Research map / Label-efficient and robust learning / Test-time adaptation
Temporal Generalization: A Reality Check
Divyam Madaan, Sumit Chopra, Kyunghyun Cho
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| model averaging of past checkpoints instead of the most recent model | recent model · temporal benchmarks (NewsRoom, Wilds-Time) | perplexity / accuracy | Hurt |
| downscale recent parameters by a scalar alpha < 1 | recent model · NewsRoom language modeling and summarization | perplexity | Improved |
| downscale recent parameters by a scalar alpha < 1 | recent model · Wilds-Time datasets | accuracy | Hurt |
| Taylor-series parameter extrapolation | recent model · temporal benchmarks | forward transfer | Hurt |
| initialize each time step from the previous one (continual learning) instead of independent training from pretrained (ablation) | interpolation and extrapolation methods · NewsRoom, T5-small | forward transfer | Improved |
| learn parameter offset for Taylor extrapolation | Taylor extrapolation · NewsRoom language modeling, 12 months | perplexity: 6.05 -> 6.09 | No clear effect |
| add learned time-varying coefficient to Taylor extrapolation | Taylor extrapolation · NewsRoom language modeling, 12 months | perplexity: 6.09 -> 6.07 | No clear effect |
| tune extrapolation factor alpha with future validation data (ablation) | Taylor extrapolation · NewsRoom summarization | optimal alpha | No clear effect |
Closest papers by meaning
- TEMPO: Prompt-based Generative Pre-trained Transformer for Time Series Forecasting (ICLR 2024 · similarity 0.84)
- LLM Processes: Numerical Predictive Distributions Conditioned on Natural Language (NeurIPS 2024 · similarity 0.84)
- TimePerceiver: An Encoder-Decoder Framework for Generalized Time-Series Forecasting (NeurIPS 2025 · similarity 0.84)
- Time-LLM: Time Series Forecasting by Reprogramming Large Language Models (ICLR 2024 · similarity 0.84)
- A decoder-only foundation model for time-series forecasting (ICML 2024 · similarity 0.84)
- Large Pre-trained time series models for cross-domain Time series analysis tasks (NeurIPS 2024 · similarity 0.83)
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.
