Organizations: College of Science, Northwest A&F University, Yangling 712100, China · Faculty of Data Science, City University of Macau, Macau, China · College of Economics and Management, Northwest A&F University, Yangling 712100, China
Multi-objective evolutionary optimization increasingly involves large populations, expensive problem evaluations, and diverse algorithmic structures. These workloads expose the limited parallelism of loop-based implementations, while rewrites for accelerator execution can alter selection order, tie handling, and population updates. AscendMOEA is presented as a scalable platform that executes multi-objective evolutionary algorithms (MOEAs) on Huawei Ascend neural processing units (NPUs) and conventional central processing units (CPUs). The platform represents the decision vectors, objectives, constraints, and auxiliary states as tensors and provides reusable interfaces for algorithms, problems, population operators, quality indicators, monitoring, and repeated experiments. Dense computations are executed in batches, such as dominance, distance, variation, and problem-evaluation calculations. Short recurrences with causal dependencies retain their prescribed order and consume precomputed tensor results, which preserves the behavior of the original algorithms. AscendMOEA contains 18 algorithms, 155 benchmark and application problems, 19 public operator interfaces, and 11 quality indicators. Experiments on eight Ascend NPUs and 192 physical CPU cores evaluate operator throughput, end-to-end runtime, memory consumption, numerical agreement, and implementation ablation. Tensorized operators obtain a median acceleration of 1.58 times on the NPU. At large population scales, complete algorithms achieve a median acceleration of 5.76 times while retaining the recorded quality indicators. Moreover, the results identify the workload scale and ordered computation that determine whether accelerator execution is beneficial. AscendMOEA is open-sourced at https://github.com/dqlme/AscendMOEA.
Figures & tables
System
Primary scope
Software organization
Parallel or accelerator support
Reported validation
PlatEMO [ 4 ]
Evolutionary multi-objective optimization
Algorithms, problems, indicators, experiments, and visualization in MATLAB
Parallel evaluation through the MATLAB ecosystem
Algorithm behavior and benchmark experiments
pymoo [ 5 ]
Multi-objective optimization in Python
Modular sampling, selection, variation, termination, and decision support
Parallel problem evaluation and vectorized numerical components
Optimization quality and software examples
EvoX [ 11 ]
General scalable evolutionary computation
Functional states, workflows, analyzers, and execution engines
Algorithms, problems, operators, indicators, monitors, and typed registries
Ascend NPU and CPU execution with ordered host control
Operator equivalence, generation state, terminal indicators, runtime, and memory
TABLE I: Representative software for multi-objective optimization and accelerator execution. The final column summarizes the validation emphasized in the corresponding publication.
Fig. 1: Layered architecture and component portfolio of AscendMOEA.
Fig. 2: Execution and validation process for semantics-preserving tensorization. Interchangeable population calculations are executed as dense NPU tensors, while calculations with causal dependencies retain their prescribed order.
Category
Count
MOEAs or MOP suites
Principal tensorized calculations
General MOEAs
3
NSGA-II [ 21 ] , MOEA/D [ 8 ] , and SPEA2 [ 9 ]
Dominance, fronts, crowding, scalarization, neighborhoods, strength fitness, and exact truncation
N=100 and FEmax=10,000 for every compatible algorithm–problem pair
20 paired seeds
Runtime, complete population histories, indicators, FE, terminal state, and memory
Scale algorithms
(N,FEmax) from (1000,1×105) to (20000,2×106)
5 paired seeds
CPU-to-NPU speedup, convergence, memory growth, OOM, and timeout
Implementation ablation
Reference, common-operator, and full tensorized implementations under matched settings
5 paired seeds
Reference/operator, operator/full, and reference/full timing gains
TABLE III: Experimental workloads and recorded evidence.
Fig. 3: Timing gain of the fully tensorized NPU operators relative to their source-consistent implementations. A value above 1× favors tensorization, and the configuration labels follow N∣M∣D .
Fig. 4: Representative scaling of the synchronized time per generation for four MOEA categories. Lines and shaded regions show medians and interquartile ranges over five float32 seeds. Each value is the synchronized solve time divided by (FE−N)/N . A missing device point indicates that none of its five runs finishes successfully at that configuration.
Fig. 5: Cumulative synchronized compute time saved by the NPU over successful matched scale runs. Each colored area gives the contribution of one MOP category, and the black line gives the total.
