Research map / Label-efficient and robust learning / Recognition and long-tail
Scientifically-Interpretable Reasoning Network (ScIReN): Discovering Hidden Relationships in the Carbon Cycle and Beyond
Joshua Fan, Haodi Xu, Feng Tao, Md Nasim, Marc Grimson, Yiqi Luo, Carla P. Gomes
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use linear (hard-sigmoid/ReLU) constraint instead of nonlinear sigmoid/softplus (ablation) | Blackbox-Hybrid · Ecosystem respiration (linear Rb) | R2 latent parameter: 0.818 -> 0.941 | Improved |
| use linear constraint instead of nonlinear constraint (ablation) | Blackbox-Hybrid · Ecosystem respiration (nonlinear Rb) | KL of functional relationships: 0.937 -> 0.199 | Improved |
| use linear constraint instead of nonlinear constraint (ablation) | ScIReN (1-layer KAN) · Ecosystem respiration (linear Rb) | KL of functional relationships: 0.002 -> 0.001 | Improved |
| use 2-layer KAN encoder instead of 1-layer (ablation) | ScIReN · Ecosystem respiration (nonlinear Rb) | R2 observed variable: 0.651 -> 0.961 | Improved |
| use 2-layer KAN encoder instead of 1-layer (ablation) | ScIReN · Ecosystem respiration (nonlinear Rb) | R2 latent parameter: -1.225 -> 0.993 | Improved |
| use 2-layer KAN encoder instead of 1-layer (ablation) | ScIReN · Soil carbon (real labels) | R2: 0.582 -> 0.571 | No clear effect |
| add process-based decoder (hybrid) to pure NN, extrapolation | ScIReN · Ecosystem respiration (linear Rb) | R2 observed variable: 0.968 -> 0.976 | Improved |
| replace blackbox NN encoder with sparse KAN encoder | ScIReN · Soil carbon (synthetic parameters) | KL of functional relationships: 1.082 -> 0.080 | Improved |
| replace blackbox NN encoder with sparse KAN encoder | ScIReN · Soil carbon (synthetic parameters) | R2 latent parameter: 0.721 -> 0.989 | Improved |
| replace blackbox hybrid with ScIReN on real labels | ScIReN · Soil carbon (real labels) | R2: 0.589 -> 0.582 | No clear effect |
| replace blackbox NN with linear model encoder (Linear-Hybrid) | Blackbox-Hybrid · Ecosystem respiration (nonlinear Rb) | R2 observed variable: 0.961 -> 0.400 | Hurt |
| use ScIReN 1-layer KAN with nonlinear Rb | ScIReN · Ecosystem respiration (nonlinear Rb) | R2 observed variable: 0.961 -> 0.651 | Hurt |
Closest papers by meaning
- CarbonSense: A Multimodal Dataset and Baseline for Carbon Flux Modelling (ICLR 2025 · similarity 0.86)
- DiSciPLE: Learning Interpretable Programs for Scientific Visual Discovery (CVPR 2025 · similarity 0.83)
- Causal discovery with endogenous context variables (NeurIPS 2024 · similarity 0.82)
- From GNNs to Trees: Multi-Granular Interpretability for Graph Neural Networks (ICLR 2025 · similarity 0.82)
- Multi-modal Learning for Geospatial Vegetation Forecasting (CVPR 2024 · similarity 0.81)
- LLMCarbon: Modeling the End-to-End Carbon Footprint of Large Language Models (ICLR 2024 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
