weak supervision
Weak supervision is a training paradigm where models learn from noisy, limited, or imprecise labels instead of relying on fully labeled data. This approach rests on leveraging large volumes of unlabeled data or imperfectly labeled data to achieve reasonable performance.
- Balancing Positive and Negative Classification Error Rates in Positive-Unlabeled Learning
- Can Diffusion Models Disentangle? A Theoretical Perspective
- Counterfactual Image Editing with Disentangled Causal Latent Space
- Disentangling Latent Shifts of In-Context Learning with Weak Supervision
- FrameShield: Adversarially Robust Video Anomaly Detection
- RAPTR: Radar-based 3D Pose Estimation using Transformer
- SCAN: Self-Denoising Monte Carlo Annotation for Robust Process Reward Learning
- Weaver: Shrinking the Generation-Verification Gap by Scaling Compute for Verification