# CoreWeave ARIA: The autoresearch loop for continuous improvement

## Executive summary

CoreWeave ARIA is an AI Research and Iteration Agent designed to automate the full autoresearch loop for continuous model and agent improvement. The demonstration shows how ARIA autonomously forms hypotheses, analyzes prior run results (including raw system metrics and plots), sets filters, and executes new training jobs via Weights & Biases Launch, allowing human users to focus on high-level problem definition.

## Key takeaways

- Autonomous Research Loop: ARIA autonomously manages the research process by forming hypotheses, running experiments, evaluating results, and executing optimal next actions without constant human intervention. This capability helps models and agents improve continuously.
- Run Analysis and Filtering: ARIA can analyze complex run data—including plots, tables, and raw system metrics—to determine what worked best. It can also set UI filters based on previous sweeps (e.g., setting a filter for 'auto research runs').
- Parallel Experimentation: Users can run multiple ARIA instances in parallel to accelerate the workflow, allowing simultaneous management of different research tracks (e.g., running two distinct ARIA variants).

## Technical details

- Autoresearch Workflow: The process involves ARIA inspecting the prior state, generating hypotheses (e.g., suggesting changes to block size, max iterations, dropout rate, and learning rates), and then triggering multiple training jobs.
- Job Execution: New research runs are triggered using Weights & Biases Launch, which manages job queues on specialized hardware (e.g., A100 GPUs). The system can manage multiple configurations running in parallel.
- Agent Interaction: ARIA interacts with the user by asking specific questions, such as how many trials should be run on a configured GPU queue (e.g., 'three trials').

## Practical implications

- Automating the iterative cycle of machine learning research (hypothesis generation, execution, evaluation) significantly reduces manual overhead.
- The system provides a mechanism for systematic hyperparameter optimization and architectural exploration by managing large sweeps of parallel jobs.
- Build engineers can leverage this concept to automate complex build or deployment testing loops that require continuous self-correction and resource management.

## Topics

AI Agents, Machine Learning Operations (MLOps), Hyperparameter Optimization, Weights & Biases, Autoresearch, Job Scheduling/CI, Weights & Biases Documentation for ARIA

Source: https://www.youtube.com/watch?v=z8zZnAcj8Ds