Research map / Learning theory and algorithms / Combinatorial and search optimization
Multi-objective optimization
Finds Pareto solutions with learned partitions, preference-guided scalarization and evolutionary methods.
93 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Combinatorial and search optimization, keyword cluster “objective · multi · solutions”
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.
25 papers in 2023–24 and 41 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 7.2% to 11.3%.
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 · Multi-Objective Self-Paced Learning (AAAI 2016 · 33 citations)
- 2017 · Self-Paced Multi-Task Learning (AAAI 2017 · 81 citations)
- 2018 · Multi-Task Learning as Multi-Objective Optimization (NeurIPS 2018 · 327 citations)
- 2019 · Evolutionary Manytasking Optimization Based on Symbiosis in Biocoenosis (AAAI 2019 · 73 citations)
- 2020 · Efficient Continuous Pareto Exploration in Multi-Task Learning (ICML 2020 · 31 citations)
- 2021 · Learning the Pareto Front with Hypernetworks (ICLR 2021 · 38 citations)
- 2022 · A First Mathematical Runtime Analysis of the Non-dominated Sorting Genetic Algorithm II (NSGA-II) (AAAI 2022 · 63 citations)
- 2023 · Runtime Analysis for the NSGA-II: Provable Speed-Ups from Crossover (AAAI 2023 · 51 citations)
- 2024 · Towards Running Time Analysis of Interactive Multi-Objective Evolutionary Algorithms (AAAI 2024 · 10 citations)
- 2025 · A Many-Objective Problem Where Crossover Is Provably Indispensable (AAAI 2025 · 5 citations)
- 2026 · Improved Runtime Guarantees for the SPEA2 Multi-Objective Optimizer (AAAI 2026 · 2 citations)
Earliest papers here
- Multi-Objective Self-Paced Learning (AAAI 2016 · 33 citations)
- Self-Paced Multi-Task Learning (AAAI 2017 · 81 citations)
- Solving High-Dimensional Multi-Objective Optimization Problems with Low Effective Dimensions (AAAI 2017 · 72 citations)
What has been tried
67 papers here report 616 measured changes (a component added, removed or swapped, compared with a baseline): helped 65%, no clear effect 9%, hurt 20%, mixed 6%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Use R-NSGA-II with N=4(n+1) rather than N=1 | OneMinMax · fitness evaluations | No clear effect | Towards Running Time Analysis of Interactive Multi-Objective Evolutionary Algorithms (AAAI 2024 · 10 citations) | |
| Use one-bit mutation instead of standard bit-wise mutation | DLTB · coverage of the full Pareto front | Hurt | How to Use the Metropolis Algorithm for Multi-Objective Optimization? (AAAI 2024 · 9 citations) | |
| Use Phase 1 only rather than mixing both phases | NYUD-v2 · MTP (△m ↑) | + 15.50 -> + 13.44 | Hurt | Quantifying Task Priority for Multi-Task Optimization (CVPR 2024 · 7 citations) |
| Disable crossover (pc = 0) | m-RRMO · generations to find any Pareto-optimal point | nΩ(n) | Hurt | A Many-Objective Problem Where Crossover Is Provably Indispensable (AAAI 2025 · 5 citations) |
| use predicted gradients without outcome evaluations or gradient-information optimizations | DTLZ1, DTLZ2, DH1 · utility regret | Hurt | User Preference Meets Pareto-Optimality in Multi-Objective Bayesian Optimization (AAAI 2025 · 5 citations) | |
| Add affine-transform dominance to node-size child ordering | Amazon River basin, ECG, ϵ=0.025 · running time (seconds) | 449 -> 369 | Helped | Scaling Up Pareto Optimization for Tree Structures with Affine Transformations: Evaluating Hybrid Floating Solar-Hydropower Systems in the Amazon (AAAI 2024 · 5 citations) |
| set smoothing parameter μ to 0.01 | Engineering design problems · ΔHV | Mixed | Smooth Tchebycheff Scalarization for Multi-Objective Optimization (ICML 2024 · 3 citations) | |
| add PL-subspace sampling | CJSP1-CJSP10 · Δ(·) | 3-6-1 | Helped | Balanced Adaptive Subspace Collaboration for Mixed Pareto-Lexicographic Multi-Objective Problems with Priority Levels (AAAI 2025 · 1 citation) |
| add CFF, the flat-fitness coping strategy | TN2 (3 objectives) · error metric | 0.00016±0.000351 -> 0.000172±0.000535 | No clear effect | Boundary Decomposition for Nadir Objective Vector Estimation (NeurIPS 2024 · 1 citation) |
| Vary the tensor rank r | RESISC45 and GTSRB, 2-model merge · test accuracy | Mixed | Pareto Merging: Multi-Objective Optimization for Preference-Aware Model Merging (ICML 2025 · 0 citations) | |
| Increase nβ from 20 to 500 without PFE | DTLZ2-M2 · HV | 0.4041 -> 0.4164 | Helped | MOBO-OSD: Batch Multi-Objective Bayesian Optimization via Orthogonal Search Directions (NeurIPS 2025 · 0 citations) |
| Increase SimPO γ from 0 to 10 | AlpacaEval, dpo-mix-7k · win-rate | 30.50 -> 33.60 | Helped | Meta-Learning Objectives for Preference Optimization (NeurIPS 2025 · 0 citations) |
What reviewers will expect
Datasets and baselines named in the 29 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- MultiMNIST · 3 of 29 papers
- DTLZ2 · 2 of 29 papers
- Taskonomy · 2 of 29 papers
- UTKFace · 2 of 29 papers
- SARCOS · 2 of 29 papers
- Cityscapes · 2 of 29 papers
- NYU-v2 · 2 of 29 papers
Baselines
- Nash-MTL · 5 of 29 papers
- linear scalarization · 3 of 29 papers
- MGDA · 3 of 29 papers
- DWA · 3 of 29 papers
- PCGrad · 3 of 29 papers
- CAGrad · 3 of 29 papers
- IMTL · 3 of 29 papers
- SMS-EMOA · 2 of 29 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 66 papers; from OpenAlex and the papers' affiliations).
- Southern University of Science and Technology · 4 papers
- Cornell University · 4 papers
- City University of Hong Kong · 4 papers
- Nanjing University · 3 papers
- University of Passau · 2 papers
- Ecole Polytechnique · 2 papers
- Stanford University · 2 papers
- Shenzhen University · 2 papers
- Korea Advanced Institute of Science and Technology · 2 papers
- Nanyang Technological University · 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
- Constrained neural network optimization (111 papers)
- Unsupervised neural combinatorial optimization (113 papers)
- Decision-focused optimization (97 papers)
- Globally optimal geometric optimization (56 papers)
Elsewhere in the topic
- Logic and answer set programming theory (150 papers)
- Heuristic and local search algorithms (147 papers)
- Classical and HTN planning (145 papers)
- Structure and complexity of constraint reasoning (92 papers)
- Predict-then-optimize learning (89 papers)
- SAT and model counting solvers (89 papers)
- Heuristics for integer programming (60 papers)
- Differentiable and symbolic reasoning (56 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).