pareto frontier
A concept from multi-objective optimization representing a set of points where no improvement can be made on one objective without worsening another. In AI, it is used to evaluate trade-offs between competing objectives such as accuracy and efficiency.
- Automated Composition of Agents: A Knapsack Approach for Agentic Component Selection
- Better Estimation of the Kullback--Leibler Divergence Between Language Models
- Blockwise Flow Matching: Improving Flow Matching Models For Efficient High-Quality Generation
- Design-Based Bandits Under Network Interference: Trade-Off Between Regret and Statistical Inference
- Distillation Robustifies Unlearning
- Improving Monte Carlo Tree Search for Symbolic Regression
- Let the LLM Stick to Its Strengths: Learning to Route Economical LLM
- Mixture-of-Recursions: Learning Dynamic Recursive Depths for Adaptive Token-Level Computation
- Parameter Efficient Fine-tuning via Explained Variance Adaptation
- R2R: Efficiently Navigating Divergent Reasoning Paths with Small-Large Model Token Routing
- Remasking Discrete Diffusion Models with Inference-Time Scaling
- Scalable Evaluation and Neural Models for Compositional Generalization
- Sparse VideoGen2: Accelerate Video Generation with Sparse Attention via Semantic-Aware Permutation