Ching Lam CHOI
Preprint 2026
Tabular Pancakes: Cryptographic Watermarks Against Dataset Fabrication
Watermarking tabular data generators by sampling latents from the keyed Pancake distribution (hCLWE) instead of Gaussian, for detecting fraudulent data fabrication.
Papers & projects
Research
Structure in data and models. Safety, robustness, and the representations that make learning work.
Google Scholar ↗8 papers
NeurIPS 2026Oral
Shaping Useful Noise: Energy Distributions Predict Visual Pretraining Quality
Synthetic data is useful when it occupies the right real-image likelihood regimes; ELD measures energy shape to predict, diagnose, curate visual pretraining data.
Preprint 2026
Fairness Aware Reward Optimization
Reducing LLM bias and toxicity by constraining the reward model to be independent of sensitive attributes, conditional on unrestricted features.
ICLR 2025Oral
UNI: Unlearning-based Neural Interpretations
Debiased path attribution by perturbing in the unlearning direction of steepest ascent.
Preprint 2023
On the Generalization of Gradient-based Neural Network Interpretations
Gradient-based XAI fails to generalise from train to test data: regularisation crucially helps.
Preprint 2022
Universal Adversarial Directions
UAD is an adversarial examples game with a pure Nash equilibrium and better transferability.
Preprint 2021
Self-distillation with Batch Knowledge Ensembling Improves ImageNet Classification
BAKE distils self-knowledge of a network, by propagating samples similarities in each batch.
CVPR 2021
DivCo: Diverse Conditional Image Synthesis via Contrastive Generative Adversarial Network
DivCo uses contrastive learning to reduce mode collapse in GANs.
No matching papers
Try a broader term, a different year, or clear the filters.