Practical Applications of Deep Learning (Sommersemester 2022)

Lecturer: Dr. Haojin Yang (Internet-Technologien und -Systeme) , Joseph Bethge (Internet-Technologien und -Systeme) , Ziyun Li (Internet-Technologien und -Systeme) , Hendrik Rätz (Internet-Technologien und -Systeme) , Jona Otholt (Internet-Technologien und -Systeme) , Gregor Nickel

General Information

• Weekly Hours: 4
• Credits: 6
• Graded: yes
• Enrolment Deadline: 01.04.2022 - 30.04.2022
• Examination time §9 (4) BAMA-O: 25.07.2022
• Teaching Form: Seminar
• Enrolment Type: Compulsory Elective Module
• Course Language: English
• Maximum number of participants: 15

Programs, Module Groups & Modules

Description

Artificial intelligence (AI) is the intelligence exhibited by computer. This term is applied when a machine mimics "cognitive" functions that humans associate with other human minds, such as "learning" and "problem solving". Currently researchers and developers in this field are making efforts to AI and machine learning algorithms which intend to train the computer to mimic some human skills such as "reading", "listening", "writing" and "making inference" etc. From the year 2006 "Deep Learning" (DL) has attracted more and more attentions in both academia and industry. Deep learning or deep neural networks is a branch of machine learning based on a set of algorithms that attempt to learn representations of data and model their high level abstractions. In a deep network, there are multiple so-called "neural layers" between the input and output. The algorithm is allowed to use those layers to learn higher abstraction, composed of multiple linear and non-linear transformations. Recently DL gives us break-record results in many novel areas as e.g., beating human in strategic game systems like Go (Google’s AlphaGo), self-driving cars, achieving dermatologist-level classification of skin cancer etc. In our current research we focus on video analysis and multimedia information retrieval (MIR) by using Deep-Learning techniques.

Course language: English and German

Topics in this seminar:

• Binary Neural Networks as Recommender Systems: On the one hand, Binary Neural Networks (BNNs) are an energy-efficient approach to run Neural Networks on devices with low computational power or applications which have real-time requirements. They have been applied to traditional Convolution Neural Networks to convert them into efficient, but slightly less accurate alternatives. On the other hand, recommender systems are widely used in real-world appplications and in the industry, being run millions of times per day, requiring overall large amounts of computation and energy. In this topic, you are going to extend a previous seminar work which introduces Binary Deep Learning Ranking Models ( the source can be found in the HPI Gitlab, it is build on BITorch, a framework for BNNs, developed by our chair and written in PyTorch). The goal is to train and evaluate more accurate and more efficient BNN models for challenging large-scale ranking tasks and provide accurate time measurements for practical scenarios.
• Document Objection Detection: The field of document object detection (DOD) aims to identify high-level semantic regions within given documents, such as paragraphs, tables, images, and equations. Using these annotated regions, archivists can filter documents based on semantics and thus locate desired information more efficiently. Additionally, DOD is a necessary preprocessing step for downstream tasks, e.g., more fine-grained image analysis. Existing approaches usually rely on supervised learning and generate satisfying results if an annotated dataset can be used for training. However, applying these methods to a different, unlabeled dataset is not trivial because the unseen documents may differ in language, layout, or even genre. In this seminar, we will evaluate methods to bridge the domain gap between different document datasets to identify objects in our art-historical dataset. 
• Long-tailed Class Discovery (LTCD): In generalized novel class discovery, given a labeled and an unlabeled set, the goal is to categorize the unlabeled set. The unlabeled samples may come from labeled classes or unknown (novel) ones. Existing recognition methods assume that the sample proportions of known and novel classes are balanced in both labeled and unlabeled sets. However, real-world data distributions are usually long-tail and open-ended. Specifically, the known classes typically are in the head classes, and the novel classes are discovered in the tail classes, with small sample size. Tail recognition robustness and new class sensitivity are the main challenges for LTCD. Our goal is to develop a method to differentiate between known and novel classes, generalize from a few samples, and recognize novel classes when they appear.

Requirements

• Strong interests in video/image processing, machine learning (Deep Learning) and/or computer vision
• Software development in C/C++ or Python
• Experience with OpenCV and machine learning applications as a plus

Literature

Books

• Alex Smola, Mu Li et al., Dive into deep learning
• Ian J. Goodfellow and Yoshua Bengio and Aaron Courville, "Deep Learning", online version
• Christopher M. Bishop “Pattern Recognition and Machine Learning” google it

Deep Learning frameworks:

• PyTorch

Examination

The final evaluation will be based on:

• Initial implementation / idea presentation, 10%

• Final presentation, 20%

• Report (research paper), 12-18 pages, 30%

• Implementation , 40%

• Participation in the seminar (bonus)

Dates

(apart from the presentations, there will be no regular meetings in our seminar room!)

25.04.2022 (15:15-16:30), Room: A 2.1 Zoom link here, code: R0ZG9V	Introduction and QA session for seminar topics. (Hybrid) [Slides]
until 29.04.2022	Registration: HPI students: via moodle Students from UP: an Studienreferat (Studienreferat(at)hpi.uni-potsdam.de) Sending your preferred and secondary topic to: haojin.yang@hpi.de
until 03.05.2022	Topics and Teams finalized Arranging individual meeting
weekly	Individual meeting with your tutor
30.05.2022 (15:15-16:30), Room: A 2.1 Zoom link here, code: R0ZG9V	Mid-Term presentation (15+5min),
25.07.2022 (15:15-16:30), Room: A 2.1 Zoom link here, code: R0ZG9V	Final presentation (15+5min)
31.08.2022	Hand in code and paper (Latex template)
until 30.09.2022	Grading finished

Zurück