Research map / Label-efficient and robust learning / Recognition and long-tail
Unsupervised Transfer Learning for Spatiotemporal Predictive Networks
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 |
|---|---|---|---|
| Transfer memory from both 1-digit and 2-digit sources instead of training from scratch | ConvLSTM on 3-digit Moving MNIST · 3-digit Moving MNIST | MSE: 120.6 -> 94.7 | Helped |
| Use transferable memory rather than finetuning from the 2-digit source | ConvLSTM on 3-digit Moving MNIST · 3-digit Moving MNIST | MSE: 110.0 -> 94.7 | Helped |
| Apply memory transfer instead of finetuning | PredRNN on 3-digit Moving MNIST · 3-digit Moving MNIST | MSE: 89.4 -> 84.9 | Helped |
| Apply memory transfer instead of finetuning | MIM on 3-digit Moving MNIST · 3-digit Moving MNIST | MSE: 83.2 -> 75.3 | Helped |
| Transfer memory from KTH and Weizmann rather than training from scratch | ConvLSTM on Human3.6M · Human3.6M | MSE: 504.2 -> 442.5 | Helped |
| Transfer memory from KTH and Weizmann rather than finetuning from KTH | ConvLSTM on Human3.6M · Human3.6M | MSE: 472.0 -> 442.5 | Helped |
| Transfer memory from Guangzhou and content-unrelated sources rather than training from scratch | ConvLSTM on Beijing radar echo · Beijing radar echo | CSI: 0.348 -> 0.408 | Helped |
| Add seemingly unrelated 2-digit MNIST and KTH source memories to Guangzhou sources | TMU ConvLSTM on Beijing radar echo · Beijing radar echo | CSI: 0.403 -> 0.408 | Helped |
| Use Human3.6M as the source rather than training from scratch | ConvLSTM on KTH · KTH | SSIM: 0.771 -> 0.774 | No clear effect |
| Use 2-digit MNIST rather than Human3.6M as the source | ConvLSTM on KTH · KTH | SSIM: 0.771 -> 0.808 | Helped |
| Use 1-digit source instead of both digit sources | TMU ConvLSTM on 3-digit Moving MNIST · 3-digit Moving MNIST | MSE: 96.1 -> 94.7 | Hurt |
| Use 2-digit source instead of both digit sources | TMU ConvLSTM on 3-digit Moving MNIST · 3-digit Moving MNIST | MSE: 97.3 -> 94.7 | Hurt |
Closest papers by meaning
- UniSTD: Towards Unified Spatio-Temporal Learning across Diverse Disciplines (CVPR 2025 · similarity 0.87)
- Wavelet-Driven Spatiotemporal Predictive Learning: Bridging Frequency and Time Variations (AAAI 2024 · similarity 0.86)
- Drawing Informative Gradients from Sources: A One-stage Transfer Learning Framework for Cross-city Spatiotemporal Forecasting (AAAI 2025 · similarity 0.86)
- SCENT: Robust Spatiotemporal Learning for Continuous Scientific Data via Scalable Conditioned Neural Fields (ICML 2025 · similarity 0.85)
- Understanding the Transferability of Representations via Task-Relatedness (NeurIPS 2024 · similarity 0.84)
- Efficient Transfer Learning for Video-language Foundation Models (CVPR 2025 · similarity 0.84)
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.