Tutorials

Dr. Hamada Naoki
Senior Researcher, Artificial Intelligence Laboratory
Fujitsu Laboratories Ltd., Japan

Title: Evolutionary Multi-objective Optimization and Topological Data Analysis for Designing Innovative Products

Abstract:
Evolutionary Computation (EC) and Machine Learning (ML) are excellent tools to model, analyze and optimize various kinds of industrial problems. Since real-world problems often involve conflicting objectives to be optimized (e.g., cost vs. performance), we need to consider a trade-off between objectives. While such a trade-off surface can be obtained by using Evolutionary Multi-objective Optimization (EMO) algorithms, the dimensionality of the surface becomes higher as the number of objectives grows. Thus after optimization, additional analysis and visualization are required for understanding the trade-off. This talk introduces an emerging ML technique to analyze EMO...

[Read More]

Mr. Sumit Misra
General Manager, RS Software (India) Limited, Kolkata, India
RS Software (India) Limited, Kolkata, India

Title: Applied Data Stream Analytics

Abstract: Business application of Data Stream Analytics opens up different types of challenges. The spectrum covers the infrastructure components as well as software solution components both for online and offline processes. While today we have a very large set of technology in AI, ML and related areas, selection of the techniques become extremely important for the business solution. As the businesses starts with deterministic methods – predominantly rule based, there is a specific need to draw a pathway from deterministic to non-deterministic models. An aspect that has to be under consideration is that the models and profiles “drift” with time as well as there are cyclic relevance of models based on seasonal activities. These and much more become...

[Read More]

Dr. Nibaran Das
Associate Professor, Department of Computer Science & Engineering
Jadavpur University, Kolkata, India

Title: Invariance: The Problem with CNN

Abstract: The insurgence of convolutional neural networks since the introduction of LeNet has been one of the most notable aspects in the field of computer vision over the last couple of decades. As more and more networks are proposed every year it is becoming more and more essential to pay attention to the shortcomings of CNN so that new avenues of research can be opened and saturation of technology can be avoided. The basis of CNNs lies in automatic learning of a set of weights arranged as kernels that are convoluted over an image to generate activations that denote the presence of a...

[Read More]