MLExpert: 35B MoE Agent Trained to Autonomously Handle Machine Learning Engineering
Shangheng Du · hf · 2026-10-16
East China Normal University and Shanghai AI Laboratory introduce MLExpert, a 35B mixture-of-experts generalist agentic model for autonomous machine learning engineering (MLE), targeting the AI4AI and recursive self-improvement path.
- Problem: In long-horizon MLE runs, each experiment's feedback reflects several preceding decisions, and runs contain many costly experiments—training must select useful supervision from long trajectories and learn from intermediate outcomes.
- Method: A post-training framework combining hierarchical supervised fine-tuning and windowed interaction reinforcement learning, plus a Kaggle MLE Factory that constructs executable tasks and a model-directed interaction harness.
- Result: MLExpert autonomously organizes experiments and adapts optimization strategies from execution feedback to optimize long-horizon MLE workflows.
More from coding & agent
- Vibe coder chains ChatGPT sessions into Claude Code CLI via Desktop Commander MCP — MidahBootyQuay · 2026-10-16
- What a single operator can do with many agents is mind blowing — natesiggard · 2026-10-16
- When is a Claude Code session too long? One 3-hour spiral reignites the context debate — Additional-Buy-2562 · 2026-10-16
- Claude Code quirk: team member agent replies never reach the lead agent — Sauers_ · 2026-10-16
- Agent wrote the full answer, but the harness silently threw it away as a 'progress note' — Stunning-Sherbet1853 · 2026-10-16
- Vibe coding gets a bad rap: an engineer draws the line between vibes and real engineering — bendee983 · 2026-10-16