Vural Erdogan

• Generative AI solutions provided using anycostgan, interfacegan, re-style, hyperstyle, Stylegan2, Stylegan3 in order to create hyperrealistic synthetic human images. GANs trained with own dataset using own local (3090 RTX) and cloud GPUs.
• LipSync, removing image's background and background replacing (image composition) models have been researched and implemented.
• Demonstrated hyperrealistic image/face editing, converted images with given styles according to business needs such as aging, smiling, head pose, de-aging etc.

• Collaborated with researchers from RAF, DSTL and +20 years experienced academicians from top tier universities; University of Oxford, Imperial College London.
• Worked on Computer Vision and other type of Deep Learning problems. Provided unique solutions and solved choke points of the projects.
• Security Check is available upon request.

• Machine Learning algorithms; Random Forest, Naive Bayes, etc. and Natural Language Processing pre-trained models; BERT, GPT-2, etc. are examined and implemented for workflow automation of client companies.
• Reinforcement Learning and Online Learning algorithms (Hoeffding trees, Learn++, etc.) investigated and experimented to create decision path optimiser for business process mining.
• Recommender system with sentiment analysis improved applying unsupervised machine learning techniques for new product suggestion.
• Financial sentiment analysis implemented which scrapes websites (HTML) to catch trends depending on daily news and numerical data. Project initialised from scratch and deployed as REST API using Flask.

• Guided on Multiple Object Detection, Pose Estimation, ReID and Tracking models. Increased accuracy, speed, and decreased occlusions during overhead detection of real-time surveillance system.
• Yolov3, Faster R-CNN, Mobile SSD Net type of pre-trained models experimented and analysed.
• Created mAP metrics and annotated data-sets from scratch using open-source tools. Rendered videos utilising Google Cloud computing on Ubuntu servers.
• Collaborated with global team of engineers working entirely remotely.

• Worked with imbalanced, limited seismic datasets and cloud computing (AWS) to create generalised deep learning models.
• Specialised in 2D and 3D image segmentation.
• Created deep learning models from scratch with own customised network architectures, loss functions, metrics and workflows in order to address specific problems in seismic interpretation.
• Improved projects' accuracies (Matthew and Jaccard scores) and increased speed x10, x20 faster by reducing concatenated layers.

• Attended project meetings, summarised project results and informed research groups.
• Conducted literature review to enhance existing circuit designs. Reproduced published papers in electronic field.