Generated from prompt:

Create a 6-slide English PowerPoint presentation based on the paper 'VDSAgents: A PCS-Guided Multi-Agent System for Veridical Data Science Automation'. Style: simple, white academic theme. Include Figure 2 (system architecture) and Figure 3 (results chart) from the paper. Each slide should include concise English bullet points and have the oral script provided below in the speaker notes. Slide 1: Title – 'VDSAgents: A PCS-Guided Multi-Agent System for Reliable Data Science' with presenter name 'Yunxuan Jiang' and course 'Academic English II'. Speaker Notes: Good afternoon everyone. My topic today is VDSAgents — a new multi-agent system that makes AI-driven data science more reliable and trustworthy. In simple words, it’s about teaching AI not only to analyze data, but to think scientifically. Slide 2: Problem – 'Unstable and non-reproducible AI systems'. Points: LLMs automate data science; AutoKaggle and DataInterpreter are often unstable; need for trustworthy automation. Speaker Notes: Many people now use large language models to automate data science — for example, cleaning data or training models. But these systems often give results that look correct, but can’t be trusted. Sometimes they crash or give different answers. So our question was — how can we make AI data science more stable and scientific? Slide 3: Idea – 'PCS-Guided Agents'. Points: Use Veridical Data Science (VDS); Apply PCS principles — Predictability, Computability, Stability; Build multi-agent collaboration. Speaker Notes: Our idea is simple — we bring human scientific thinking into AI. We use the PCS principles: Predictability means generalization, Computability means it runs reliably, and Stability means results don’t break under small changes. We design several AI agents that follow these principles at every step. Slide 4: How It Works – include Figure 2 (architecture diagram). Points: Five agents – Define, Explore, Model, Evaluate, PCS; PCS-Agent checks stability and logic. Speaker Notes: Please notice this part — this is the overall workflow of VDSAgents. We have five agents: Define, Explore, Model, Evaluate, and one special PCS-Agent in the center. The PCS-Agent acts like a coach — it checks if every step follows scientific logic. For example, when cleaning data, it tests different methods and checks which one is more reliable. Slide 5: Results – include Figure 3 (comparison chart). Points: Tested on nine datasets; Compared with AutoKaggle and DataInterpreter; Higher scores in almost all tasks. Speaker Notes: Now let’s look at our results — notice this part, the blue bars. This chart compares VDSAgents with AutoKaggle and DataInterpreter. You can see that VDSAgents clearly outperforms the others in almost every dataset. It achieves an average score of 0.82 — meaning it’s accurate and reliable. Slide 6: Takeaway – 'Reliable AI for Data Science'. Points: First PCS-guided multi-agent system; Improves stability and reproducibility; A step toward trustworthy AI. Speaker Notes: To sum up, VDSAgents combines automation and scientific reasoning. It improves both performance and trustworthiness in data science tasks. We believe this approach can help build a more reliable future for AI-driven research. Thank you very much for listening!