TimeSeek pretrains one model for multi-task time series analytics
After large-scale pretraining, authors report state-of-the-art forecasting, anomaly detection, and classification in few-shot and zero-shot settings.
Stories per day · 30 days
3 RSS
Relatedtime-series13foundation-model4foundation-models4benchmarks3mistral3open-weights3timestral3flow-matching2llm-agents2pathfm2timesfm2activations1agent-evolution1anomaly-detection1benchmark1bots1coding-agents1conformal-prediction1
38 stories, newest first · one card per story
Stories tagged #forecasting
Paper measures sampling-frequency invariance of four time-series models
MOMENT, Chronos-2, Moirai-MoE and TimesFM are scored with new FSI, FDI and CDI metrics.
Interviews with 15 builders map forecasting-bot construction problems
Successful competition participants struggled to ground bots in facts, keep reasoning coherent, and judge real performance.
Conformal prediction improves wind forecasts
Regime-specific calibration fixes high-wind undercoverage while narrowing bands.
AI forecasts grain growth under non-isothermal conditions
Recurrent neural networks conditioned on thermal data predict grain growth with mean relative errors below 5%.
Context-Answer Metamodels Forecast LLM Activations Pre-Answer
Context activations can forecast mean-pooled answer activations with high fidelity and generalize zero-shot to out-of-domain datasets.
Random features approximate Koopman generators of continuous-time systems
A data-informed dictionary for gEDMD is reported faster than iterative optimization, with convergence proofs and a boundedness counterexample.
FACTS-MAS weights forecasts by each evidence channel's skill
On housing inventory and unemployment claims for 15 U.S. metros, it beats TimesFM, while factor-adjusted exponential smoothing matches the fused forecast.
Investor's Last Exam tests equity agents on frozen decision-time evidence
GPT-6 Astra leads at 68.1 of 100, but scores only 52.5 on Forecast & Model.
Timestral models lead GIFT-Eval, TIME and fev-bench
Four probabilistic forecasters from 5M to 209M parameters ship with Hugging Face weights and GitHub inference code.
Timestral releases four open time-series models, topping GIFT-Eval
Authors say the 209M and 123M variants rank first and second on every metric of GIFT-Eval, TIME, and fev-bench.
DISC combines time-series similarity metrics to predict transfer
After calibration, DISC scores pairs from metric values and tracks forecasting error better than any single metric.
NVIDIA introduces Kumo Forecast for time-series prediction
The post says it pairs pretrained time-series representations with historical pattern retrieval and interpretability.
Listing text predicts Polymarket volume but not later dispute risk well
On 802,863 resolved markets, log-volume R² is 0.747, while dispute ranking reaches AUC 0.78 at a 0.31% base rate.
DeepEvolve agents self-improve by forecasting live events
Authors say the forecaster led FutureX for three weeks, with rating gains up to 195%.
Mistral releases Timestral time-series foundation models
Mistral says its 209M model ranks first on GIFT-Eval, TIME and fev-bench while using about two-thirds the parameters of the prior leader.
Mistral releases Timestral, a 209M forecasting model family
The title says the family outperforms Google’s TimesFM-3; the article body was not in the provided text.
LagNet uses lag-structured operators for long-term forecasting
LagLayer ties history-to-future weights by relative lag and can replace temporal linear layers in existing models.
PathFM: Fast Generative Forecasting of Multivariate Trajectories
PathFM generates full forecast trajectories with fast flow-matching sampling and leads a new path-score benchmark.
PathFM: Fast Generative Forecasting of Multivariate Trajectories
PathFM uses flow matching to forecast full multivariate trajectories and beats baselines on a new path-score benchmark.
TimeRSI: Towards Self-Improved Agentic Deployment for Time-Series Foundation Models
TimeRSI uses coding agents to iteratively improve frozen time-series pipelines and leads 198 forecasting tasks.
Graph-Enhanced Disease Trajectory Discovery and Temporal Endpoint Forecasting from Longitudinal Electronic Health Records
A MIMIC-IV study finds graph branching motifs mainly improve long-range event timing, not endpoint discrimination.
Mistral releases Timestral-209M time-series forecasting model
The 209M checkpoint forecasts up to 1,024 steps from 32,768 history steps across 21 quantiles.
JockeyBench finds LLM racing agents lose under capped-Kelly rules
On 324 historical Japanese races, all 28 model-scaffold setups had negative mean profit, despite some revision gains.