Skip to content

AISSAI Hackathon 2025

Date: November 3 to 7 - 2025

Location: Seolane - Barcelonnette, Alpes de Haute Provence, France

Proposed Hackathon Topics

Symbolic Regression for Stellar Systems

Principal Investigator: Paolo Bianchini

Focus: Discovering interpretable physics from N-body simulations

Scientific Challenge:
N-body simulations are essential tools in astrophysics for studying stellar system formation and evolution, but face the challenge of simultaneously accounting for complex multi-physics phenomena (stellar dynamics, stellar evolution, external interactions) across vastly different spatial and temporal scales.

Project Overview:
Use symbolic regression to discover interpretable analytic relations that capture stellar system evolution while disentangling the key driving parameters. The goal is to identify fundamental physical mechanisms and potentially discover new physics from simulation data.

Dataset & Methodology:
- Simulations: N-body simulations of globular clusters evolved over 14 Gyr using Nbody6++GPU
- Tool: Phi-SO symbolic regression: https://physo.readthedocs.io/en/latest/index.html
- Target Relations: Derive analytic formulas for global properties evolution: - Total mass evolution with time
- Angular momentum changes
- Black hole population dynamics
- Cluster structural parameters

Data Specifications:
- Reduced dataset: <10 GB (sufficient for hackathon purposes)
- Full simulation data: ~10 TB (star-by-star detailed properties)
- Simulation visualization example

Preliminary Results:
Previous Phi-SO tests successfully identified mathematical relations describing rotation profiles across different cluster snapshots, demonstrating the method's potential for discovering universal scaling laws.

Resource Requirements:
- Multiple CPU nodes preferred for parallel testing (different parameter setups)
- Standard laptop performance adequate for basic Phi-SO operations
- No GPU requirements (minimal performance improvement observed)

Note: Simulation paper currently under internal collaboration review; PI will provide necessary data access for hackathon participants.

[CANCELLED] Multi-Agent System with Humans-in-the-Loop

Principal Investigator: Ioana Ciucă

Focus: Scientific workflow automation with AI agents

Preparation:
- 2-hour lecture + 1.5-hour tutorial on transformer architecture
- Materials: AI4Astro IMPRS 2025

Project Summary:
Build an efficient and robust multi-agent system with humans-in-the-loop using state-of-the-art LLMs to undertake astrophysics research through clear-defined steps: generate ideas, run analyses, interpret results, and write research reports.

Project Structure:
1. Research Task Identification: Identify specific astrophysics research tasks (e.g., analyzing new observational data with testable hypotheses)
2. Evaluation Framework Development: Establish comprehensive evaluation rubric inspired by ResearchBench assessing: - Research quality and scientific rigor
- Information retrieval accuracy and citation trustworthiness
- Experimental reproducibility
- System collaboration effectiveness
3. System Design & Implementation: Multi-agent architecture featuring: - Orchestrator Agent: Decomposes complex research questions into actionable subtasks
- Execution Agent: Spawns specialized agents for subtask resolution (inspired by CMB Agent, AstroCoder)
- Human-in-the-Loop Control: Expert validation checkpoints and guidance
4. Capabilities Demonstration: Complete research workflow execution including hypothesis refinement, computational experiments, statistical analysis, and report generation

Technical Implementation:
- Foundation Models: Mistral LLMs
- Integrations: AstroCoder, Allen AI tools, astrophysics-specific libraries
- Infrastructure: Vector stores for scientific literature and documentation

Resource Requirements:
- GPU resources (AMU mesocentre + cloud compute backup)
- Storage for scientific datasets and vector embeddings

References: CMB Agent, VirtualLab, ResearchBench, Octotools, ColabLLM, AstroCoder, arXiv:2412.00431

Euclid Q1 – Anomaly Detection

Principal Investigator: Marc Huertas-Company and Malgorzata Siudek

Focus: Unsupervised identification of outliers

  • Based on the Euclid Q1 release
  • Work on instrumental artifacts and astrophysical anomalies
  • Use a multimodal approach (images, spectra, metadata)

Notes

  • Breakfast from 8:00 to 9:00
  • Morning tag up from 9:30 to 10:00 by group
  • Morning coffee break from 10:30 to 11:00
  • Lunch break from 12:30 to 14:00
  • Breakout sessions from 13:30 to 14:00
  • Afternoon coffee break from 15:30 to 16:00
  • Wrap-up session from 17:00 to 17:30 all together
  • Dinner from 19:00 to 20:30
  • Last day will devote to the final presentation of the hackathon