CmpE Events


  1. CmpE Seminar: Unsupervised Feature Extraction with Competitive CNNs
    • Start time: 03:00pm, Monday, December 18th
    • End time: 04:00pm, Monday, December 18th
    • Where: AVS Conference Room, BM
    • Speaker: Taner Eskil, PhDTaner Eskil received his BSc in Mechanical Engineering andMSc in Systems and Control Engineering from Boğaziçi University. During his MScstudies he was also employed as a research assistant at Boğaziçi UniversityPattern Analysis and Machine Vision Laboratory (BUPAM). He completed his PhD inComputer Science at Michigan State University in 2005. His thesis subject wasmultiple routine design and simulation using off-the-shelf components that aredistributed over the Internet. He then joined Sabancı University as apost-doctoral fellow where he lead the Vision and Pattern Analysis Laboratorywith funding from EU 6th framework. In 2006, he was appointed by IşıkUniversity Computer Engineering Department as a faculty member. Taner Eskil completed one joint TÜBİTAK ARDEB and EU COSTproject and 2 scientific research projects internally funded by IşıkUniversity. He founded Pattern Recognition and Machine Intelligence Laboratoryfrom which 1 PhD and 8 MSc students graduated to date. His researchconcentration was facial expression recognition until he received his AssociateProfessorship in 2015. Since then, he has been studying unsupervised featureextraction with multi layered neural networks. Abstract:The most critical stage in machine learning with ConvolutionalNeural Networks (CNNs) is the algorithmic training of the filters in the hiddenlayers. Early approaches such as Fukushima’s Neocognitrons and Hinton’s multilayer generative models focused solely on the inputs to extract a set ofrepresentative features. This is intuitive considering the early stages ofhuman development, when an infant tries to make sense of the environment bylearning the modes of variation in his/her sensory inputs. State of the art inCNN studies however revolve around supervised training through gradient basedalgorithms. In gradient based learning (1) the success depends on randominitialization of both the number of neurons and their weights, (2) learningstages are vulnerable to the credit assignment problem and (3) training is slowas it requires numerous epochs on samples.I propose add-vote, an unsupervised feature extractionalgorithm that has its roots in adaptive resonance theory and Neocognitrons.Add-vote completely eliminates the random initialization stage, avoidspropagation of updates in the network, and is extremely fast, converging inonly a few epochs in most cases. I will introduce the algorithm and present ourearly results on 3 different tasks; MNIST handwritten digits data set, BRATSmulti modal brain tumor segmentation challenge and textile defectdetection.

    • View this event in Google Calendar

Contact us

Department of Computer Engineering, Boğaziçi University,
34342 Bebek, Istanbul, Turkey

  • Phone: +90 212 359 45 23/24
  • Fax: +90 212 2872461

Connect with us

We're on Social Networks. Follow us & get in touch.