Fig. 6: NPU timing gains from common-operator replacement and tensorization inside each algorithm. Violin envelopes show the distributions, internal boxes show interquartile ranges, and zero on the log2 scale denotes parity.
Experiment
Pairs
Clusters
Median
95% CI
Standard workload
20,100
1,005
0.980×
0.976 – 0.983×
Scale workload
373
77
5.76×
3.26 – 9.09×
Real-world subset
2,920
146
0.965×
0.959 – 0.972×
TABLE IV: Aggregate synchronized timing results for the full float32 implementation.
Optimizing AscendC (Ascend C) operators for Ascend NPUs is difficult for two reasons. First, unlike CUDA, the ecosystem offers few public kernels to learn from. Second, performance depends on a coupled two-part implementation: a host-side tiling program that controls data movement and a kernel program that schedules and pipelines computation. We present AscendOptimizer, an episodic agent that builds missing optimization knowledge from execution itself. For kernel optimization, AscendOptimizer rewinds strong implementations by removing optimizations in a controlled way, then keeps the changes whose removal measurably hurts performance as reusable experience for later rewriting. For host-side optimization, it runs profiling-in-the-loop evolutionary search to find valid, fast tiling and data-movement configurations directly from hardware feedback. This combination lets the agent improve kernel structure and host-side scheduling together. On a benchmark of 101 real AscendC operators, AscendOptimizer achieves a 1.21x geometric-mean speedup over the open-source baseline, and 53.47% of operators run faster than their references. Given a same budget of evaluations per operator, AscendOptimizer consistently outperforms Best-of-N sampling and OpenEvolve in terms of geometric mean speedup, fast_p tail speedup ratios, and overall optimization progress across varying budgets.
Jiehao Wu, Zixiao Huang, Wenhao Li +3
School of Computer Science and Technology, East China Normal University · School of Computer Science and Technology, Tongji University · Shanghai University of International Business and Economics +3
To facilitate the solving of complex problems, evolutionary multitask optimization is regarded as a promising technique due to its capacity for cross-task knowledge transfer. In contrast to empirical evidences, no mathematical runtime support exists for this technique within the domain of multi-objective optimization. In this work, we aim to fill this important gap. We analyze how the representative multiobjective multifactorial evolutionary algorithm (MO-MFEA) covers the Pareto fronts of two bi-objective problems simultaneously. For the two-task problem of maximizing the classic bi-objective OMM and the multimodal OJZJ benchmarks, the MO-MFEA requires only an expected runtime of O(n2logn)---the same as optimizing the simple OMM alone---yet achieves a speedup by a factor of Θ(nk−1/logn) compared to optimizing OJZJ alone (with gap size k). Similarly, for the task pair comprising the classic LOTZ and the multimodal DLTB benchmarks, a speedup of Θ(n) is attained over optimizing DLTB alone, while the runtime matches that of optimizing LOTZ alone.
Danyang Zhang, Weijie Zheng
School of Computer Science and Technology, State Key Laboratory of Smart Farm Technologies and Systems, International Research Institute for Artificial Intelligence, Harbin Institute of Technology, Shenzhen, China · Pengcheng Laboratory, Shenzhen, China
LLM agents have become capable kernel engineers on mainstream GPU stacks, but they struggle on emerging accelerators such as Ascend NPUs, whose programming model is barely represented in pretraining data. Recent NPU agents address this gap largely through experience: they accumulate memory from their own generation and optimization trials and reuse it on later tasks. We argue that experience alone is not enough for kernel generation. An agent can only remember what it has tried, and it rarely tries a mechanism it does not know. As a result, such memory offers little help when a new operator calls for a different mechanism, and even familiar operators raise unseen situations as optimization proceeds. We present ASCENDEVOLVE, which grounds every step of an evolutionary kernel search in a large external knowledge base built from platform documentation, toolkit headers, and operator repositories. Because this corpus far exceeds any context window, ASCENDEVOLVE indexes it and routes each agent to a small, task-specific slice of evidence instead of the corpus itself. The retrieved knowledge seeds diverse implementation strategies, guides repair when a candidate fails, and constrains how strong candidates are recombined. On 53 Ascend C operators from CANN-Bench spanning four difficulty levels, ASCENDEVOLVE achieves an overall score of 71.39, improving over Reflexion and OpenEvolve by 11.6% and 17.2%.
Xi Chen, Ming Shen, Haotian Yan +2
Shenzhen Loop Area Institute (SLAI) · Guangzhou University · Shanghai Jiao Tong University +2