Haris Šikić

2 accepted papers

2026

Cut Less, Fold More: Model Compression through the Lens of Projection Geometry

Olga Saukh, Dong Wang, Haris Šikić, Yun Cheng, Lothar Thiele

ICLR 2026poster

Compressing neural networks without retraining is vital for deployment at scale. We study calibration-free compression through the lens of projection geometry: structured pruning is an axis-aligned projection, whereas model folding performs a low-rank projection via weight clustering. We formalize b…

Cited by 0 Source ↗Scholar ↗

2025

Forget the Data and Fine-Tuning! Just Fold the Network to Compress

Dong Wang, Haris Šikić, Lothar Thiele, Olga Saukh

ICLR 2025poster

We introduce model folding, a novel data-free model compression technique that merges structurally similar neurons across layers, significantly reducing the model size without the need for fine-tuning or access to training data. Unlike existing methods, model folding preserves data statistics during…

Cited by 0 Source ↗PDF ↗code ↗Scholar ↗