ABSTRACT. Emotions are a very important factor that can significantly affect the level of success of students, their ability to think and analyze information during learning and make appropriate decisions. The emotional states of students studying online and the impact of different types of educational content presentation during a lecture on their emotions were studied. Using MorphCast, seven main emotions (Neural, Happy, Surprise, Sad, Angry, Fear, Disgust) representing the facial expressions of students during an online lecture were obtained, based on which a report on the level of atten-tion and engagement was generated. The majority of students were ready to use modern AI-related technologies to collect information to help them in their studies and were willing to share their data with teachers. Verbal con-sent was obtained from the participants before the study began. Participants who did not want to participate were not included in this study. The pedagog-ical experiment on the implementation of facial emotion recognition (FER) of students lasted during the 2023-2024 academic year at the National Uni-versity of Life and Environmental Sciences of Ukraine (NULES). To evalu-ate AI-based emotion recognition tools, 17 teachers and 39 students were surveyed on certain assessment indicators after a series of lecture classes. As a result of the pedagogical experiment, recommendations have been made that to improve the quality of online classes, it is worth introducing interac-tive teaching methods, and various forms of presenting different types of educational content during online lectures.

ABSTRACT. Constant Constant life threats are disrupting the daily routines of preschool-ers and pose significant risks to their mental and physical health. Neverthe-less, Ukraine's educational institutions continue to operate, demonstrating re-silience by adapting to these challenging conditions. Scientific studies indi-cate that such stressful conditions can severely strain the nervous systems of children and their teachers. As a result, it's crucial for future teachers to un-derstand how to alleviate children's anxiety, manage their emotional states, and prioritize their own well-being while providing support. Borys Grinchenko Kyiv Metropolitan University annually reviews and updates its educational programs in accordance with current societal chal-lenges and surveys results of stakeholders, students, and graduates. In 2023, the Department of Preschool Education introduced a new compulsory course "Child in Society" as a part of the educational and professional program 012 Preschool Education, aimed at training pre-service teachers of toddlers and preschoolers. This involves fostering social and civic competence through personalized interactions that prioritize individual connections between adults and children. Additionally, the course integrates a psychosocial sup-port component "Well-being of Children and Teachers: Effective Tools and Practices of Psychosocial Support," designed by the MHPSS Collaborative. The use of Information and Communication Technology (ICT) has become crucial in addressing these challenges. It delves into the effective use of ICT tools and resources to facilitate the "Child in Society" course and foster active engagement with students. By sharing these insights, the article aims to provide valuable guidance to educa-tors and educational institutions navigating the complexities of teaching dur-ing conflict.

ABSTRACT. Qualitative education is one of the efficient tools to ensure the sustainable development of the economy and society. Given that, special significance is attributed to the use of labor market data in order to make academic decisions. The integration of information technologies, big data, and artificial intelligence methods into the decision-making process enhances this efficiency and reduces costs. However, suggested approach requires the training of data specialists and university administrators. The novelty of the research lies in the implementation of machine learning methods, which enable the classification and specification of sets of skills revealed through the analysis of open vacancies for data engineers. This is achieved through clusterization (unsupervised learning) and the development of a model for predicting possible areas of professional growth based on existing skills using the XGBoost method (supervised learning). The developed model possesses high accuracy (99.53%), which means it can be implemented on real data (students) to enhance the mastery of skills recommended based on machine learning. The results of structuring vacancies according to the most relevant skills (five clusters have been revealed) are relevant for improving individual educational trajectories of academic programs for future data engineers. The article presents an example of the implementation of the created model for specifying elective courses in the academic program "Information Systems and Technologies" for bachelor's degree students at Kherson State University. Although, the implementation of models of machine learning and Data-driven Decisions significantly depends on data, the results of the research can be implemented in various educational fields.

ABSTRACT. The article presents the experience of creating an intellectual assistant for educational management. The RDF model of educational management, its subsequent processing in JSONL for Azure AI Studio, and the learning process in the environment of a higher education institution are described. Understanding the online educational environment as a virtual space in which students of higher education can interact with each other, access information and use online tools has allowed the introduction of online tutoring mechanism. Such a toolkit allows you to use a personalized approach to learning, taking into account the individual needs and capabilities of each student. Compliance with the developed algorithm of using an intelligent online tutor revealed the effectiveness of supporting edu- cational and educational goals of higher education seekers, which made it possi- ble to take into account their further competitiveness in the labor market.

