When is Warmstarting Effective for Scaling Language Models?

Neeratyoy Mallik, Maciej Janowski, Johannes Hog, Herilalaina Rakotoarison and 3 more

Sydney Poster Session 1, Tue, Dec 8, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

On the Depth of Monotone ReLU Neural Networks and ICNNs

Monotone ReLU networks cannot compute or approximate maximum, ICNNs need depth n for it, and depth-k ICNNs cannot simulate some depth-2 ReLU networks.

Egor Bakaev, Florestan Brunck, Christoph Hertrich, Daniel Reichman and 1 more

Sydney Poster Session 1, Tue, Dec 8, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

I Have a Stream: Making Self-Supervised Learning Work on Continuous Video

Self-supervised video-stream pretraining fails due to intra-batch near-duplicate frames, but proposed StreamMAE with motion-biased crops matches i.i.d. MAE and scales to 95 hours.

Ivan Martinović, Lukas Knobel, Yuki Asano

Paris Poster Session 2, Wed, Dec 9, 5:00 PM–7:00 PM, Paris Poster Hall · Published 2026 · ▲ 14 on Hugging Face

Emergence of a Shared Canonical Object Frame from In-the-Wild Videos

Self-supervised training on 160,000 in-the-wild videos via a shared coarse mesh yields emergent canonical object frames without pose labels, matching supervised category-level pose estimation accuracy.

Tom Fischer, Martin Sundermeyer, Adam Kortylewski, Eddy Ilg

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