Research map / Large language models / AI and society
Towards Trustworthy, Efficient, and Scalable Machine Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
A doctoral consortium-style summary of the author's work on trustworthy (fair, private, explainable) and scalable (graph and image dataset condensation) machine learning, plus planned projects.
Problem
Trustworthiness (bias, privacy, explainability) and efficiency/scalability of machine learning, especially for graph neural networks and large image datasets.
Approach
Summarizes several prior and planned works: a data-centric fair graph learning approach, FGU for fair graph unlearning, DGExplainer for dynamic GNN explanation via LRP, and joint condensation methods (TinyGraph, TinyData) that condense both nodes/samples and features.
Key results
- On CIFAR10 at 1% condensation ratio, FairGraph-style joint condensation reports 52.4% accuracy vs 27.3% full-feature baseline, 48.8% DC-PCA (table from planned-work section, as reported).
- On MNIST at 0.83% condensation ratio, the reported accuracy is 97.8% versus 96.4% for the Coarsening baseline and 94.1% for DC-PCA.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use FairGraph instead of Coarsening | FairGraph · MNIST, data condensation ratio 0.017% | accuracy (%): 88.4 ± 0.3 -> 88.3 ± 0.7 | No clear effect |
| Use FairGraph instead of DC-PCA | FairGraph · MNIST, data condensation ratio 0.17% | accuracy (%): 91.3 ± 0.4 -> 95.7 ± 1.1 | Helped |
| Use FairGraph instead of Coarsening | FairGraph · MNIST, data condensation ratio 0.83% | accuracy (%): 96.4 ± 0.5 -> 97.8 ± 1.6 | Helped |
| Use FairGraph instead of DC-PCA | FairGraph · CIFAR10, data condensation ratio 0.02% | accuracy (%): 24.6 ± 0.7 -> 27.6 ± 0.8 | Helped |
| Use FairGraph instead of DC-PCA | FairGraph · CIFAR10, data condensation ratio 0.2% | accuracy (%): 40.4 ± 1.1 -> 43.6 ± 1.0 | Helped |
| Use FairGraph instead of DC-PCA | FairGraph · CIFAR10, data condensation ratio 1% | accuracy (%): 48.8 ± 0.9 -> 52.4 ± 0.7 | Helped |
Limitations
- Very short summary; method details and results for most works are not given.
- Table 1 mixes baselines with unnamed method columns, making attribution unclear.
Experimental setup
Datasets: MNIST CIFAR10
Benchmarks: MNIST CIFAR10
Metrics: accuracy
Baselines: Coarsening DC-PCA
Builds on: layer-wise relevance propagation gradient matching
Closest papers by meaning
- The Mainstays of Trustworthy Machine Learning (AAAI 2025 · similarity 0.83)
- Position: Application-Driven Innovation in Machine Learning (ICML 2024 · similarity 0.82)
- Scalable and Trustworthy Learning in Heterogeneous Networks (AAAI 2025 · similarity 0.80)
- Towards Trustworthy Machine Learning Under Distribution Shifts (AAAI 2025 · similarity 0.79)
- Data-Driven Discovery of Design Specifications (Student Abstract) (AAAI 2024 · similarity 0.79)
- What Do Machine Learning Researchers Mean by “Reproducible”? (AAAI 2025 · similarity 0.79)
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.