Time Series Forecasting
PulseAugur coverage of Time Series Forecasting — every cluster mentioning Time Series Forecasting across labs, papers, and developer communities, ranked by signal.
7 day(s) with sentiment data
Time Series Forecasting models will increasingly incorporate RAG for improved accuracy
Recent research highlights two new papers (SERAF and Cross-RAG) that leverage Retrieval-Augmented Generation (RAG) to enhance time series forecasting accuracy. This suggests a growing trend towards integrating RAG techniques, which combine external knowledge with forecasting models, to improve performance across various datasets and forecasting tasks.
Diffusion models and LLMs will be integrated for enhanced time series forecasting capabilities
The development of Diffusion-LLM, which combines diffusion models with LLMs, addresses limitations in handling multimodal data and improves probabilistic modeling for time series forecasting. Its demonstrated success in ultra-long-term and few-shot forecasting suggests this hybrid approach will see further adoption for robust and generalized forecasting.
Metalearning and abstention strategies will become more prevalent in time series forecasting
A new metalearning framework enables selective time series forecasting by allowing models to abstain from making predictions on difficult data points. This approach, which uses scale-invariant statistics for transferability, indicates a potential shift towards more nuanced forecasting methods that recognize model limitations and improve overall reliability.
-
Deep learning performance distributions reveal tail risk for Trustworthy AI
A new study published on arXiv explores the non-deterministic nature of deep learning algorithms, revealing that model performance is best understood as a distribution rather than a single value. The research conducted …
-
New TempShap framework offers detailed feature attribution for time-series forecasting
Researchers have introduced TempShap, a novel framework for explaining time-series forecasting models. Unlike existing methods that treat features as single entities, TempShap analyzes individual input cells across the …
-
New DBCM module enhances time series forecasting adaptation
Researchers have developed a new Dual-Branch Calibration Module (DBCM) designed to improve test-time adaptation for time series forecasting. This module addresses the issue of forecast residuals caused by non-stationari…
-
New research refines time series forecasting with agentic orchestration and robust correction
Three new research papers explore advancements in time series forecasting. One paper introduces FairAMO, a protocol for evaluating agentic model orchestration, finding that while orchestration improves accuracy, its gai…
-
AI research advances in materials science, time-series forecasting, and neural network training
Researchers are developing new methods for analyzing complex scientific data using AI. One paper introduces a compositional neural network framework that combines X-ray and neutron scattering data for materials science,…
-
New Error Feedback Calibration framework improves time series forecasting
Researchers have developed a new framework called Error Feedback Calibration (EFC) to improve time series forecasting models. EFC uses past forecasting errors to causally correct future predictions by constructing a Cau…
-
Nixtla releases TimeGPT for AI-powered time series forecasting
Nixtla has released TimeGPT, a tool designed for time series forecasting and anomaly detection. The tool aims to transform time series data into actionable insights. The release includes code examples and visualizations…
-
New SCORE framework efficiently approximates covariance matrices for Gaussian targets
Researchers have developed SCORE, a novel framework for approximating covariance matrices in high-dimensional Gaussian targets for neural network-based predictive modeling. This method decomposes the learning task into …
-
New MARO model prioritizes recent data for advanced time series forecasting
Researchers have introduced MARO, a novel model for time series forecasting that prioritizes recent data. Unlike traditional models that apply uniform processing to all historical data, MARO processes the look-back wind…
-
New paper calls for scenario-grounded stress testing for time series forecasting models
A new paper published on arXiv argues that current benchmarks for time series forecasting (TSF) are insufficient for real-world deployment. The authors contend that existing stress tests, which often focus on Gaussian n…
-
New minimal architecture outperforms DS models, tackles regime changes
Researchers have developed a novel, minimal architecture called DynaBase for zero-shot reconstruction of dynamical systems. This architecture, featuring a single-parameter piecewise affine map and a context selector, ca…
-
New research on arXiv introduces advanced time series forecasting techniques · 9 sources tracked
Multiple research papers published on arXiv introduce novel methods for improving time series forecasting. SteerCast enhances decoder-only models by using a database of past data to guide predictions without retraining.…
-
New PIRNN model leverages historical physical data for improved time series forecasting
Researchers have developed a Physics Informed Recurrent Neural Network (PIRNN) that improves time series forecasting by incorporating physical knowledge from historical data. Unlike previous Physics Informed Neural Netw…
-
New method reduces redundant dependencies in Transformer time series forecasting
Researchers have developed a new strategy to improve Transformer-based time series forecasting models by reducing redundant token dependencies. This method jointly applies an attention entropy constraint and a predictio…
-
New framework treats time series forecasting as visual inpainting task
Researchers have introduced ICI-Time, a new framework that treats time series forecasting as a visual inpainting problem. This approach leverages the capabilities of large vision models by converting time series data in…
-
MetaCaster framework enables few-shot learning for lightweight time series forecasters
Researchers have introduced MetaCaster, a novel multi-agent framework designed to facilitate few-shot learning for lightweight time series forecasters. This system addresses the challenge of training effective forecaste…
-
ICI-Time framework reframes time series forecasting as visual inpainting
Researchers have developed ICI-Time, a new framework that treats time series forecasting as a visual inpainting problem. This approach utilizes large vision models by converting time series data into area charts, which …
-
Deep learning models benchmarked for smart meter energy forecasting
Researchers have conducted an empirical benchmark of nine deep learning models for smart meter energy forecasting, evaluating their performance on two public datasets. The study found that while extending historical inp…
-
New regularization technique combats suboptimal collapse in time series models
Researchers have introduced a new technique called Ground-Truth Neighborhood Regularization (GTN-R) to improve the performance of time series foundation models (TSFMs) when using reinforcement learning (RL) for post-tra…
-
New frameworks enhance time-series forecasting with retrieval and novel architectures · 4 sources tracked
Researchers have introduced three novel frameworks for time-series forecasting, each leveraging different techniques to improve accuracy and efficiency. TimePre integrates the speed of Multilayer Perceptrons with the di…