Research map / Learning theory and algorithms / Prediction and decision losses
Optimal decision tree learning
Compute optimal or sparse decision trees with algorithmic and generalization guarantees.
75 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Prediction and decision losses, keyword cluster “trees · decision tree · forests”
State History Tried Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
18 papers in 2023–24 and 20 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic held steady, from 3.4% to 3.5%.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2016 · A Communication-Efficient Parallel Algorithm for Decision Tree (NeurIPS 2016 · 69 citations)
- 2017 · Variable Importance Using Decision Trees (NeurIPS 2017 · 61 citations)
- 2018 · Alternating optimization of decision trees, with application to learning sparse oblique trees (NeurIPS 2018 · 50 citations)
- 2019 · Compiling Bayesian Network Classifiers into Decision Graphs (AAAI 2019 · 56 citations)
- 2020 · Learning Optimal Decision Trees Using Caching Branch-and-Bound Search (AAAI 2020 · 110 citations)
- 2021 · Optimal Decision Trees for Nonlinear Metrics (AAAI 2021 · 16 citations)
- 2022 · Fast Sparse Decision Tree Optimization via Reference Ensembles (AAAI 2022 · 28 citations)
- 2023 · Optimal Decision Diagrams for Classification (AAAI 2023 · 12 citations)
- 2024 · Symbolic Regression Enhanced Decision Trees for Classification Tasks (AAAI 2024 · 15 citations)
- 2025 · Leveraging Predictive Equivalence in Decision Trees (ICML 2025 · 1 citation)
- 2026 · DFDT: Dynamic Fast Decision Tree for IoT Data Stream Mining on Edge Devices (AAAI 2026 · 1 citation)
Earliest papers here
- A Communication-Efficient Parallel Algorithm for Decision Tree (NeurIPS 2016 · 69 citations)
- Sparse Perceptron Decision Tree for Millions of Dimensions (AAAI 2016 · 17 citations)
- Random Composite Forests (AAAI 2016 · 8 citations)
What has been tried
56 papers here report 436 measured changes (a component added, removed or swapped, compared with a baseline): helped 66%, no clear effect 8%, hurt 21%, mixed 5%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| add pretrained equation initialization from the AI-Feynman corpus | Cancer · accuracy | 0.9644 -> 0.9546 | Hurt | Symbolic Regression Enhanced Decision Trees for Classification Tasks (AAAI 2024 · 15 citations) |
| Use fitness function (2) instead of fitness function (1) | Sensitivity analysis datasets · accuracy | .832 ± .12 -> 0.638 ± .16 | Hurt | Generative Model for Decision Trees (AAAI 2024 · 5 citations) |
| Tune the NE robustness parameter λ for each noise setting | MNIST digits · test accuracy | Mixed | Robust Loss Functions for Training Decision Trees with Noisy Labels (AAAI 2024 · 4 citations) | |
| increase the number of internal nodes | Occupational gender bias case study · training error | Helped | Extracting PAC Decision Trees from Black Box Binary Classifiers: The Gender Bias Study Case on BERT-based Language Models (AAAI 2025 · 0 citations) | |
| Increase lookahead depth | Aggressively binarized datasets · regularized training loss | Mixed | Near-Optimal Decision Trees in a SPLIT Second (ICML 2025 · 0 citations) | |
| Use λ = 0.0, omitting recourse-risk-aware greedy splitting before set-cover relabeling | FICO, Bail · accuracy | null | Hurt | Learning Decision Trees and Forests with Algorithmic Recourse (ICML 2024 · 0 citations) |
What reviewers will expect
Datasets and baselines named in the 13 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- COMPAS · 3 of 13 papers
- Adult · 3 of 13 papers
- Bank · 3 of 13 papers
- HELOC · 2 of 13 papers
- Netherlands · 2 of 13 papers
- Bike · 2 of 13 papers
- FICO · 2 of 13 papers
Baselines
- CART · 4 of 13 papers
- TreeFARMS · 2 of 13 papers
- STreeD · 2 of 13 papers
- AxTAO · 2 of 13 papers
- DPDT · 2 of 13 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 38 papers; from OpenAlex and the papers' affiliations).
- Duke University · 5 papers
- University of British Columbia · 3 papers
- Duke · 3 papers
- University of Milan · 2 papers
- Stanford University · 2 papers
- Fujitsu Ltd · 2 papers
- Fujitsu Limited · 2 papers
- Delft University of Technology · 2 papers
- University of Toronto · 2 papers
Use it
Download BibTeX of 39 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Tree ensemble optimization (77 papers)
- Compact interpretable model classes (26 papers)
Elsewhere in the topic
- Adaptive testing and experiments (201 papers)
- Active learning and subset selection (199 papers)
- Distributionally robust optimization (184 papers)
- Data attribution and recourse (160 papers)
- Theory of neural network sensitivity (143 papers)
- Ranking models from comparisons (127 papers)
- Determinantal point process inference (122 papers)
- Conformal prediction sets (116 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).