Ilya Makarov

Kevin Wang, Anna Thöni, Benjamin Kempinski et al.

Large language models (LLMs) are increasingly deployed as interactive agents, yet their capacity for social and strategic reasoning over extended interaction remains poorly understood. Existing evaluations rely on static vignettes or single-game benchmarks that cannot capture the sustained, multi-faceted reasoning that real-world multi-agent settings demand. We introduce Mindgames, a multi-game arena and evaluation platform for LLM agents that operationalizes complementary reasoning demands relevant to ``theory of mind'': belief attribution under hidden information, opponent modeling through repeated strategic interaction, cooperative inference under knowledge asymmetries, and sustained deception in social deduction. Built on TextArena, Mindgames provides a unified interaction interface, TrueSkill-based rating, and full trajectory logging across four game environments. We instantiate Mindgames through a 2025 competition cycle hosted at a major AI conference, which assessed 944 submitted agents from 76 teams across four games: Colonel Blotto, Iterated Prisoner's Dilemma, Codenames, and Secret Mafia. Our analysis surfaces both agent-level and evaluation-level limitations: brittle rule adherence remains a major bottleneck, top-performing systems repeatedly rely on explicit structural scaffolding, and leaderboard validity differs sharply across environments. In particular, failure-heavy environments can reward robustness to opponent errors as much as strategic ability, with Secret Mafia exhibiting a pronounced error-survival confound in this cycle. We release a dataset of 29,571 multi-agent games with turn-level observations, actions, and rewards, together with MG-Ref, a deterministic offline tournament protocol that scores new agents against a frozen reference pool of top-ranked, low-error Stage~II submissions under the same error-attribution lens used in this analysis.

Maksim Golyadkin, Vitaliy Pozdnyakov, Leonid Zhukov et al.

Modern industrial facilities generate large volumes of raw sensor data during the production process. This data is used to monitor and control the processes and can be analyzed to detect and predict process abnormalities. Typically, the data has to be annotated by experts in order to be used in predictive modeling. However, manual annotation of large amounts of data can be difficult in industrial settings. In this paper, we propose SensorSCAN, a novel method for unsupervised fault detection and diagnosis, designed for industrial chemical process monitoring. We demonstrate our model's performance on two publicly available datasets of the Tennessee Eastman Process with various faults. The results show that our method significantly outperforms existing approaches (+0.2-0.3 TPR for a fixed FPR) and effectively detects most of the process faults without expert annotation. Moreover, we show that the model fine-tuned on a small fraction of labeled data nearly reaches the performance of a SOTA model trained on the full dataset. We also demonstrate that our method is suitable for real-world applications where the number of faults is not known in advance. The code is available at https://github.com/AIRI-Institute/sensorscan.

Aliaksei Korshuk, Alexander Buyantuev, Ilya Makarov

Training language model agents for multi-agent strategic interaction presents a core difficulty: the quality of any action may depend on future events that never materialize, on moves that violate game rules, or on decisions made by other players. Standard reinforcement learning assumes that rewards can be assigned at each step, but this assumption fails in settings where outcomes are entangled across time and agents. We introduce delayed per-step reward attribution with eligibility gating, an episode lifecycle and postprocessing pipeline that computes rewards only at episode end, propagates them back to originating steps according to task-specific semantics, and excludes steps that lack valid dependent information from training. Together with asynchronous rollout generation via vLLM's continuous batching, curriculum-based opponent sampling, and multi-level stratified batch construction, this approach enables stable, sample-efficient RL training in multi-agent environments. We evaluate on the MindGames Arena benchmark at NeurIPS 2025, where a single 8-billion-parameter open-source model trained with our method matched or surpassed substantially larger proprietary systems, including GPT-5, in head-to-head play and took first place in both the Open (unrestricted) and Efficient (<=8B parameters) tracks.

Arman Martirosyan, Shahane Tigranyan, Maria Razzhivina et al.

Micro-gesture recognition and behavior-based emotion prediction are both highly challenging tasks that require modeling subtle, fine-grained human behaviors, primarily leveraging video and skeletal pose data. In this work, we present two multimodal frameworks designed to tackle both problems on the iMiGUE dataset. For micro-gesture classification, we explore the complementary strengths of RGB and 3D pose-based representations to capture nuanced spatio-temporal patterns. To comprehensively represent gestures, video, and skeletal embeddings are extracted using MViTv2-S and 2s-AGCN, respectively. Then, they are integrated through a Cross-Modal Token Fusion module to combine spatial and pose information. For emotion recognition, our framework extends to behavior-based emotion prediction, a binary classification task identifying emotional states based on visual cues. We leverage facial and contextual embeddings extracted using SwinFace and MViTv2-S models and fuse them through an InterFusion module designed to capture emotional expressions and body gestures. Experiments conducted on the iMiGUE dataset, within the scope of the MiGA 2025 Challenge, demonstrate the robust performance and accuracy of our method in the behavior-based emotion prediction task, where our approach secured 2nd place.

Chandler Smith, Marwa Abdulhai, Manfred Diaz et al.

Large Language Model (LLM) agents have demonstrated impressive capabilities for social interaction and are increasingly being deployed in situations where they might engage with both human and artificial agents. These interactions represent a critical frontier for LLM-based agents, yet existing evaluation methods fail to measure how well these capabilities generalize to novel social situations. In this paper, we introduce a method for evaluating the ability of LLM-based agents to cooperate in zero-shot, mixed-motive environments using Concordia, a natural language multi-agent simulation environment. Our method measures general cooperative intelligence by testing an agent's ability to identify and exploit opportunities for mutual gain across diverse partners and contexts. We present empirical results from the NeurIPS 2024 Concordia Contest, where agents were evaluated on their ability to achieve mutual gains across a suite of diverse scenarios ranging from negotiation to collective action problems. Our findings reveal significant gaps between current agent capabilities and the robust generalization required for reliable cooperation, particularly in scenarios demanding persuasion and norm enforcement.