ABSTRACT. The paper examines the implementation of the practical part of the program in the discipline "physics" regarding the study of light dispersion in the con-ditions of online learning, specifically during the martial law in the regions close to the front line. A software application has been created based on a real experiment on the study of light dispersion. The application program-ming interface (API) of Windows Forms, which is part of the Microsoft .NET Framework in the integrated development environment (IDE) of Microsoft Visual Studio 2022, was chosen for it.

ABSTRACT. To foster student motivation, it is crucial to infuse learning with enjoyment during the education process. The advent of information and communication technologies has made access to digital tools increasingly more convenient. Teachers frequently employ free digital resources, such as simulations, pictures, videos, and games, in their teaching-learning process. By incorporating simulations into the Predict-Observe-Explain (POE) method, the POE method can positively impact student success, particularly in bolstering students' belief in their ability to learn and perform well in chemistry courses. This study examined the impact of a simulation-supported POE intervention on high school students' self-efficacy in learning and performance. Two groups of students were randomly selected as the experimental and control groups. While the experimental group received lessons using the POE method supported by simulations, the control group received lessons using the teacher's traditional method. The Self-Efficacy for Learning and Performance Scale, one of the subscales of the Motivated Strategies for Learning Questionnaire (MSLQ), was used to assess high school students’ self-efficacy levels. The findings indicate that simulation-supported POE led to a significant increase in student self-efficacy. The results of this study suggest that employing the POE method with digital tools, such as simulations, in teaching-learning environments can enhance teachers' effectiveness in teaching chemistry compared to the control group.

ABSTRACT. The article provides a brief description of the MathPartner service. This freely available cloud-based Mathematics is a universal system for symbolic-numeric calculations. Its Mathpar language is a subset of the LaTeX language, but allows you to create mathematical texts that contain “computable” mathematical operators. This opens up completely new opportunities for improving the educational process for all natural science disciplines, for the use of mathematics in scientific and engineering calculations. To save and freely exchange educational and other texts in the Mathpar language, a GitHub repository has been created. It is concluded that MathPartner cloud mathematics is a new breakthrough technology for science education, scientific and engineering applications.

ABSTRACT. We delve into our research on AI safety, focusing on advancements aimed at ensuring the robustness, alignment, and fairness of large language models (LLMs). The talk will start with an exploration of the challenges posed by sensitivity in AI systems and strategies for providing provable guarantees against worst-case adversaries. Building upon this, we navigate through the alignment challenges and safety considerations of LLMs, addressing both their limitations and capabilities, particularly following techniques related to instruction prefix tuning and their theoretical limitations towards alignment. At last, I will talk about fairness across languages in common tokenizers in LLMs.

The talk will be based on these three papers:

1. Language Model Tokenizers Introduce Unfairness Between Languages (NeurIPS23)
2. When do prompting and prefix-tuning work? a theory of capabilities and limitations (ICLR24)
3. Prompting a pretrained transformer can be a universal approximator (ICML24)

ABSTRACT. Discovering mutations enhancing protein-protein interactions (PPIs) is critical for advancing biomedical research and developing improved therapeutics. While machine learning approaches have substantially advanced the field, they often struggle to generalize beyond training data in practical scenarios. The contributions of this work are three-fold. First, we construct PPIRef, the largest and non-redundant dataset of 3D protein-protein interactions, enabling effective large-scale learning. Second, we leverage the PPIRef dataset to pre-train PPIformer, a new SE(3)-equivariant model generalizing across diverse protein-binder variants. We fine-tune PPIformer to predict effects of mutations on protein-protein interactions via a thermodynamically motivated adjustment of the pre-training loss function. Finally, we demonstrate the enhanced generalization of our new PPIformer approach by outperforming other state-of-the-art methods on new, non-leaking splits of standard labeled PPI mutational data and independent case studies optimizing a human antibody against SARS-CoV-2 and increasing the thrombolytic activity of staphylokinase.

ABSTRACT. This work aims at leveraging instructional video to guide the solving of complex multi-contact task-and-motion planning tasks in robotics. Towards this goal, we propose an extension of the well-established Rapidly-Exploring Random Tree (RRT) planner, which simultaneously grows multiple trees around grasp and release states extracted from the guiding video. Our key novelty lies in combining contact states, and 3D object poses extracted from the guiding video with a traditional planning algorithm that allows us to solve tasks with sequential dependencies, for example, if an object needs to be placed at a specific location to be grasped later. To demonstrate the benefits of the proposed video-guided planning approach, we design a new benchmark with three challenging tasks: (i) 3D re-arrangement of multiple objects between a table and a shelf, (ii) multi-contact transfer of an object through a tunnel, and (iii) transferring objects using a tray in a similar way a waiter transfers dishes. We demonstrate the effectiveness of our planning algorithm on several robots, including the Franka Emika Panda and the KUKA KMR iiwa.

