Held for the 3rd year, exchange of the latest research trends and strategies
On-site demonstration of on-device AI interpretation technology, etc.
A poster session for outstanding domestic papers has also been organized.


Samsung Electronics will hold the 'Samsung AI Forum 2019' from the 4th to the 5th.

The first day of the Samsung AI Forum 2019 will be held at the Samsung Electronics Seocho Building, hosted by the Samsung Electronics Advanced Institute of Technology, and the second day will be held at the Samsung Electronics Seoul R&D Campus, hosted by Samsung Research.

The Samsung AI Forum, now in its third year, is a venue that invites world-renowned AI scholars to share the latest research trends and explore future innovation strategies.

This year, approximately 1,700 people, including AI experts, professors, and students, are expected to attend.

In his opening remarks, Samsung Electronics Executive Vice Chairman Kim Ki-nam stated, “AI technology is already having a widespread impact on society as a whole,” adding, “I hope that through this forum, we can present the future direction of AI technology development and contemplate strategies that can benefit the world.”


Day 1 of Evolved and Expanded AI Technology Announcements
On the first day of the forum, leading authorities in the field of deep learning participated and presented research achievements in evolved and expanded AI technologies, such as deep learning-based world understanding and autonomous systems.br />
Speakers on the first day of the forum included Professor Yoshua Bengio of the University of Montreal, Canada; Professor Trevor Darrell of UC Berkeley, USA; Professor Kyunghyun Cho of New York University, USA; and Professor Simon Lacoste-Julien of the University of Montreal, Canada, who is also the head of the Montreal AI Lab at Samsung Electronics Advanced Institute of Technology.

At a seminar titled 'Towards Compositional Understanding of the World by Deep Learning,' Professor Bengio proposed key technologies in the field of deep learning, such as meta-learning and reinforcement learning, that enable AI agents to understand the world, much like a young child comes to understand the world through experience. Meta-learning is a method of training a model that can effectively solve new problems even when they are given.

Professor Darrell gave a lecture on the topic of 'Adapting and Explaining Deep Learning for Autonomous Systems'. He introduced the latest research trends on technology that enables systems, such as autonomous vehicles that plan and operate based on sensor information, to analyze and make decisions on their own when faced with unexpected and complex situations.

At a seminar titled 'Three Flavors of Neural Sequence Generation,' Professor Cho Kyung-hyun proposed techniques to improve sentence generation speed during machine translation, such as parallel decoding and insertion-based generation.

Professor Lacoste-Julien gave a presentation on the topic of "New Perspectives on Generative Adversarial Networks." Generative Adversarial Networks are deep learning algorithms that generate and learn data through competition between a Generator and a Discriminator. He also presented reasons why Generative Adversarial Networks are suitable for high-dimensional data such as images and introduced optimization theories that can resolve the difficulties of training Generative Adversarial Networks.

Samsung Electronics demonstrated deep learning technology on-site. While machine translation was demonstrated in 2017 and end-to-end speech recognition technology in 2018, this year, 'On-Device AI interpretation technology' was introduced, which performs AI functions on the device itself without going through a server.


Day 2 to present distinctive and diverse AI technologies
On the second day, hosted by Samsung Research, you can enjoy various lectures by AI experts from around the world. The keynote speeches will be delivered by Professor Noah Smith of the University of Washington and Professor Abhinav Gupta of Carnegie Mellon University.

In his keynote address, Professor Smith will present on the topic of "Rational Recurrences for Empirical Natural Language Processing." We will explain the vulnerabilities of existing natural language processing deep learning models and introduce a rational Recurrent Neural Network (RNN) as a new model capable of deep language processing while also being capable of linguistic understanding.

In his lecture "Supersizing and Empowering Visual and Robot Learning," Professor Gupta proposes a large-scale self-visual learning method that overcomes the limitations of existing supervised learning and shares strategies for applying it to future agents.

Professor Vaishak Belle of the University of Edinburgh, UK, introduces a method for integrating human knowledge and data-based learning to apply AI to a wider range of fields in people's daily lives.

Following this, Professor Joan Bruna of New York University will present research findings on Graph Neural Networks (GNNs), which are recognized as a breakthrough technology for the realization of Artificial General Intelligence (AGI), as well as her views on various applications such as personalization.

In the afternoon, two tracks themed 'Vision & Image' and 'On-Device, IoT & Social' will be held simultaneously.

In the Vision and Image track, the following topics are covered: △Projects focused on deep network capabilities to bridge the perception gap between humans and deep learning; △Deep learning-based real-time tracking and synthesis technologies for building realistic 3D telepresence; △Recent research trends and new deep learning-based imaging pipelines to address the limitations of existing mobile device video capture; and △Inferring the overall flow based on a portion of the data and situations from various perspectives.A lecture on AGI-based object recognition and context description technology that can interpret and utilize [it] will be held.

In the On-Device, IoT, and Social tracks, lectures will cover: deep learning technologies that mobile embedded devices must overcome for on-device computing in the era of general artificial intelligence; methods for high-resolution location tracking based on deep learning and Wi-Fi; methods for applying deep learning-based inference and control technologies to solve various problems in 5G and IoT wireless systems; and recent trends in conversational modeling, which is gaining attention in the field of social sciences.

Meanwhile, in addition to lectures by distinguished scholars, this forum also features a poster session showcasing selected outstanding papers from domestic universities and graduate schools chosen by major global academic societies in the field of AI.