sparsity
Sparsity refers to the condition when a matrix or vector contains a significant number of zero or near-zero values. In neural networks, promoting sparsity can lead to more efficient computations and reduced memory usage, often enhancing model interpretability.
- Adam Reduces a Unique Form of Sharpness: Theoretical Insights Near the Minimizer Manifold
- ChromFound: Towards A Universal Foundation Model for Single-Cell Chromatin Accessibiltiy Data
- DiCoFlex: Model-Agnostic Diverse Counterfactuals with Flexible Control
- Elastic ViTs from Pretrained Models without Retraining
- Exploring Neural Granger Causality with xLSTMs: Unveiling Temporal Dependencies in Complex Data
- FPSAttention: Training-Aware FP8 and Sparsity Co-Design for Fast Video Diffusion
- GraSS: Scalable Data Attribution with Gradient Sparsification and Sparse Projection
- In-Context Compositional Learning vis Sparse Coding Transformer
- Less is More: Unlocking Specialization of Time Series Foundation Models via Structured Pruning
- Making Classic GNNs Strong Baselines Across Varying Homophily: A Smoothness–Generalization Perspective
- On the Expressive Power of Mixture-of-Experts for Structured Complex Tasks
- Place Cells as Multi-Scale Position Embeddings: Random Walk Transition Kernels for Path Planning
- Rethinking Circuit Completeness in Language Models: AND, OR, and ADDER Gates
- SPACE: SPike-Aware Consistency Enhancement for Test-Time Adaptation in Spiking Neural Networks
- SPOT-Trip: Dual-Preference Driven Out-of-Town Trip Recommendation
- SeerAttention: Self-distilled Attention Gating for Efficient Long-context Prefilling
- Sparse Autoencoders Learn Monosemantic Features in Vision-Language Models
- Stochastic Shortest Path with Sparse Adversarial Costs
- Structured Linear CDEs: Maximally Expressive and Parallel-in-Time Sequence Models
- Understanding and Enhancing Message Passing on Heterophilic Graphs via Compatibility Matrix
- Variational Task Vector Composition