31 articles tagged with #time-series-forecasting. AI-curated summaries with sentiment analysis and key takeaways from 50+ sources.
Researchers have released Toto 2.0, a family of five open-source time series forecasting models that demonstrate reliable improvements across a scaling range of 4M to 2.5B parameters. The models achieve state-of-the-art performance on three major benchmarks and represent a significant advance in applying foundation model scaling principles to forecasting tasks.
Researchers propose a novel framework combining importance-aware news compression and process reward models to improve LLM-based time series forecasting across finance, energy, and cryptocurrency markets. The method addresses practical limitations of existing approaches by intelligently filtering news articles within context windows and guiding iterative retrieval, achieving better accuracy with fewer refinement iterations.
FreqLite is a new lightweight linear model for long-term time-series forecasting that uses frequency decomposition and adaptive normalization to achieve better accuracy than larger transformer models while requiring 4x fewer parameters and significantly less computational resources. The method introduces Adaptive Reversible Instance Normalization (A-RevIN) to handle non-stationary data more effectively than existing approaches.
Researchers present VLBM, a machine learning framework designed to improve multivariate time series forecasting under out-of-distribution (OOD) conditions by separating stable patterns from anomalous deviations. The model demonstrates 15% average improvement over existing methods across real-world datasets, addressing a critical gap where standard forecasting fails during rare but high-impact events.
Researchers propose a new family of learnable Koopman operators that combine linear dynamical systems theory with deep learning for time series forecasting. The approach integrates with existing transformer architectures like Patchtst and Autoformer, offering improved stability and interpretability in predictive models.
Researchers develop a large language model framework for predicting vessel trajectories and destinations up to 30 days in advance using reinforcement learning with verifiable rewards. The approach outperforms traditional deep learning methods by maintaining route feasibility and destination accuracy over extended maritime forecasting horizons.
Researchers introduce FAME, a sparse mixture-of-experts framework that dynamically routes time series forecasting tasks to specialized models based on data characteristics. Tested on a production retail dataset with 5,000+ vending machines, the system achieves 12.4% MSE improvement over single-model baselines while using only 1.92 experts per series, demonstrating practical advantages for large-scale commercial forecasting systems.
Researchers present VFEM, a cross-modal forecasting model that combines pre-trained vision models with time series data to improve multivariate forecasting by capturing cross-channel dependencies. The approach transforms time series into visual representations and uses cross-modal attention fusion, achieving competitive performance while training only 7.45% of total parameters.
A new research paper challenges the effectiveness of adaptive patching in time-series Transformers, demonstrating that well-tuned uniform patching strategies often match or exceed the performance of dynamic approaches. The study provides theoretical and empirical evidence that adaptive patching requires specific conditions to outperform simpler baselines and questions whether the added complexity delivers meaningful forecasting improvements.
A research paper reveals a fundamental trade-off in multi-step time series forecasting: models optimized for mean squared error (MSE) produce unrealistic predictions under conditional uncertainty, failing to capture actual market variability. The study demonstrates that relaxing MSE constraints by just 5% can yield 17-30% improvements in forecast realism without sacrificing practical accuracy.
Researchers introduce Signed Dual Attention, a novel transformer attention mechanism that captures both positive and negative dependencies in time series data without requiring additional parameters. By using a dual message-passing approach inspired by correlation structures, this technique achieves greater expressiveness while maintaining parameter efficiency, potentially improving forecasting accuracy in applications requiring signed relational modeling.
Researchers introduce PC-MambaSDE, a machine learning framework designed to predict remaining useful life in industrial equipment by combining continuous-time neural networks with physics-based constraints. The model handles irregular sensor data and prevents physically impossible degradation patterns, outperforming existing methods especially when observation data is sparse.
Researchers present an LLM-agent framework that enhances time series forecasting by incorporating business context and expert judgment into statistical predictions. The system bridges the gap between raw forecasts and decision-ready outputs through structured reasoning, contextual evidence retrieval, and auditable revision mechanisms.
Researchers demonstrate that time series forecasting models require longer context windows not merely to capture long-range dependencies, but fundamentally to identify which generative process is producing the data. They prove that even for processes with memory length P, window sizes strictly larger than P are necessary to achieve minimum error, and propose decoupling generative process identification from conditional forecasting to improve computational efficiency.
Researchers introduce ODTQA-FoRe, a new dataset and TimeFore framework enabling large language models to perform future-oriented numerical predictions on tabular data using time-series forecasting. The innovation addresses a critical gap where existing LLM systems excel at historical analysis but struggle with predictive reasoning, demonstrated through real estate data scenarios.
Researchers introduce Unicorn, a universal correlation network that addresses a key limitation in time series forecasting by enabling models to scale across high-dimensional datasets while capturing inter-channel dependencies. The framework uses a latent prototype codebook to learn identity-agnostic patterns that transfer across diverse domains, significantly outperforming existing architectures in few-shot transfer scenarios.
Researchers introduce KairosAgent, an agentic framework combining large language models with time series foundation models to improve multimodal forecasting across domains. The system uses semantic reasoning from LLMs fused with numerical forecasting capabilities, achieving superior zero-shot performance through reinforcement learning and structured tool integration.
Researchers have developed a mathematical framework that preserves closed-form variational inference when composing multiple probabilistic models together, traditionally a challenge that breaks analytical tractability. By identifying five core factor-graph primitives and proving their composability, the work enables Bayesian mixture-of-experts models with inferred gating functions, demonstrated through improved ensemble forecasting with calibrated uncertainty.
Researchers introduce Dr-CiK, a benchmark for testing whether AI agents can independently retrieve relevant context from noisy document sources to improve time series forecasting. Evaluation reveals current information retrieval agents recover less than 5% of supporting evidence and are frequently misled by irrelevant information, highlighting a critical gap in foresight-driven AI development.
Researchers propose Under-Cali, a machine learning framework for forecasting irregular multivariate time series data in real-time online settings. The system uses uncertainty estimation and dual-expert calibration to maintain accuracy despite dynamic data distribution shifts, achieving improvements over existing methods with minimal computational overhead.
Researchers introduce a novel predictability-aligned evaluation framework for time series forecasting that separates model performance from data's inherent unpredictability. The framework reveals that complex AI models excel with difficult-to-predict data while linear models perform comparably on more predictable tasks, suggesting current benchmark rankings conflate model capability with task difficulty.
Falcon-X is a new time series foundation model that improves multivariate forecasting by mapping heterogeneous data types into a unified latent space rather than processing raw variables directly. The model uses novel attention mechanisms to capture both positive and negative relationships between variables, achieving state-of-the-art performance on forecasting benchmarks.
Researchers at the KATRIN experiment applied advanced deep learning models to predict source stability in tritium monitoring, identifying N-BEATS as the optimal forecasting algorithm. This application demonstrates how temporal learning models can optimize real-world physics experiments by improving measurement scheduling and maintenance planning through accurate long-horizon time-series predictions.
Researchers introduce MS-FLOW, a machine learning framework that improves multivariate time series forecasting by using sparse, selective connections between variables rather than dense interactions. The approach addresses the problem of spurious correlations that plague existing methods, achieving state-of-the-art accuracy on 12 benchmarks while identifying fewer but more reliable dependencies.
Researchers introduce STRIDE, a framework that integrates large language model reasoning into time series foundation models by projecting LLM reasoning into continuous embedding spaces rather than discrete tokens. The approach achieves state-of-the-art forecasting performance while providing interpretable reasoning, addressing the modality gap that previously limited combining LLMs with numerical time series data.
