Technology Category
Machine Learning: Causal LearningReasoning under Uncertainty: CausalityCognitive Modeling & Cognitive Systems: Neural Spike Coding
Gradient Regularization-based Neural Granger Causality
🤖 AI Summary
Existing neural Granger causality methods require separate models for each pair of time series, incurring prohibitive computational overhead, and rely on weight sparsity regularization, limiting their capacity to capture complex dynamic interactions. To address these limitations, we propose GRNGC—a unified predictive model (compatible with MLPs, LSTMs, KANs, etc.) that directly quantifies causal influences between variables via L1-regularized input-output gradients, eliminating the need for weight sparsity assumptions while preserving both interpretability and representational power. Evaluated on benchmark datasets including DREAM and fMRI BOLD, GRNGC achieves substantial improvements in causal network reconstruction accuracy and reduces computational cost by an order of magnitude. GRNGC thus establishes a new paradigm for high-dimensional time-series causal inference—efficient, architecture-agnostic, and scalable.
Technology Category
Application Category
📝 Abstract
With the advancement of deep learning technologies, various neural network-based Granger causality models have been proposed. Although these models have demonstrated notable improvements, several limitations remain. Most existing approaches adopt the component-wise architecture, necessitating the construction of a separate model for each time series, which results in substantial computational costs. In addition, imposing the sparsity-inducing penalty on the first-layer weights of the neural network to extract causal relationships weakens the model's ability to capture complex interactions. To address these limitations, we propose Gradient Regularization-based Neural Granger Causality (GRNGC), which requires only one time series prediction model and applies $L_{1}$ regularization to the gradient between model's input and output to infer Granger causality. Moreover, GRNGC is not tied to a specific time series forecasting model and can be implemented with diverse architectures such as KAN, MLP, and LSTM, offering enhanced flexibility. Numerical simulations on DREAM, Lorenz-96, fMRI BOLD, and CausalTime show that GRNGC outperforms existing baselines and significantly reduces computational overhead. Meanwhile, experiments on real-world DNA, Yeast, HeLa, and bladder urothelial carcinoma datasets further validate the model's effectiveness in reconstructing gene regulatory networks.
Problem
Research questions and friction points this paper is trying to address.
High computational costs in neural Granger causality models
Weakness in capturing complex interactions due to sparsity penalties
Need for flexible, architecture-agnostic causality inference methods
Innovation
Methods, ideas, or system contributions that make the work stand out.
Uses gradient regularization for causality
Single model for all time series
Flexible with various architectures
🔎 Similar Papers
No similar papers found.
