Research map / Time series / Foundation models for time series
Universal Redundancies in Time Series Foundation Models
Anthony Bao, Venkata Hasith Vattikuti, Jeffrey B. Lai, William Gilpin
Read the paperCheck your idea against 137,217 papers
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 |
|---|---|---|---|
| zero-ablate 28% of heads (middle layers, stable-rank ordering) and 2 MLPs (ablation) | TimesFM 2.5 · GIFT-Eval | % change in MASE: +6.12% MASE, +5.86% CRPS | No clear effect |
| zero-ablate 24% of heads and 2 MLPs (ablation) | Chronos 2 · GIFT-Eval | % change in MASE: +4.71% MASE, +3.92% CRPS | No clear effect |
| ablate 29% of heads in Toto, then also one MLP (ablation) | Toto · GIFT-Eval | % change in MASE: 3.54% -> 8.54% | Hurt |
| increase heads ablated in Toto from 29% to 43% (ablation) | Toto · GIFT-Eval | % change in MASE: 3.54% -> 10.10% | Hurt |
| increase heads ablated in Chronos Bolt from 28% to 42% (with 6 MLPs) (ablation) | Chronos Bolt · GIFT-Eval | % change in CRPS: 0.19% -> 6.42% | Hurt |
| ablate whole middle layers instead of first or last layers (ablation) | Chronos, Chronos-Bolt, Chronos 2, TimesFM 2.5, Toto, Moirai 1.1 · GIFT-Eval | Spearman distance to original predictions | Improved |
| ablate MLPs in encoder-decoder vs decoder-only models (ablation) | Chronos Bolt vs Toto and TimesFM 2.5 · GIFT-Eval | MASE geometric mean | Mixed |
| order heads for ablation from high to low stable rank rather than low to high (ablation) | TimesFM 2.5, layers 7, 10, 13 · GIFT-Eval | MASE geometric mean | Mixed |
| ablate a single sharp (low-entropy) attention head (ablation) | Chronos · context parroting forecasts | forecast quality | Hurt |
| ablate diffuse (high-entropy) heads, 32% of heads (ablation) | Chronos · context parroting forecasts | forecast quality | No clear effect |
| ablate heads by intrinsic stable rank instead of random head selection (ablation) | TSFMs · GIFT-Eval | forecast seasonality structure | Improved |
Closest papers by meaning
- Exploring Representations and Interventions in Time Series Foundation Models (ICML 2025 · similarity 0.92)
- Towards Neural Scaling Laws for Time Series Foundation Models (ICLR 2025 · similarity 0.90)
- Less is More: Unlocking Specialization of Time Series Foundation Models via Structured Pruning (NeurIPS 2025 · similarity 0.90)
- Multi-Scale Finetuning for Encoder-based Time Series Foundation Models (NeurIPS 2025 · similarity 0.89)
- MOMENT: A Family of Open Time-series Foundation Models (ICML 2024 · similarity 0.88)
- Learning the Context of Errors: Black-Box Online Adaptation of Time Series Foundation Models (NeurIPS 2026 · similarity 0.88)
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.
