گالری

Keynote Speakers

  • Amir Hossein Payberah
  • Assistant Professor of Computer Science
  • KTH Royal Institute of Technology - Google-Scholar

  • Lecture title: Generalized Reinforcement Learning for Gameplay
  • Lecturer biography:
  • Amir H. Payberah is an assistant professor (docent) of computer science at the division of Software and Computer System (SCS) of KTH Royal Institute of Technology in Sweden. He is also a member of the Distributed Computing at KTH (DC@KTH) and the Center on Advanced Software Technology Research (CASTOR). Prior to that, Amir was a machine learning scientist at University of Oxford (2017-2018), and a senior researcher at the Swedish Institute of Computer Science (2013-2017). Amir got his PhD from KTH Royal Institute of Technology in June 2013.

  • Location: Online presentation in AUT Central Amphitheater Hall, or can participate via https://bluemeet.aut.ac.ir/ch/aaic_lectures/guest from anywhere. The lecture is open to everyone.
  • Date: May 18 2022, 11:00:00
Abstract:

Reinforcement Learning (RL) is becoming ever more prevalent in game development. However, there exist many challenges to overcome in order to use RL to its full potential. For example, an RL agent trained in one game environment cannot easily generalize to replicate the same level of performance in new game environments with different levels and variations. In this talk, we present a generalized solution for match-3 games, such as Candy Crush Friends Saga (CCFS). Our solution is a two-step process, inspired by human behavior in approaching such games: (i) learning basic skills while progressing through the levels, and (ii) combining and reusing them in different game states. These skills are not necessarily related to a level’s objective, but their execution can indirectly help the player win the level. Here, we propose various basic skills for CCFS with intrinsic rewards, which do not necessarily have the same objective as the game level. We show that an agent trained with intrinsic rewards outperforms the agents that are trained with extrinsic rewards, despite not knowing how to win a level. Moreover, we show that by making a hybrid model and combining these basic skills, the agent can significantly outperform the baselines, winning more than twice as much as an agent trained with extrinsic rewards.

  • Nicolas Gillis
  • Professor
  • University of Mons - Google-Scholar

  • Lecture title: Learning with Nonnegative Matrix Factorizations
  • Lecturer biography:
  • Nicolas Gillis is a professor in the department of Mathematics and Operational Research at the University of Mons in Belgium. He is a recipient of the Householder Award and an ERC Starting Grant. His research interests include optimization, numerical linear algebra, machine learning, signal processing, and data mining. He serves as an associate editor of the SIAM Journal on Matrix Analysis and Applications, the SIAM Journal on Mathematics of Data Science, and the IEEE Transactions on Signal Processing. See https://sites.google.com/site/nicolasgillis/ for more detail.

  • Location: Online presentation in AUT Central Amphitheater Hall, or can participate via https://bluemeet.aut.ac.ir/ch/aaic_lectures/guest from anywhere. The lecture is open to everyone.
  • Date: May 18 2022, 13:30:00
Abstract:

Given a nonnegative matrix X and a factorization rank r, nonnegative matrix factorization (NMF) approximates the matrix X as the product of a nonnegative matrix W with r columns and a nonnegative matrix H with r rows such that WH approximates X as well as possible. NMF has become a standard linear dimensionality reduction technique in data mining and machine learning. In this talk, we first review the motivations behind NMF and several applications, including topic modeling, hyperspectral imaging, and facial feature extraction.

Then, we present two recent contributions about NMF. The first one is about the uniqueness of NMF decompositions, also known as the identifiability, which is crucial in many applications. We provide a new model and algorithm based on sparsity assumptions that guarantee the uniqueness of the NMF decomposition. The second contribution is the generalization of NMF to non-linear models. We consider the linear-quadratic NMF (LQ-NMF) model that adds as basis elements the component-wise product of the columns of W, that is, W(:,j).*W(:,k) for all j,k where .* is the component-wise product. We show that LQ-NMF can be solved in polynomial time, even in the presence of noise, under the separability assumption which requires the presence of the columns of W as columns of X. We illustrate these new results on the blind unmixing of hyperspectral images. This is joint work with Maryam Abdolali, Christophe Kervazo and Nicolas Dobigeon.

