Learn Locally, Recurse Globally: Neural Circuit Synthesis Beyond Training Depth

Emile Richard

Atlanta Poster Session 6, Fri, Dec 11, 4:30 PM–7:30 PM, Hall C1 · Published 2026

CREF: Forecasting Benchmarks for the Age of Agents

Andreas Auer, Abdul Fatir Ansari, Oleksandr Shchur, Xiyuan Zhang and 5 more

Sydney Poster Session 5, Thu, Dec 10, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

IFDECORATOR: Wrapping Instruction Following Reinforcement Learning with Verifiable Rewards

Xu Guo, Tianyi Liang, Jian Tong, Xiaogui Yang and 5 more

Sydney Poster Session 4, Wed, Dec 9, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026

What Fits (Into Few Tokens) Doesn't Overfit: Compression and Generalization in ML Research Agents

LLM research agents find high-performance models via compressed prompts and feedback, supporting a description-length explanation for limited overfitting.

Martin Bertran, Aaron Roth, Steven Wu

Atlanta Poster Session 5, Fri, Dec 11, 10:00 AM–1:00 PM, Hall C1 · Published 2026

MURPHY: Feedback-Aware GRPO with Retrospective Credit Assignment for Multi-Turn Code Generation

MURPHY extends GRPO to multi-turn code generation via feedback-conditioned rollout trees with retrospective credit assignment, achieving up to 6% absolute pass@1 gains over prior methods.

Chanakya Ekbote, Vijay Lingam, Sujay Sanghavi, Luke Huan and 3 more

Sydney Poster Session 3, Wed, Dec 9, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

Consilience for Verifier-Free Test-Time Scaling

Confidence-based verifier-free test-time scaling fails on complex tasks because high initial confidence signals no exploration; consilience selects rollouts by requiring low early but high final confidence, improving reasoning and coding.

Lecheng Kong, Like Hui, Haitao Mao, Luke Huan

Sydney Poster Session 4, Wed, Dec 9, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026 · ▲ 2 on Hugging Face

Synthetic Pre-Pre-Training Improves Language Model Robustness to Noisy Pre-Training Data

Synthetic pre-pre-training improves language model robustness to noisy pre-training data by inhibiting noise self-modeling and reducing required natural-text tokens by up to 49%.

Xu Guo, Runyu Peng, Jian Tong, Yunhua Zhou and 3 more

Sydney Poster Session 6, Thu, Dec 10, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026