ABSTRACT. Large Language Models (LLMs) have demonstrated impressive capabilities in various applications such as dialogue systems and story writing. However, their deployment remains cost-prohibitive, particularly due to the extensive memory requirements associated with long-content generation. This talk will present innovative approaches aimed at reducing the memory footprint and improving the throughput of LLMs, focusing on the management of the KV cache, a key component stored in GPU memory that scales with sequence length and batch size. I will first introduce the Heavy Hitter Oracle (H2O), a novel KV cache eviction policy that significantly reduces memory usage by dynamically retaining a balance of recent and “Heavy Hitter” (H2) tokens—tokens that contribute most to the attention scores. We will discuss how the H2O approach, based on submodular optimization, outperforms existing in-ference systems. We will then address the compound effects of combining sparsification and quantization techniques on the KV cache, proposing the Q-Hitter framework. Q-Hitter enhances the H2O approach by identifying to-kens that are not only pivotal based on accumulated attention scores but also more suitable for low-bit quantization. This dual consideration of attention importance and quantization friendliness leads to substantial memory savings and throughput improvements. We will conclude the talk by reviewing more recent efforts on efficient LLM inference and the future opportunities.

ABSTRACT. We propose MFT -- Multi-Flow dense Tracker -- a novel method for dense, pixel-level, long-term tracking. The approach exploits optical flows estimated not only between consecutive frames, but also for pairs of frames at logarithmically spaced intervals. It selects the most reliable sequence of flows on the basis of estimates of its geometric accuracy and the probability of occlusion, both provided by a pre-trained CNN. We show that MFT achieves competitive performance on the TAP-Vid benchmark, outperforming baselines by a significant margin, and tracking densely orders of magnitude faster than the state-of-the-art point-tracking methods. The method is insensitive to medium-length occlusions and it is robustified by estimating flow with respect to the reference frame, which reduces drift.

The talk is based on the paper

• MFT: Long-Term Tracking of Every Pixel (WACV-24)

ABSTRACT. Machine Learning (ML) models are being used to make decisions in increasingly critical domains. To determine whether models are production-ready, they must be comprehensively evaluated on a number of performance dimensions. Since measuring only accuracy and fairness is not enough for building robust ML systems, each model involves at least three overall dimensions (correctness, stability, uncertainty) and three disparity dimensions evaluated on subgroups of interest (error disparity, stability disparity, uncertainty disparity). Adding to the complexity, these model dimensions exhibit trade-offs with one another. Considering the multitude of model types, performance dimensions, and trade-offs, model developers face the challenge of responsible model selection.

In this paper, we present a comprehensive software library for model auditing and responsible model selection, called Virny, along with an interactive tool called VirnyView. Our library is modular and extensible, it implements a rich set of performance and fairness metrics, including novel metrics that quantify and compare model stability and uncertainty, and enables performance analysis based on multiple sensitive attributes, and their intersections. The Virny library and the VirnyView tool are available at https://github.com/DataResponsibly/Virny and https://r-ai.co/VirnyView.

ABSTRACT. This talk will cover research at the forefront of biomedical imaging data analysis, leveraging machine learning and deep learning to create state-of-the-art software solutions for diverse biomedical imaging modalities. Collaborating with academic institutions, private companies, and hospitals, we address a wide range of biomedical challenges. Our partnerships with Revvity have led to breakthroughs in microscopy image analysis, including a pioneering AI model for brightfield microscopy. In histopathology, our work with Estonian hospitals aims to automate and expedite pathology workflows, improving diagnostic accuracy and patient outcomes. Additionally, our AI research in medical imaging, in collaboration with a MedTech startup - Better Medicine, focuses on developing the General Purpose Medical AI (GPAI) system, which uses weakly supervised models to streamline radiological workflows and reduce the burden on radiologists. Our efforts are particularly impactful in Estonia, a nation with progressive digitalization policies and centralized electronic health records, as we advance the digital transformation of healthcare, ultimately improving patient care and accelerating therapeutic discoveries through innovative AI-driven analysis.

The talk will be based on these papers:

1. Counterfactual inpainting for weakly supervised semantic segmentation for medical images (XAI-2024)
2. Metadata Improves Segmentation Through Multitasking Elicitation (DART 2023)