Here is a list of my courses. Scroll down for course briefs.
- Data Mining for Visual Media (Bogazici University EE Dept., latest edition: Spring 2017, since 2010)
- Computer Vision (Spring 2015)
- Image Analysis (Bogazici University EE Dept., Spring 2012)
- The Digital Shape or Mind the Gap Reloaded (as part of the Digital Architectural Design Studio, Istanbul Technical University, Architectural Design Program, Spring 2011)
- Data Mining for Architecture and Urban Plannning (as part of the Digital Architectural Design Studio, Istanbul Technical University, Architectural Design Program, Spring 2010)
Data Mining for Visual Media (Latest Edition: Spring 2017)
This graduate-level course presents an integrated perspective of data mining for applications where data arise from visual media. Production and sharing of digital visual data have never been as easy as today thanks to recent technological developments. As humans, we are very good at performing tasks that involve visual processing: We can efficiently organize images when they are a handful, we can accurately identify objects and patterns, or we can easily describe the scene depicted in a video. However, concurrent to the increase in the amount of visual data, its manipulation for the above purposes is becoming harder, more time-consuming, and consequently frustrating for humans. Visual media also involve less easily interpretable modalities such as satellite, multispectral and medical imagery. Data arising in these modalities convey extremely rich information that can be very useful in several applications ranging from urban planning to medical diagnosis. Since all these intelligent tasks require at the outset that raw pixel information be transformed into useful information, visual media constitutes an ideal niche for data mining.
The pedagogical objective of this course is to provide the students with the basic theoretical knowledge and the necessary computing skills related to data mining with a particular focus on visual media. Accordingly, we exclusively deal with problems that involve images, video, or 3D models. The methodological bulk of the course consists of two main parts: visual content description techniques and machine learning methods. The course is illustrated with several state-of-the-art techniques and applications from the real world. Students are given weekly computer assignments and a term project so that they can gain hands-on experience in the domain. The course also involves case studies and reading research papers.
Lecture Notes 2017
- Lecture 1. Basic Concepts and Tasks in Data Mining, Applications for Visual Media, Visual Recognition: Problems & Architectures - updated on February 10, 2017
- Lecture 2. Visual Features: Description, Levels of Visual Description, Taxonomy of Visual Descriptors, Algorithm Examples: SIFT, SURF, HOG, LBP, PCA - updated on February 17, 2017
- Lecture 3. Visual Features: Detection - updated on March 19, 2017
- Lecture 4. Visual Features: Matching - updated on March 31, 2017
- Lecture 5. Visual Recognition: History, Models, and Architectures - updated on April 06, 2017
- Lecture 6. Intro to Machine Learning: Classification as Function Approximation, Basics of Statistical Decision Theory, The Bayes Classifier, Naive Bayes, Linear & Quadratic Classifiers - updated on May 13, 2017
- Lecture 7. Polynomial Classifier, Bias-Variance Trade-off and Model Complexity, Model Selection: Cross-Validation, How to Train a Classifier for Visual Recognition Tasks? - updated on May 13, 2017
- Lecture 8a and Lecture 8b. Support Vector Machines - updated on May 13, 2017
- Lecture 9a and Lecture 9b. Boosting, Feature Selection - updated on May 13, 2017
- Lecture 10. Intro to Deep Learning, Convolutional Neural Nets - updated on May 13, 2017
Lecture Notes (Previous
- Lecture I. Basic Concepts and Tasks in Data Mining, Applications for Visual Media - updated on February 26, 2015
- Lecture II. Visual Data Description: MPEG-7 Visual Descriptors - updated on February 26, 2015
- Lecture III. Visual Data Description: Invariance, Interest Point Detection - updated on February 26, 2015
- Lecture IV. Visual Data Description: SIFT, Bag-of-Words Model, Shape Contexts, Textons, Subspace Methods - updated on March 06, 2015
- Lecture V. Introduction to Machine Learning: Regression and Classification as Function Approximation, Basics of Statistical Decision Theory, Linear Methods for Regression and Classification.
- Lecture VI. Polynomial Regression, Bias-Variance Trade-off and Model Complexity, Cross-validation, Logistic Regression, Naive Bayes Classifier.