  • Maryam Abdolali
  • PhD in Artificial Intelligence
  • Department of Mathematics and Operational Research at University of Mons - Google-Scholar

  • Lecture title: Subspace Clustering: Algorithms, Applications and Challenges
  • Lecturer biography:
  • Maryam Abdolali received her PhD and MSc in Artificial Intelligence from Amirkabir University of Technology in 2019 and 2013, respectively. Currently, she is a postdoctoral researcher in University of Mons, Belgium. Her research interests include subspace clustering, low-rank matrix approximation and sparse representation.

  • Location: Online presentation in AUT Central Amphitheater Hall, or can participate via https://bluemeet.aut.ac.ir/ch/aaic_lectures/guest from anywhere. The lecture is open to everyone.
  • Date: May 18 2022, 14:45:00
Abstract:

Data clustering is one of the fundamental tasks in unsupervised machine learning. Among the many clustering methodologies, subspace clustering has been established as a major approach for grouping samples according to their underlying subspaces. Subspace clustering is based on the assumption that the samples approximately lie on several low-dimensional subspaces. In the past two decades, a wide variety of algorithms have been proposed to tackle subspace clustering. As a vital tool in understanding and processing high-dimensional data, subspace clustering has applications in many domains including machine learning, signal processing and computer vision. In this talk, we will focus on the state-of-the-art algorithms for subspace clustering and the existing challenges in developing efficient subspace clustering algorithms for the real-world data.

  • Fatemeh Torabi Asr
  • PhD in Artificial Intelligence
  • Postdoctoral Researcher @ Discourse Processing Lab, Simon Fraser University - Google-Scholar

  • Lecture title: Fake News Detection: Methods and Challenges
  • Lecturer biography:
  • Fatemeh Torabi Asr studie Computer Science at Shiraz University and pursued a PhD with specialization in Computational Linguistics at Saarland University, Germany. She has explored a variety of research topics in this area and continued her academic work as a postdoctoral researcher at the Cognitive Computing Lab, Indiana University, USA, and the Discourse Processing Lab, Simon Fraser University, Canada. Her most recent research has been focused on the topics of Fake News Detection and Gender Gap Tracking in Media. She is currently working in industry on several applications of Natural Language Processing such as communication data analysis and consumer classification.

  • Location: Online presentation in AUT Central Amphitheater Hall, or can participate via https://bluemeet.aut.ac.ir/ch/aaic_lectures/guest from anywhere. The lecture is open to everyone.
  • Date: May 18 2022, 16:30:00
Abstract:

While hoaxes and scandals have always existed, sharing them online has become a dominant feature of social media. With no central control over online content, websites have emerged that publish a mixture of factual and misleading information, and often have relatively high traffic through referrals from social media sites such as Facebook. Given the speed and scale of fake news broadcasts, insightful human fact checking (the classic approach) is not always a feasible solution; therefore automatic approaches using machine learning and natural language processing have become essential. One challenge, however, regarding the automated detection of fake news is low accuracy. Fake news can have very similar features to their legitimate equivalents; and these similarities can easily mislead uninformed readers as well as automated systems. We will explain the extent to which a combination of techniques from discourse analysis, psycholinguistics and machine learning can be employed to differentiate fake from genuine content and what still remains unsolved.

