Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Meta-Sketch: A Neural Data Structure for Estimating Item Frequencies of Data Streams
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use basic meta-training on synthetic Zipf tasks instead of conventional basic sketches | Basic Meta-Sketch · Real-stream tests, Wq and It | ARE: 12.30 vs 32.94 (Wq, n=5K); 1.13 vs 24.49 (It, n=40K) | Helped |
| Adapt the pretrained meta-sketch using sampled real-stream tasks | Advanced Meta-Sketch · Wq, Tr, n=40K | ARE: 5.20 vs 104.54 | Helped |
| Adapt the pretrained meta-sketch using sampled real-stream tasks | Advanced Meta-Sketch · It, Tr, n=40K | ARE: 2.29 vs 16.71 | Helped |
| Adapt during changing item-frequency correspondences rather than rely on a fixed learned classifier | Advanced Meta-Sketch vs. learned augmented sketches · Wq, n=5K, B=9KB, increasing shuffle ratio | AAE: 21.28 and 21.68; learned sketch AAEs start above 37 | Helped |
| Replace the learned addressing matrix with randomly generated unit vectors (ablation) | Meta-Sketch · Wq, n=5K, B=9KB | AAE | Hurt |
| Replace the learned addressing matrix with K-means cluster-center vectors (ablation) | Meta-Sketch · Wq, n=5K, B=9KB | AAE | No clear effect |
| Train the decoding module on a fixed-size, fixed-mean task instead of freezing its parameters (ablation) | Advanced Meta-Sketch · Wq, evaluation tasks with varying item size | AAE | Hurt |
| Train the decoding module on a fixed-size, fixed-mean task instead of freezing its parameters (ablation) | Advanced Meta-Sketch · Wq, evaluation tasks with varying frequency mean | ARE | Hurt |
| Use sparse addressing instead of dense SoftMax addressing | Meta-Sketch · Storage access and back-propagation | storage access overhead | Helped |
Closest papers by meaning
- Lego Sketch: A Scalable Memory-augmented Neural Network for Sketching Data Streams (ICML 2025 · similarity 0.88)
- Efficient Sketches for Training Data Attribution and Studying the Loss Landscape (NeurIPS 2024 · similarity 0.84)
- SketchINR: A First Look into Sketches as Implicit Neural Representations (CVPR 2024 · similarity 0.83)
- Sketch Down the FLOPs: Towards Efficient Networks for Human Sketch (CVPR 2025 · similarity 0.82)
- CTSketch: Compositional Tensor Sketching for Scalable Neurosymbolic Learning (NeurIPS 2025 · similarity 0.82)
- Learning-Augmented Frequent Directions (ICLR 2025 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.