- Lecture VII. Support Vector Machines. [ Constrained Optimization ] [ Tips & Tricks ] [ Andrew Tutorial on SVM (adapted) ] - updated on April 18, 2013
- Lecture VIII. Boosting - updated on May 02, 2013
- Lecture IX. Feature Selection - updated on May 02, 2013
Course Time and Venue: Spring 2017: Fri 1200-1500, Shannon Room, KB 2nd Floor, Bogazici University EE Dept.
Computer Vision (Latest Edition: Spring 2015)
This graduate-level course offers a sampling of computer vision topics from the basics of image formation, point operators and image filtering, visual features and matching, stereo vision, segmentation and grouping, and visual recognition. Lectures notes are progressively published below.
- Lecture I. Introduction: What Is Computer Vision? What Are The Real World Applications? - updated on February 11, 2015
- Lecture II. Image Formation, Light and Color - updated on March 04, 2015
- Lecture III. Linear Filters - updated on March 11, 2015
- Lecture IV. Template Matching, Image Pyramids - updated on March 20, 2015
- Lecture V. Edge and Boundary Detection - updated on March 25, 2015
- Lecture VI. Interest Points, Hough Transform - updated on April 15, 2015
- Lecture VII. Local Image Descriptors - updated on April 15, 2015
- Lecture VIII. Feature Matching - updated on April 22, 2015
- Lecture IX. Recognition 01 - updated on April 29, 2015
- Lecture X. Recognition 02 - updated on May 06, 2015
The Digital Shape or Mind the Gap Reloaded (Spring 2011)
The main goal of this course is to provide architecture students with a theoretical and practical knowledge on shape and its many possible geometries by means of computational techniques. The course will flow in two phases in which students will develop a deeper understanding and skills on the concept of shape, digital shape representation, and computational shape generation. Phase 1 will consist of lectures and assignments focusing on the core concepts and techniques, while in Phase 2, students will be expected to develop design projects.
In Phase 1, concurrent to a series of lectures, students will be given weekly assignments in which they will experiment with digital shape representation and computational shape generation techniques. The core studio part is in the Phase 2 and consists of a series of digital shape design experiments with a particular focus on the exploration of volumetric potentials to fill in the urban gaps. The outputs from Phase 1 provide the motivation, the conceptual background, and associated techniques for Phase 2, the Design Studio.
In Phase 2, a series of urban plots standing as gaps between buildings will be introduced. Students will be working on these urban gaps by exploring the existing volumetric potentials of the gaps such as boundaries, openings and closings, obstacles. By defining their individual approaches in terms of sensitivity for the volumetric potentials of the gaps, each student will be producing a series of shapes. Production of architectural spaces within these urban gaps is not in the scope of the design studio but the aim is to run a series of experiments in the production of digital shape by means of external inputs. Hence the urban gap is treated not as a building plot but an empty volume with various constraints and possibilities, as inputs to the shape generation process.
Go to course blog.
Data Mining for Architecture and Urban Planning (Spring 2010)
We designed this course with Ahu Sokmenoglu, M. Arch, as a graduate-level digital architectural design studio course aiming to explore the patterns and trends of socio-spatial activities of architecture student community in Istanbul. The subject of the studio is to design a thematic city map of activity patterns and trends for different groups of architecture students. Our starting process will be to collect raw data on the daily activities of students through dedicated questionnaires, surveys, and interviews. We will then use data mining methodologies in order to understand "what this specific data say" about questions related to socio-spatial activities. When applied to discover relationships between spatial and non-spatial attributes, we believe that data mining can support the analysis of cities and lead to novel opportunities in the formulation of new design approaches.
Our main motivation behind this studio is that architects and urban designers can profit a lot from a deeper understanding of individual activities occurring in urban space so that innovative urban design solutions can be initiated. Cities, how we think of them, and how we can work strategically to guide their development require deeper analysis in understanding the social and spatial patterns of movements performed in every-day life. What we propose here is an analysis of the urban space use at the level of individuals by evaluating their social characteristics and preferences.
We have two main pedagocical goals in this studio. One is to clearly define what data mining is and to give the student a feeling about what it is capable of, as well as is not, with several examples from a broad spectrum of applications. Our second goal is to use data mining on a specific socio-spatial analysis problem.
Go to course blog.