CompJudge: Fine-Grained Comparative Evaluation using Multimodal LLM for Subject-Driven Generation

Nam Hyeon-Woo, Wenbin Ouyang, Ciprian A Corneanu

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

Phase Transitions in Attention: A Bayesian Theory of Copy Head Emergence

Bayesian theory reveals softmax attention learns copy heads via a first-order data phase transition, unlike linear attention's second-order transition and crossover.

Itay Lavie, Kirsten Fischer, Andrey Lekov, Frederic Van Maele and 2 more

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

Croissant Baker: Metadata Generation for Discoverable, Governable, and Reusable ML Datasets

Croissant Baker generates validated Croissant metadata locally from dataset directories via modular handlers, achieving 97, 100% agreement with ground truth across 140+ datasets including MIMIC-IV.

Rafi Al Attrach, Rajna Fani, Sebastian Lobentanzer, Joan Giner-Miguelez and 16 more

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

Sample Efficient Generative Molecular Optimization with Joint Self-Improvement

Joint Self-Improvement uses a joint generative-predictive model and self-improving sampling to reduce distribution shift and efficiently generate optimized molecules under limited evaluation budgets.

Serra Korkmaz, Adam Izdebski, Jonathan Pirnay, Rasmus Møller-Larsen and 5 more

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