data distribution
Data distribution describes how data is dispersed across different classes or feature dimensions. Understanding the data distribution is crucial for training models effectively, as it affects how the model learns and generalizes to new samples.
- A Geometric Analysis of PCA
- A Gradient Guided Diffusion Framework for Chance Constrained Programming
- Benign Overfitting in Single-Head Attention
- Beyond the Seen: Bounded Distribution Estimation for Open-Vocabulary Learning
- Compositional Discrete Latent Code for High Fidelity, Productive Diffusion Models
- Conservative classifiers do consistently well with improving agents: characterizing statistical and online learning
- Consistent Sampling and Simulation: Molecular Dynamics with Energy-Based Diffusion Models
- Contribution of task-irrelevant stimuli to drift of neural representations
- Discretization-free Multicalibration through Loss Minimization over Tree Ensembles
- Emergence of Linear Truth Encodings in Language Models
- Estimating Model Performance Under Covariate Shift Without Labels
- Fair Deepfake Detectors Can Generalize
- Fast-Slow Thinking GRPO for Large Vision-Language Model Reasoning
- Information-Theoretic Reward Decomposition for Generalizable RLHF
- LLM-DAMVC: A Large Language Model Assisted Dynamic Agent for Multi-View Clustering
- Low Precision Streaming PCA
- Neural Entropy
- On the creation of narrow AI: hierarchy and nonlocality of neural network skills
- Performative Validity of Recourse Explanations
- Principled Fine-tuning of LLMs from User-Edits: A Medley of Preference, Supervision, and Reward
- The Good, the Bad and the Ugly: Meta-Analysis of Watermarks, Transferable Attacks and Adversarial Defenses
- Tight Lower Bounds and Improved Convergence in Performative Prediction
- Tighter CMI-Based Generalization Bounds via Stochastic Projection and Quantization
- Towards Large-Scale In-Context Reinforcement Learning by Meta-Training in Randomized Worlds
- Training-Free Safe Denoisers for Safe Use of Diffusion Models
- Understanding Representation Dynamics of Diffusion Models via Low-Dimensional Modeling
- un$^2$CLIP: Improving CLIP's Visual Detail Capturing Ability via Inverting unCLIP