Smart World Congress Tutorials
Inside Deep Learning: Methods, Technologies, Tools and Applications
Deep learning is about learning multiple levels of representation and abstraction that help to contextualize different types of data such as images, sound, and text.
As data amounts increase, traditional machine learning algorithms require heavy pre-processing, feature selection /engineering and fail to generalize. On the other hand,
the ability of deep neural networks to extract features from the raw data, has shown a significant performance increase in classification and regression tasks.
The purpose of the tutorials is to introduce researchers in the academia or in the industry the core of deep learning, the neurons and the backpropagation. Furthermore, the tutorials will focus on applications of deep learning algorithms in the fields of audio processing, computer vision and natural language processing. The speakers come from well-established research institutes with publications at venues such as NIPS, ICRA and distinctions in international competitions (Google Brain/TensorFlow Speech Recognition Challenge).
• Anastasios Vafeiadis, Center for Research & Technology Hellas - Information Technologies Institute, Thessaloniki, Greece
• Mohammad Reza Loghmani, Vision for Robotics laboratory - Vienna University of Technology (TUW), Vienna, Austria
• Erinc Merdivan, Austrian Institute of Technology – Austria, CentraleSupelec, France
Title: An Overview of Deep Learning
Duration: 1 hour
Introduction: There is a tremendous hype around deep learning. Deep learning started in the 1980s but it started becoming popular after 2010 with modern hardware (Graphical Processing Units) allowing it to be applied to industrial applications (e.g., speech recognition). In this talk, we will discuss the main idea behind deep learning, trying to understand what is inside the so-called “black box”. This talk aims to cut through the hype by focusing on how deep learning approaches are being successfully used across a range of disciplines, illustrated with real-world examples.
Speaker: Anastasios Vafeiadis
Anastasios holds a BSc in Electrical & Computer Engineering from Worcester Polytechnic Institute (WPI) and a MSc in Electrical & Computer Engineering from Northeastern University. He has worked as a Research Assistant at Toyota InfoTechnology Center, USA; conducting research regarding vehicular communications and Dynamic Spectrum Access (DSA). He has also worked as an Advance Development Engineer at Bose Corporation, developing algorithms for active noise cancellation and engine harmonics enhancement in vehicles with Continuous Variable Transmission (CVT) system. His research interests include digital signal processing, machine learning, deep learning and acoustics.
Title: Deep learning for Computer Vision
Duration: 1 hour
Introduction: Computer vision is one of the most innovative and fast-paced fields in computer science. Advancements in deep learning are often originated by innovative solutions in computer vision tasks that later echo in other fields. In this talk, we will cover some of the milestones of deep learning in computer vision and provide a working understanding of the standard tools that are omnipresent in current state-of-the-art algorithms. More specifically, we will dive into Convolutional Neural Networks (CNNs) and their applications for image classification, object detection and semantic segmentation. Finally, we will touch on Generative Adversarial Networks (GANs) and their impact on the computer vision community. This talk aims at providing a basic overview of deep learning tools for computer vision for beginner practitioners that are interested in immersing themselves into this exciting field.
Speaker: Mohammad Reza Loghmani
Mohammad Reza holds a BSc in Electronic & Telecommunication Engineering from University of Genova (Italy) and a double MSc in Robotics Engineering from University of Genova and École Centrale de Nantes (France). He has completed an internship as a junior researcher in the GV lab at Tokyo University of Agriculture and Technology, Tokyo, Japan; conducting research on emotion recognition from affordable sensors. He has also completed an internship as junior researcher in the VANDAL lab at the Italian Institute of Technology, Milan, Italy; conducting research on visual domain adaptation. His research interests include object recognition, visual domain adaptation and robotics.
Title:Deep learning for Sequential Modelling
Duration: 1 hour
Introduction: Sequence modelling covers sequential data where each sample is series of data. Sequence modelling covers a wide range of areas such as natural language, speech, video and sensors. Recent advances in RNNs and their most common variations LSTMs and GRUs showed great performance increase on sequence modelling applications. We will start our talk with RNNs then focus mainly on two special variations of RNNs, LSTM and GRU. Due to different modalities in sequence modelling we will cover how LSTMS and GRUs are applied to different modalities such as language translation and video processing. Finally, we will briefly go over recent mechanisms on language modelling which do not use RNNs but rely on Transformer architecture and how they can outperform RNNs on language modelling. This talk aims at providing basics of sequential modelling with deep learning models and how to combine different architectures of deep learning for modelling sequential data with varying modalities.
Speaker: Erinc Merdivan
Erinc Merdivan holds a BSc in Electronics Engineering and a MSc in Computer Science and Engineering from Sabanci University in Istanbul (Turkey). He worked on IBM Watson as Core Algorithms developer. He later joined to AIT as Marie Curie Fellow. He is currently working on deep learning and deep reinforcement learning for dialogue modelling, activity recognition and ambient assisted living.
Title: Blockchain: A Disruptive Solution for Building Consensus and Trust
Duration: 1 hour
Introduction: In this talk, Dr. Zhao will first give an overview of the blockchain technology as it is used in Bitcoin. Then he will examine the evolution of the blockchain technology, from the dream of a digital currency, to the creation of Bitcoin, to the smart contract based Ethereum, and a new crop of mechanisms for building distributed consensus and trust. He will conclude his talk by discussing the role played by randomization in reaching consensus and trust in an open untrusted environment, and the impact of asynchrony to the consensus process. Specific topics this talk will cover include the data structure of the blockchain, proof of work, proof of stake, proof of authority, forks and conflict resolution, attacks on blockchain, the FLP impossibility result for asynchronous distributed system, the Ben-Or randomized consensus algorithm, optimistic replication, and Byzantine fault tolerance.
Speaker: Wenbing Zhao
Dr. Zhao is a Professor at the Department of Electrical Engineering and Computer Science, Cleveland State University. He got his BS and MS degrees from the Physics Department in Peking University. He earned his Ph.D. at University of California, Santa Barbara in 2002. He has over 200 peer-reviewed publications. Dr. Zhao’s research spans from dependable distributed systems to human centered smart systems. His research has been funded by the US NSF, US Department of Transportation, Ohio Bureau of Workers’ Compensation, Ohio Department of Higher Education, the Ohio Development Services Agency, and Woodruff Foundation. He has delivered more than 10 keynotes, tutorials, public talks and demonstrations in various conferences, industry and academic venues. Dr. Zhao is an associate editor for IEEE Access, MDPI Computers, and PeerJ Computer Science, and a member of the editorial board of several international journals, including Applied System Innovation, Internal Journal of Parallel, Emergent and Distributed Systems, and International Journal of Distributed Systems and Technologies. He is currently an IEEE Senior Member and serves on the executive committee of the IEEE Cleveland Section.