PatchTST
PulseAugur coverage of PatchTST — every cluster mentioning PatchTST across labs, papers, and developer communities, ranked by signal.
6 day(s) with sentiment data
-
New MoE framework enhances time series forecasting with integrated expert losses
Researchers have developed a new Mixture-of-Experts (MoE) framework for time series forecasting that improves training efficiency and predictive performance. This framework integrates expert-specific losses directly int…
-
Foundation models show mixed results for pedestrian crowd forecasting
A new study published on arXiv evaluates the effectiveness of time-series foundation models (FMs) for pedestrian crowd count forecasting. The research compares seven different forecasting approaches, including tradition…
-
New JEPA model tackles cross-machine industrial AI transfer
Researchers have developed a Schema-Adaptive Action-Conditioned JEPA (SAAC-JEPA) designed for transferring industrial world models between different CNC machines. This model addresses challenges like differing dynamics,…
-
Foundation models require fine-tuning for diabetes glucose forecasting
A new study published on arXiv evaluates the effectiveness of time-series foundation models for continuous glucose monitoring (CGM) forecasting, particularly for individuals with type 1 and type 2 diabetes. The research…
-
New augmentation technique boosts multivariate forecasting models
Researchers have developed a novel time-domain augmentation technique for multivariate forecasting models. This method, called Sliding-Window Reordering with Overlap Averaging, involves unfolding sequences into overlapp…
-
New research details online adaptation for edge time-series forecasting
A new research paper published on arXiv explores the effectiveness of online adaptation techniques for time-series forecasting on edge devices. The study highlights how evaluation methodologies, such as warmup budgets a…
-
Neuro-symbolic framework predicts academic risk with interpretable AI
Researchers have developed EduRiskX, a novel neuro-symbolic framework designed to predict academic risk in online education. This system combines a Transformer-based neural network for analyzing student activity sequenc…
-
New benchmark improves AI-driven anomaly detection for research networks
Researchers have developed a new forecasting framework to improve anomaly detection in research networks, which often struggle with distinguishing legitimate high-traffic scientific bursts from malicious attacks. Using …
-
New ESN techniques boost time-series forecasting efficiency and accuracy
Researchers have developed new methods for improving Echo State Networks (ESNs), a type of reservoir computing model efficient for time-series forecasting. One approach, Dynamical Mode Pruning (DMP), refines ESNs by ran…
-
LithoFormer uses transformers for robust geological stratigraphic inference
Researchers have developed LithoFormer, a novel framework for stratigraphic inference in geological well log data. This system utilizes a Seq2Seq transformer model, specifically a PatchTST backbone with rotary positiona…
-
New research tackles explainability and adaptation in continual time series forecasting
Two new research papers explore the challenges and solutions for continual learning in time series forecasting models. The first paper introduces an attention-based experience replay framework to help models adapt to ch…
-
New AI frameworks enhance blood glucose forecasting for diabetes management
Two new research frameworks, GlucoTune and GlyRAG, aim to improve blood glucose forecasting for diabetes management. GlucoTune standardizes preprocessing and evaluation pipelines for reproducible experiments with time-s…
-
Simple models outperform LLMs in time series forecasting
A recent analysis highlights the significant challenges in time series forecasting, revealing that simple statistical models and zero-shot foundation models often outperform complex neural networks and even large langua…
-
New principle establishes stability threshold for residual neural network architectures
Researchers have introduced the 'sublinear-growth principle' for deep residual architectures, establishing a sharp stability threshold for the velocity field's input-magnitude exponent. This principle, supported by ODE …
-
New framework assesses AI forecasting model robustness against weather prediction errors
A new framework for evaluating the robustness of AI forecasting models in photovoltaic (PV) power generation has been developed. This framework addresses the challenge of numerical weather prediction (NWP) errors, which…
-
Deep learning models show resilience to weather forecast errors in PV power prediction
A new study evaluates the robustness of various deep learning models, including PatchTST, GRU, N-HITS, and LightGBM, when subjected to errors in numerical weather prediction (NWP) data. The research introduces a physica…
-
SEER framework tackles noisy, missing, and shifted time series data
Researchers have introduced SEER, a Transformer-based framework designed to enhance time series forecasting robustness. SEER addresses common data quality issues such as noise, anomalies, missing values, and distributio…
-
New research explores adaptive deployment for financial volatility forecasting models
A new research paper explores the impact of deployment strategies on the performance of multi-horizon volatility forecasting models in finance. The study demonstrates that different inference-time rollout rules can sign…
-
Deep learning models underperform simpler AI in stock market analysis
A recent research project compared three distinct eras of quantitative finance strategies—rule-based, classical machine learning, and deep learning—using 10 years of Apple stock data. Surprisingly, the most complex deep…
-
New Temporal Operator Attention framework enhances time-series analysis
Researchers have introduced Temporal Operator Attention (TOA), a novel framework designed to improve time-series analysis by addressing limitations in standard attention mechanisms. TOA explicitly incorporates learnable…