Informatics Institute faculty member Behçet Uğur Töreyin coauthored paper titled 'PURSUhInT: In Search of Informative Hint Points Based on Layer Clustering for Knowledge Distillation' has been published in 'Expert Systems with Applications' on 1st March 2023. 

DOI: https://doi.org/10.1016/j.eswa.2022.119040

Abstract:
One of the most efficient methods for model compression is hint distillation, where the student model is injected with information (hints) from several different layers of the teacher model. Although the selection of hint points can drastically alter the compression performance, conventional distillation approaches overlook this fact and use the same hint points as in the early studies. Therefore, we propose a clustering based hint selection methodology, where the layers of teacher model are clustered with respect to several metrics and the cluster centers are used as the hint points. Our method is applicable for any student network, once it is applied on a chosen teacher network. The proposed approach is validated in CIFAR-100 and ImageNet datasets, using various teacher–student pairs and numerous hint distillation methods. Our results show that hint points selected by our algorithm results in superior compression performance compared to state-of-the-art knowledge distillation algorithms on the same student models and datasets.

İTÜ Informatics Institute

bilisim-anasayfa-hakkimizda

ITU Informatics Institute provides graduate-level education and research in applied informatics, computer sciences, computational science and engineering, communication systems under the following programs.

Faculty members and students conduct research supported by national and international organızatıons in the fields of electromagnetic fields, communication systems/regulations, computational materials design, computational chemistry/biology, cryptography, signal/data processing/visualization, big data management, climate and ocean sciences, 

  • The paper titled “Non-Invasive Complete Hemodynamic Model to Investigate the Effect of Multi-Stenosis in Patient-Specific Coronary Arteries”, jointly prepared by research assistants Hacer Duzman and E. Cenk Ersan, andProf. Dr. M. Serdar Çelebi from the Computational Science and Engineering Program, has been awarded the Best Paper Award at the ESM’25. (22-24 Oct. 2025 Belgium)
  • Hacer Duzman and Muhammed Enis Şen, PhD students of Computational Science and Engineering Program, received awards in the "5 Minute Thesis Competition" organized at the "Başarım 2024 8th National High Performance Computing Conference".

There is also a High Performance Computing Laboratory established with the support of the State Planning Organization within the Institute.