation Tasks

Instruction Following by Boosting Attention of Large Language Models

Probabilistic Stability Guarantees for Feature Attributions

DOLPHIN: A Programmable Framework for Scalable Neurosymbolic Learning

CTSketch: Compositional Tensor Sketching for Scalable Neurosymbolic Learning

Once Upon an Input: Reasoning via Per-Instance Program Synthesis

Logicbreaks: A Framework for Understanding Subversion of Rule-based Inference

Sum-of-Parts: Self-Attributing Neural Networks with End-to-End Learning of Feature Groups

Dolphin: A Programmable Framework for Scalable Neurosymbolic Learning

The FIX Benchmark: Extracting Features Interpretable to eXperts