Ian Osband's 3-minute Colab shows horizon loss beats policy gradient and cross-entropy
redpony · x · 2026-10-10
DeepMind researcher Ian Osband released a Colab for his "Planning to Learn" work that runs in 3 minutes on a free T4 GPU. redpony shared it, noting single-Colab experiments still reveal important things in 2026.
Results, fully reproducible:
- Exact policy gradient loses to cross-entropy, even on accuracy.
- The horizon loss beats both methods.
- Planning CE even beats plain CE on its own NLL metric.
Nothing is hidden; readers are encouraged to run it themselves.
More from Research
- MathEconLib open-sourced: Axiom Math and top economists bring machine-checked proofs to economics in Lean 4 — soumitrashukla9 · 2026-10-23
- Ken Ono invites the community to build MathEcon on the new Lean 4 economics library — soumitrashukla9 · 2026-10-23
- IFM open-sources RL360: 8,192 concurrent sandboxes and ~1.3M rollouts for agentic RL — rupspace · 2026-10-23
- Science lifts expression of concern on 2024 AI debunking-conspiracies paper; conclusions unchanged — DG_Rand · 2026-10-23
- Embodied First Exam benchmarks coding agents on 1,090 tasks across 44 robots — ManlingLi_ · 2026-10-23
- Anima Anandkumar on neural operators: AI weather models beat supercomputers — AnimaAnandkumar · 2026-10-23