  • Mahdi Javanmardi
  • Assistant Professor
  • Amirkabir University of Technology - Google-Scholar

  • Lecture title: Map-based Localization for Autonomous Vehicles
  • Lecturer biography:
  • Mahdi Javanmardi is an assistant professor in the Department of Computer Engineering at the Amirkabir University of Technology. Before joining AUT, he was a postdoctoral researcher at the Institute of Industrial Science, the University of Tokyo, hosted by Shunsuke Kamijo. Mahdi received his M.Sc. degree from the Sharif University of Technology, Iran, in 2013, and his Ph.D. degree in information and communication engineering from the University of Tokyo, Japan, in 2017. He was a visiting researcher with the California PATH, University of California, Berkeley, from 2016 to 2017. His research interests include localization and mapping for an autonomous vehicle, autonomous vehicle perception, and computer vision.

  • Location: Oral presentation in AUT Central Amphitheater Hall, or can participate via https://bluemeet.aut.ac.ir/ch/aaic_lectures/guest from anywhere. The lecture is open to everyone.
  • Date: May 19 2022, 09:15:00
Abstract:

بیش از یک قرن پیش، کارل بنز با تولید اولین نسل از خودرو تحول بزرگی در نحوه جابجایی انسان به‌وجود آورد. با گذشته بیش از یک قرن از آن روز، امروزه خودروهای خودران در صدد ایجاد انقلابی مشابه در زیست و حمل و نقل بشر می‌باشند. سالیانه بیش از یک میلیون نفر به دلیل حوادث رانندگی در دنیا جان خود را از دست می‌دهند، این در حالی است که ظهور خودروهای هوشمند و خودران می‌تواند با کاهش خطاهای انسانی تعداد این حوادث را به حداقل برساند. نقشه‌های سه بعدی (HD Map) یکی از ارکان اصلی رانندگی خودران است که به خودرو این امکان را می‌دهد تا در هر لحظه مکان خود را با دقت بالایی محاسبه کرده و همچنین ادراک محیط و برنامه‌ریزی مسیر بهتری داشته باشد. در این سخنرانی قصد داریم ضمن معرفی اجزای فناوری خودروی خودران و سطوح مختلف آن، به معرفی کاربرد نقشه‌های سه بعدی در مکان‌یابی دقیق خودرو بپردازیم.

  • Ahmad Nickabadi
  • Assistant Professor
  • Amirkabir University of Technology - Google-Scholar

  • Lecture title: Semantic Facial Attribute Editing Using Generative Models
  • Lecturer biography:
  • Ahmad Nickabadi received the B.S. degree in computer engineering and the M.S. and Ph.D. degrees in artificial intelligence from the Amirkabir University of Technology (AUT), Tehran, Iran, in 2004, 2006, and 2011, respectively. Since 2012, he has been an Assistant Professor with the Computer Engineering Department, AUT. His research interests include the analysis of image and video content using deep learning and probabilistic graphical models with a special focus on activity recognition, face recognition, and face synthesis.

  • Location: Oral presentation in AUT Central Amphitheater Hall, or can participate via https://bluemeet.aut.ac.ir/ch/aaic_lectures/guest from anywhere. The lecture is open to everyone.
  • Date: May 19 2022, 11:00:00
Abstract:

Facial attribute editing (FAE) is the task of manipulating single or multiple attributes of a face image while the other attributes remain untouched.  FAE has applications in many areas including data augmentation in facial image processing tasks, adding effects to facial images in the social networks, improving the performance of automatic face recognition systems, creating and animating faces in game and animation industries. A human face can be digitally represented in many forms including 2D images of grayscale or RGB pixels, parametric or nonparametric 3D models, low-level feature vectors extracted by image processing models, and high-level descriptors given as the values of some facial attributes (e.g. gender, age, hair color, and face shape). FAE may happen at any of these representation levels. For example, while it is possible to make a face look older by manually adding some wrinkles to the face image in a drawing software or application, it is also possible to do this by requesting a high-level semantic facial attribute manipulation model to just increase the age. The focus of this presentation is on the semantic facial attribute editing models in which the required changes are given as the values of the target attributes (e.g. “the hair color should be brown”) or in the form of a driving face image with the desired attribute values and the whole image modification task is then performed by the model without any user intervention.