DSWorld: A Data Science World Model for Efficient Autonomous Agents
Quick Answer
The DSWorld framework introduces a Data Science World Model that predicts data science operation effects, enhancing autonomous agent efficiency.
Quick Take
It accelerates RL agent training by 14x and inference by 3-6x, outperforming the top baseline by 35.6% in transition prediction tasks.
Key Points
- DSWorld combines structured state construction and cost-aware routing for efficiency.
- Introduces an 8K-scale transition trajectory dataset for training purposes.
- Employs Reflective World Model Optimization for improved transition prediction.
- Achieves a 14x acceleration in RL-based agent training.
- Outperforms the strongest LLM baseline by 35.6% on transition prediction tasks.
DeepSignal Analysis
What happened
The DSWorld framework introduces a Data Science World Model aimed at improving the efficiency of autonomous data science agents. It predicts the effects of data science operations, significantly accelerating reinforcement learning (RL) agent training and inference times compared to existing models.
Key evidence
- DSWorld accelerates RL agent training by approximately 14 times and inference by 3 to 6 times.
- The framework outperforms the strongest large language model (LLM) baseline by 35.6% in transition prediction tasks.
- An 8K-scale transition trajectory dataset was constructed to support the training of the DSWorld framework.
Why it matters
The reliance on trial-and-error methods in autonomous data science can lead to high computational costs. By predicting the outcomes of data science operations, DSWorld could reduce these costs and improve the efficiency of data science workflows. This advancement may influence how autonomous agents are developed and utilized in various applications.
What to watch
Future developments should focus on the practical applications of DSWorld in real-world scenarios. Observing how this framework integrates with existing data science tools and its performance in diverse environments will be crucial. Additionally, the long-term implications of its error-aware reinforcement learning strategy warrant attention.
Paper Resources
📖 Reader Mode
~2 min readAbstract:Despite strong capabilities in data understanding and decision-making, autonomous data science agents still heavily rely on trial-and-error workflows that involve expensive computation. This bottleneck motivates models that can anticipate the effects of data science operations before real execution. In this paper, we introduce the concept of Data Science World Model, which model the data science execution environment by predicting environment state transitions conditioned on current workflow states and candidate operations. We further propose DSWorld, a practical framework that combines structured state construction, cost-aware routing, lightweight real execution, and an LLM-based simulator for expensive operations. To support training, we construct an 8K-scale transition trajectory dataset and introduce Reflective World Model Optimization, an error-aware reinforcement learning strategy for improving transition prediction. Experiments show that DSWorld accelerates RL-based agent training by approximately $14\times$ and search-based inference by approximately $3$-$6\times$ while maintaining competitive performance, and outperforms the strongest LLM baseline by 35.6% on transition prediction tasks. The code is available at this https URL.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2607.15901 [cs.AI] |
| (or arXiv:2607.15901v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2607.15901 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Zherui Yang [view email]
[v1]
Fri, 17 Jul 2026 12:14:55 UTC (1,354 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.AI
See more →HOBA: Hierarchical On-Policy Bidding Agents for Adaptive Online Advertising
HOBA (Hierarchical On-policy Bidding Agents) is a novel hierarchical reinforcement learning framework that enhances online advertising bidding systems by improving adaptability and reducing hyperparameter tuning costs. It utilizes a for hyperparameter inference, a SARSA agent for expert model selection, and a dynamic expert pool for bid execution, achieving a +3.6% increase in target cost during large-scale deployment and outperforming state-of-the-art baselines on AuctionNet.