The list of the seminars that will be given under BBL 596E class is given below. Presentations will be performed in Class 409.
Student No: 704181008
Name Surname: Esra Ergün
Tittle: Feature Diversification on Sparse Progressive Neural Networks for Continual Learning
Abstract: Human brain effectively integrate prior knowledge with new skills by transferring experience across tasks without suffering from catastrophic forgetting. Performance of state-of-the-art neural networks are impressive on computer vision and natural language processing and they outperm humans on number of tasks. However, they perform modest on knowledge transfer and continual learning. The goal of this work is to address memory overhead of existing methods and to improve forward transfer through tasks with feature diversification. This study combines sparsity and additional loss terms for feature diversification on progressive neural networks [0] to continually learn multiple tasks. This approach will be evaluated on permuted MNIST, class MNIST, and class CIFAR10 datasets.
Advisor Name Surname: Behçet Uğur Töreyin
Date: 26.11.2019
Hour: 09.30
Student No: 704181010
Name Surname: Gözde Filiz
Tittle: Reordering Buffer Management Problem
Abstract: In the Reordering Buffer Management (RBM) Problem, we are given a service station and a random-access buffer with a limited capacity. An input sequence of items which are characterized by a specific attribute has to be processed by the service station which benefits from consecutive items with the same attribute value. We use the buffer to minimize the cost of service station i.e., the output sequence has maximal subsequences of items with the same attribute. This problem has many applications in computer science and economics. In this seminar, we will visit many variants of RBM problem, introduce the algorithms that produced and discuss the complexity analysis.
Advisor Name Surname: Muhammed Oğuzhan Külekçi
Date: 26.11.2019
Hour: 10.15
Student No: 704181005
Name Surname: Behnaz Ghaderkalankesh
Tittle: Can text mining be an antidote for radiology error?
Abstract: Radiology reports contain detailed information about patients’ health status. This critical patient health data is recorded as free text format. Radiology reports differ based on the experiences of the radiologists and, the normative information given by the medical school. Especially, the accuracy rate in radiology reports of cancer patients is a critical issue. We want to develop an unsupervised machine learning model to understand the main concepts in a radiology report based on their similarity. In this model, we collect cancer diagnosed reports of head, neck and abdomen region. Based on the model, we try to understand which classification method is more effective in classifying low sample sized radiology reports. All in all, we aimed to implement and train a model that can be easily utilized for rare disease and cancers that normally a sufficient amount of biomedical information and database is not exist.
Advisor Name Surname: Sefer Baday
Date: 26.11.2019
Hour: 11.00