Speakers
Opening Ceremony
The IOAI Opening Ceremony will take place on August 10th at 14:30pm at the Burgas Theatre. It will feature welcome addresses from prominent figures, impressive performances, and a keynote by Prof Preslav Nakov, a leading researcher in fake news and LLMs.
Keynote speaker: Preslav Nakov
What to expect: In his keynote, Preslav will share his own AI journey: from competitive programmer to author of the favorite textbooks on algorithms in Bulgaria, to PhD in Computer Science with an AI major at UC Berkeley, and Professor at MBZUAI, the world’s first specialized AI university. He will share advice for young people passionate about programming and AI and will share insights from the main research problems he works on: fake news, factuality, and large language models, especially within the broader LLM research at MBZUAI.
Closing Ceremony
The IOAI Closing Ceremony will take place on August 14th at 14:30pm at the Burgas Congress Center. It will feature a keynote by Misha Denil, a prominent research scientist at Google DeepMind, followed by the Awards ceremony for winning teams.
Keynote speaker: Misha Denil
Speaker bio: Misha Denil is a Research Scientist at Deepmind where he has worked on many aspects of machine learning, including reinforcement learning, meta learning, learning from demonstrations. For the past few years he has been focused on learning for robotic manipulation, specifically on the challenges associated with applying cutting edge machine learning techniques to robotics.
What to expect: After the tremendous success of large language models in cognitive domains like language understanding, reasoning, planning and theorem proving one of the next frontiers for LLMs is understanding the physical world. Is the path to Spatial AI the same as the path to Cognitive AI? Is more scale and data really all we need? There is reason to believe that Spatial AI will require more than just repeating the same recipe that worked for language. In his keynote, Misha will talk about what some of those reasons are, and cover some recent works that he believes are promising directions for achieving Spatial AI.
Conference
The IOAI will be hosting a Conference on August 12th in between the two competition days. The conference is not related to the competition part of the Olympiad and the information covered in it will not be tested. Instead, the purpose of the conference is to enrich the experience of the teams during the Olympiad. The conference will start with two debates in which students will be introduced to some of the big questions in AI nowadays and will be able to express their opinions in response to questions posed by the presenters.
10:00 - 11:00 Large hall (open to the public)
Debate 1: The evolution NLP: What does the future hold?
Speaker bio: Prof Dr Ruslan Mitkov is a Professor in Computing and Communications at Lancaster University, one of the top-10 UK universities. He has been working in NLP, Computational linguistics, and related areas since the early 1980s and is best known for his seminal contributions to key areas such as anaphora resolution. He has authored more than 320 publications including 15 books, 35 journal articles and 35 book chapters.
What to expect: In this entertaining talk, Ruslan will the sketch the history of Natural Language Processing and Machine Translation and will review latest advances powered by Deep Learning and Large Language Models. He will share his original research which compares LLMs, Deep Learning and rule-based approaches for selected NLP tasks and applications. These studies will serve as platform for a follow-up discission on the future of Natural Language Processing. The speaker will emphasise that he is not a clairvoyant but on the basis of his experience in the field he will attempt to predict the likely future of artificial intelligence as compared to human intelligence taking language as a testbed.
11:00 - 12:00 Large hall (open to the public)
Debate 2: Is ethical AI even a thing?
Speaker bios: Ivo Emanuilov is a legal practitioner and information technology expert with over 10 years of international experience as a consultant and legal advisor to large companies in the field of smart manufacturing, telecommunications, and media. In 2024, he established I/O Legal as an informal network of experts and a hub of legal, scientific and technological expertise with a focus on IP, AI and data.
Katerina Yordanova is a legal practitioner and senior researcher at the Belgian nanotechnology company imec and KU Leuven. Her work is focused on artificial intelligence and experimental and anticipatory regulation, regulatory laboratories and human rights protection. Katerina advises small and start-up businesses on data management, privacy protection and ethical issues related to AI systems.
What to expect: Can machines tell good from bad? Whose responsibility is it to develop artificial intelligence that is aligned with basic human values? Is this even possible? Can artificial intelligence replace human ingenuity and creativity? In this talk, Ivo and Katerina will discuss what ethics means in the world of artificial intelligence and will lead an engaging debate with the audience on how to develop technologies that solve people’s real problems.
12:30 - 14:00 Large hall (open to the public)
Open lecture for the Burgas public: "Democratizing creativity: Open-source AI art tools"
Speaker bios: Stoyan Stoyanov is a digital artist specialized in motion graphics, photorealistic illusions, and particle simulations. In 2011, Stoyan cofounded Elektrick.me, a motion graphics collective based in Bulgaria. Over the next decade, he spearheaded dozens 3D mapping projects, ranging from interactive installations for brands like Mercedes, Intel, Pernod Ricard, BAT, etc. to large-scale projections across global venues.
Vladimir Grancharov is a designer and co-founder of Elektrick.Me. He graduated in engineering design at the Technical University of Sofia. He has directed major productions worldwide – interactive installations, multimedia performances for concerts, operas and theater, TV productions, and 3D projection mappings on iconic buildings in Bulgaria, Kazahstan, Japan, China, Russia, etc.
