Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Learning to Generalize: An Information Perspective on Neural Processes
Authors: Hui Li, Huafeng Liu, Shuyang Lin, Jingyue Shi, Yiran Fu, Liping Jing
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | To address this, we propose an information-theoretic framework to analyze the generalization bounds of NPs, introducing dynamical stability regularization to minimize sharpness and improve optimization dynamics. Additionally, we show how noise-injected parameter updates complement this regularization. The proposed approach, applicable to a wide range of NP models, is validated through experiments on classic benchmarks, including 1D regression, image completion, Bayesian optimization, and contextual bandits. The results demonstrate tighter generalization bounds and superior predictive performance, establishing a principled foundation for advancing generalizable NP models. |
| Researcher Affiliation | Academia | 1State Key Laboratory of Advanced Rail Autonomous Operation, Beijing, China 2School of Computer Science and Technology, Beijing Jiaotong University, Beijing, China 3Beijing Key Laboratory of Traffic Data Mining and Embodied Intelligence, Beijing, China EMAIL |
| Pseudocode | Yes | E Algorithm Pseudocode We present the complete algorithm for Generalization Neural Processes (Gen-NPs) that integrates both the Risk-Aware Dynamical Stability Regularization (DSR) and Optimization-Aware Noise Injection Learning Strategy (NILS) components. Algorithm 1 provides a comprehensive pseudocode implementation that practitioners can follow to apply our method to various Neural Process variants. |
| Open Source Code | Yes | Codes: https: //github.com/Allen0497/Gen-NPs |
| Open Datasets | Yes | We evaluate Gen-NPs on image completion tasks using Celeb A [32] and EMNIST [5], formulated as a 2-D regression problem where pixel coordinates are inputs and intensities are outputs [15]. |
| Dataset Splits | Yes | For each function fi, N random locations are selected for evaluation, and an index m is chosen to divide the sequence into context points and target points. The parameters are set as follows: ℓ U[0.6, 1.0), σf U[0.1, 1.0), B = 16, N U[6, 50), and m U[3, 47). |
| Hardware Specification | Yes | GPU Model(s): Model: NVIDIA RTX A4000 Count: 8 GPUs Memory per GPU: 16 GB CPU Model(s): Model: Intel(R) Xeon(R) Platinum 8358P Core Count: 32 cores |
| Software Dependencies | Yes | CUDA: Version 11.8 cu DNN: Version 8.6 Py Torch: Version 2.0.0 Scikit-learn: Version 1.5.0 Num Py: Version 1.26.3 Pandas: Version 2.2.2 |
| Experiment Setup | Yes | The hyperparameters λ1 and λ2 control the relative importance of these two components, with empirical values typically set as λ1 [0.01, 0.1] and λ2 [0.001, 0.01] to maintain an appropriate balance between the two terms. |