I work on applied machine learning problems, in particular in the intersection of computer vision and machine learning: handwriting recognition (online, offline; English, Turkish), biometric verification (signature, fingerprint), biometric privacy and image understanding (face, plant…), attribute-based classifiers and NLP applications (sentiment analysis, information summarization).

Some of our recent work is listed below:

Recent Projects

• Ottoman document transcription – project page
• Hate Speech Detection – project page
• Face Attribute Understanding – project page
• Covid-19 Detection on CT Images – project page

Image Understanding/Deep Learning:

• Multimodal Deception Detection using Real-Life Trial Data
  M. U. Sen, V Perez-Rosas, B Yanikoglu, M Abouelenien, M Burzo, …
  IEEE Transactions on Affective Computing, 2021
• Skin Lesion Diagnosis with Imbalanced ECOC Ensembles
  SAA Ahmed, B Yanikoglu, C Zor, M Awais, J Kittler
  International Conference on Machine Learning, Optimization, and Data Science …
• Plant identification using deep neural networks via optimization of transfer learning parameters, MM Ghazi, B Yanikoglu, E Aptoula, Neurocomputing 235, 228-235 (2017) pdf
  • We explore the effects of increasing iteration, data augmentation or batch sizes on transfer learning performance.
• Open-set plant identification using an ensemble of deep convolutional neural networks, M Mehdipour Ghazi, B Yanıkoğlu, E Aptoula, CLEF Workshop (2016) pdf

   Face Expression/Attributes Understanding:

• Relative Attribute Classification with Deep-RankSVM
  Sara Atito Ali Ahmed, B. Yanikoglu, 2020
  ICPR International Workshops and Challenges 12662 (Lecture Notes in Computer …
• Within-Network Ensemble for Face Attributes Classification, Sara Atito Ali, B Yanikoglu, ICIAP (1) 2019: 466-476 pdf
  • We explore how to reap benefits of ensembles within a single deep network, by adding extra output layers from several earlier layers of one deep network.
• Multi-Label Networks for FaceAttributes Classification, Sara Atito Ali, B Yanikoglu, 2018
  IEEEInternational Conference on Multimedia & Expo Workshops (ICMEW), 1-6(2018) pdf
  • We explore indicating the location of a face attribute to the network. It increases training speed significantly and slightly improves accuracy.

Handwriting Recognition:

• Large vocabulary recognition for online Turkish handwriting with sublexical units, EF Bilgin Tasdemir, B Yanikoglu, Turkish Journal of Electrical Engineering and Computer Science 26 (5), 2218-2233 pdf
  • Turkish has a very large number of word forms due to its agglutinative nature. We explore how to deal with the large lexicon size in handwriting classification.
• A comparative study of delayed stroke handling approachesin online handwriting, EFB Tasdemir, B Yanikoglu, International Journal on Document Analysis and Recognition (IJDAR), 1-14 (2018) pdf
  • Delayed strokes cause the main source of variation in online handwriting recognition. This is the first study that evaluates different delayed stroke handling methods.

Signature Verification

• Score level fusion of classifiers in off-line signature verification, MB Yılmaz, B Yanıkoğlu, Information Fusion 32, 109-119 (2016)  pdf

Biometric Privacy/Template Protection

• Keystroke Dynamics Based Biometric Identification
  HA Boz, M Gürkan, B Yanıkoğlu
  2020 28th Signal Processing and Communications Applications Conference (SIU …
• Biometric layering with fingerprints: template security and privacy through multi-biometric template fusion, M Yildiz, B Yanikoğlu, A Kholmatov, A Kanak, U Uludağ, H Erdoğan, The Computer Journal 60 (4), 573-587 (2016), pdf
  • We propose layering multiple biometrics, and specifically multiple fingerprints to obtain a biometric template that can vary from application to application to become unlinkeable across different databases.

Machine Learning

• Document classification of SuDer Turkish news corpora, M U Şen, B Yanıkoğlu, 26th Signal Processing and Communications Applications Conference (SIU) 2018 pdf
• FLIP-ECOC: a greedy optimization of the ECOC matrix, C Zor, B Yanikoglu, T Windeatt, E Alpaydin, Computer and Information Sciences, 149-154 (2011) pdf
  • We optimize the ECOC matrix so as to better match the trained base classifiers, obtaining consistent improvements in several UCI problems.

Sentiment Analysis

• Sentiment analysis using domain-adaptation and sentence-based analysis, G Gezici, B Yanikoglu, D Tapucu, Y Saygın, Advances in Social Media Analysis, M. Gaber Eds. 45-64 pdf

var sc_project=11890835;
var sc_invisible=1;
var sc_security=”11237e86″;
var sc_https=1;