What to expect: In recent years, we’ve witnessed a remarkable transformation in the world of digital art and creativity. At the forefront of this revolution are open-source AI art tools, which have dramatically lowered the barriers to entry for creating stunning visual content. This talk will explore how these powerful technologies, with a particular focus on Stable Diffusion, are reshaping the creative landscape. Stoyan and Vlado will delve into the ecosystem of tools that has rapidly evolved around AI-generated art, examining how they’re empowering artists, designers, and even those with no formal artistic training to bring their visions to life. From text-to-image generation to advanced editing techniques, they will showcase the versatility and potential of these tools.
Workshops
The second part of the IOAI Conference will be four practical workshops which will run simultaneously. Teams will have to fill out a form in which they will rank their preferences, and they will be assigned to a workshop on a first-come first-serve basis. The purpose of the workshops is to give students an opportunity to try out techniques and tools in a hands-on way, under the guidance of experienced AI practitioners.
12:30 - 14:00 Foros hall (for contestants only)
Workshop 1: The Art of adaptation: Parameter efficient fine tuning using Gemma
Speaker bio: Parisa Haghani is a Senior Staff member at Google Speech, where she leads a team of research scientists and engineers on automated speech recognition (ASR) research. Her research interests span areas of speech recognition, understanding, and transferring successful research ideas to production. Parisa received her Ph.D. from EPFL Switzerland after which she spent a year at UIUC before joining Google.
What to expect: Gemma is a family of lightweight, state-of-the art open models built from the same research and technology used to create the Gemini models by Google. In this workshop, students will go over the fundamentals of parameter efficient training and use that paradigm to enable a new task on Gemma. Depending on the time and interests of the participants, they will tap into making Gemma multi-modal: to make it hear and transcribe audio. Parisa will be assisted by Roshan Sharma, Xinjian Li, and Tongzhou Chen, who will be joining remotely.
12:30 - 14:00 Foros hall (for contestants only)
Workshop 2: An introduction to Reinforcement Learning
Speaker bio: Todor Davchev is a Senior Research Scientist at Google DeepMind where he works on Robot Learning. Prior to that, Todor completed a PhD at the University of Edinburgh through which he did two Google X AI residencies (project Intrinsic) and a research scientist internship with DeepMind. Broadly, Todor’s interests lie in the intersection of robotics and machine learning.
What to expect: Reinforcement learning has been widely successful in solving particular tasks defined with a reward function – from superhuman Go playing and refining ChatGPT to magnetic confinement for plasma control and solving complex tasks in Robotics. This is a branch of Machine Learning that allows for our agents to learn from interacting with completely unknown situations when deployed in the real world. In this workshop, students will learn the fundamentals of Reinforcement Learning, the standard learning paradigm allowing for dynamic online acquisition of knowledge, purely through reward signals. Examples relevant to robotics will be shared!
12:30 - 14:00 SU hall (for contestants only)
Workshop 3: Transforming healthcare with Computer Vision AI
Speaker bios: Georgi Kostadinov, a renowned computer vision expert, specializes in developing sophisticated algorithms for image and video analysis and has led the creation of Imagga Cloud AI, processing over 30 billion data points. As co-founder and technical director at Kelvin Health, he develops AI-driven solutions for early diagnosis and preventive medicine using thermal image analysis.
Georgi Kadrev is an innovator in the area of visual AI, pioneering concepts such image recognition API and coining terms like image auto-tagging back in 2008. Most recently Georgi co-founded Kelvin Health with the mission to make diagnostics accessible for everyone, in every context. The company has received multiple global innovation awards since its inception in 2020.
What to expect: Join Georgi and Georgi for a dynamic workshop on the groundbreaking applications of computer vision AI in the healthcare industry. Discover how their HealthTech company Kelvin Health leveraged convolutional neural networks (CNNs) to address critical challenges in medical diagnostics and patient care. Through hands-on sessions, participants will learn about the fundamentals of CNNs, their architectural evolution, and real-world applications that are transforming healthcare. Students will gain insights into the future of AI-driven healthcare solutions and how they can improve patient outcomes and operational efficiency.
12:30 - 14:00 SoftServe hall (for contestants only)
Workshop 4: LLM-based agents as digital assistants
Speaker bio: Mareike Hartmann is a PostDoc at Saarland University, Saarbrücken Germany. Her current research interests are interactive LLM-based agents, e.g. for API orchestration, code generation, or web navigation, and retrieval-augmented generation. Prior to that, she was a researcher at the German Research Center for Artificial Intelligence in Saarbrücken, and received her PhD from Copenhagen University.
What to expect: Large language models like GPT4 have shown to be useful backbones for goal-directed interactive agents, which perceive their environment and take actions in order to fulfill a task. In this workshop, Mareike will focus on llm-based agents acting as digital assistants by operating common Apps like Spotify or Amazon, given high-level app descriptions, the ability to look up more detailed app documentation, and some examples. Experiments will be based on the tasks + software from the AppWorld engine, with interactive and non-interactive prompting techniques for zero and few-shot in-context learning.