HSE University Hosts Fall into ML 2023 Conference on Machine Learning

Over three days, more than 300 conference participants attended workshops, seminars, sections and a poster session. During panel discussions, experts deliberated on the regulation of artificial intelligence (AI) technologies and considered collaborative initiatives between academic institutions and industry to advance AI development through megaprojects.

The conference brought together experts in fundamental science (from HSE University, MSU, RAS, MIPT, ITMO, Skoltech, and NSU), industry (Sber, Yandex), and state institutions (ANO 'Digital Economy').

Ivan Arzhantsev

While addressing the conference’s opening session, Ivan Arzhantsev, Dean of the HSE Faculty of Computer Science, highlighted the significance of Fall into ML, not only for the FCS, but for the entire university. ‘Since the establishment of the FCS, our aspiration has been for HSE University to host a top-tier international conference on machine learning. Its initiator, Alexey Naumov (Head, International Laboratory of Stochastic Algorithms and High-Dimensional Inference — Ed.), suggested assembling experts who had already been featured speakers at prominent international events. As a result, everything worked out as planned,' Arzhantsev noted.

This year, Fall into ML brought together over 50 authors, who have published papers in A* level conferences, which are considered flagship events in this field. Over the course of three days, more than 300 conference participants attended thematic workshops, seminars, sections, and a poster session.

For instance, at the 'NN Diagnostics' workshop, various approaches to managing processes that occur during the training of neural networks were presented. In addition, the 'AI in Physics' workshop focused on the problem of recounting dynamic systems through the application of machine learning methods. During the 'Reinforcement Learning' workshop, speakers discussed the challenges associated with constructing effective reinforcement learning algorithms, and explored their practical applications. During the 'AI in Medicine' session, various AI methods employed for the analysis of medical data were considered.

Maria Poptsova

‘It is impossible to cover all issues associated with medical tasks in a single session. However, we successfully brought together key figures in this field, resulting in diverse conference presentations united by a common theme,' comments Maria Poptsova, Head of the International Laboratory of Bioinformatics at the AI and Digital Science Institute of HSE University. ‘Machine learning algorithms not only serve to analyse vast amounts of data and aid medics in diagnosis, but also propose optimal treatment options while considering multiple factors, as well as enable disease prediction. The subject of AI in medicine is of immediate interest, presenting numerous opportunities to enhance the quality of healthcare, improve diagnostic accuracy, and implement a personalised approach to patient care and treatment.'

The panel discussions at Fall into ML 2023 provided a platform for experts to engage in open and free communication. The panel discussion 'Strong AI: risks and benefits' covered the key aspects of strong artificial intelligence, ranging from defining strong AI to discussing its potential to addressing ethical and social considerations, as well as the risks, associated with its development, e.g., concerns over privacy, security, and autonomy.

'One of our shared concerns is the regulation of AI technologies,' comments Karen Ghazaryan, Director of Analytics at ANO ‘Digital Economy’, noting: 'In various domains, we observe how diverse algorithms replace human operators, accomplishing more complex tasks in a fraction of the time. The degree of technology integration in all areas has been expanding, with models becoming automated and operating without human intervention. At the same time, the possibility of software errors and information security risks cannot be ruled out. Given the current state of technological progress and strategic planning horizon, achieving strong AI remains elusive. However, today we are developing risk-based approaches within specific industries to ensure the safe implementation of such technologies. During the discussion at Fall into ML 2023, we were able to explore the advantages and risks associated with AI.'

During the panel discussion 'Science in the academy vs industry,' experts deliberated on the megaprojects that universities can collaboratively undertake with industry in order to advance AI technologies. It was emphasised that the progress of science and industry in the field of AI should be inseparable so as to ensure mutually beneficial cooperation, innovative solutions, and ethical standards for the general advancement of this field.

‘During the discussion, my colleagues and I noted the need for advancing the theory of machine learning. Currently, theory in this area is in its infancy, and there is a notable deficiency, particularly in the development of trusted intelligent systems. As we witness increasing AI implementations, this problem will become even more pronounced. For instance, in the development of self-driving cars, developers cannot currently guarantee that the algorithm will not fail in certain highly unlikely circumstances, which were not considered during the model's training,' explains Denis Turdakov, Head of ISP RAS Research Centre for Trusted Artificial Intelligence, adding: ‘Therefore, both academia, including research centres, and industry should align their efforts to address this issue. Fall into ML has served as a platform for interaction among scientists, who work in both academia and industry. We have not only identified current trends in the advancement of AI, but also discussed career opportunities for young specialists in today’s world.’

All authors of papers presented at the conference participated in a poster session. The participants presented their research, work outcomes, and new AI technologies and methods on posters featuring concise information about each paper, including its objectives, methods and results.

Inna Kolinko, UserGate, SB RAS Institute of Mathematics, Novosibirsk State University:

‘I was particularly impressed by the mini-course “Multimodal multitask models as a tool for generative artificial intelligence”, delivered by Elizaveta Goncharova and Vladimir Arkhipkin, and the presentations “Layerwise Universal Adversarial Attack on NLP Models” by Olga Tsymboi and Ivan Oseledets, and “SpQR: A Sparse-quantized Representation For Near-Lossless LLM Weight Compression” by Denis Kuznedelev. I would also like to highlight the excellent organization of the conference itself, with fantastic coffee breaks, and I particularly appreciated the format of the discussions.’

Nikolay Smolyanov, ITMO:

‘The conference has been highly productive. As a newcomer to machine learning, I have gained valuable insights in terms of both professional competencies such as multimodal models and various applications of neural networks, and a better understanding of the threats, advantages, and distinctive features in the field of artificial intelligence. Clearly, this conference attracts experts with extensive experience, making it truly interesting to listen and learn. The event was organized at the highest level—thank you!’

Nikolay Plyuta, Moscow State University of Civil Engineering:

‘The organization is top-notch; nothing less would be expected from HSE University! As a novice in machine learning, I found most of the presentations to be accessible and easy to comprehend. I would like to participate in this event next year as well.’

Marina Mikitchuk, Vega Institute Foundation, MSU Moscow School of Economics, RAS Central Economics and Mathematics Institute:

‘Fall into ML 2023 left the most incredible impressions! Many thanks to the organizers for the exceptional quality of the presentations, the warm atmosphere, and opportunities for new and engaging discussions.’

The video can be viewed on the official website of the conference.