Ivan Lukyanenko, Mikhail Mozikov, Yury Maximov et al.

Hail risk assessment is necessary to estimate and reduce damage to crops, orchards, and infrastructure. Also, it helps to estimate and reduce consequent losses for businesses and, particularly, insurance companies. But hail forecasting is challenging. Data used for designing models for this purpose are tree-dimensional geospatial time series. Hail is a very local event with respect to the resolution of available datasets. Also, hail events are rare - only 1% of targets in observations are marked as "hail". Models for nowcasting and short-term hail forecasts are improving. Introducing machine learning models to the meteorology field is not new. There are also various climate models reflecting possible scenarios of climate change in the future. But there are no machine learning models for data-driven forecasting of changes in hail frequency for a given area. The first possible approach for the latter task is to ignore spatial and temporal structure and develop a model capable of classifying a given vertical profile of meteorological variables as favorable to hail formation or not. Although such an approach certainly neglects important information, it is very light weighted and easily scalable because it treats observations as independent from each other. The more advanced approach is to design a neural network capable to process geospatial data. Our idea here is to combine convolutional layers responsible for the processing of spatial data with recurrent neural network blocks capable to work with temporal structure. This study compares two approaches and introduces a model suitable for the task of forecasting changes in hail frequency for ongoing decades.

Matvey Gerasyov, Ilya Makarov

Deep reinforcement learning in partially observable environments is a difficult task in itself, and can be further complicated by a sparse reward signal. Most tasks involving navigation in three-dimensional environments provide the agent with extremely limited information. Typically, the agent receives a visual observation input from the environment and is rewarded once at the end of the episode. A good reward function could substantially improve the convergence of reinforcement learning algorithms for such tasks. The classic approach to increase the density of the reward signal is to augment it with supplementary rewards. This technique is called the reward shaping. In this study, we propose two modifications of one of the recent reward shaping methods based on graph convolutional networks: the first involving advanced aggregation functions, and the second utilizing the attention mechanism. We empirically validate the effectiveness of our solutions for the task of navigation in a 3D environment with sparse rewards. For the solution featuring attention mechanism, we are also able to show that the learned attention is concentrated on edges corresponding to important transitions in 3D environment.

Ilia Indyk, Ignat Penshin, Ivan Sosin et al.

Fisheye cameras are increasingly adopted in robotics for near-field manipulation, navigation, and immersive perception, yet indoor depth benchmarks with accurate ground truth are still missing. To address this, we introduce WideDepth - the first indoor dataset for fisheye depth estimation, featuring 101 scenes containing 5K high-resolution stereo pairs labeled with millimeter-level ground truth depth and disparity. Our dataset also includes paired pinhole and fisheye samples across varying fields of view and baselines in both horizontal and vertical stereo setups. We further propose a method to adapt pinhole-trained stereo models to fisheye images and introduce a novel stereo fisheye image generation pipeline based on high-resolution LiDAR scans. Leveraging these methods, we thoroughly evaluate state-of-the-art monocular depth, stereo matching, and depth completion models on our benchmark. Additionally, we provide 18K LiDAR-derived sparse depth training samples, achieving up to a 62% performance boost on fisheye data when fine-tuning pinhole-based stereo models. In summary, the high precision and versatility of our benchmark set a strong foundation for advancing research in fisheye depth estimation and robotics perception. Project page: https://ilyaind.github.io/WideDepth

Alexander Kovalenko, Vitaliy Pozdnyakov, Ilya Makarov

Timely detected anomalies in the chemical technological processes, as well as the earliest detection of the cause of the fault, significantly reduce the production cost in the industrial factories. Data on the state of the technological process and the operation of production equipment are received by a large number of different sensors. To better predict the behavior of the process and equipment, it is necessary not only to consider the behavior of the signals in each sensor separately, but also to take into account their correlation and hidden relationships with each other. Graph-based data representation helps with this. The graph nodes can be represented as data from the different sensors, and the edges can display the influence of these data on each other. In this work, the possibility of applying graph neural networks to the problem of fault diagnosis in a chemical process is studied. It was proposed to construct a graph during the training of graph neural network. This allows to train models on data where the dependencies between the sensors are not known in advance. In this work, several methods for obtaining adjacency matrices were considered, as well as their quality was studied. It has also been proposed to use multiple adjacency matrices in one model. We showed state-of-the-art performance on the fault diagnosis task with the Tennessee Eastman Process dataset. The proposed graph neural networks outperformed the results of recurrent neural networks.

Nikita Severin, Danil Kartushov, Vladislav Urzhumov et al.

Sequential recommender systems have achieved significant success in modeling temporal user behavior but remain limited in capturing rich user semantics beyond interaction patterns. Large Language Models (LLMs) present opportunities to enhance user understanding with their reasoning capabilities, yet existing integration approaches create prohibitive inference costs in real time. To address these limitations, we present a novel knowledge distillation method that utilizes textual user profile generated by pre-trained LLMs into sequential recommenders without requiring LLM inference at serving time. The resulting approach maintains the inference efficiency of traditional sequential models while requiring neither architectural modifications nor LLM fine-tuning.

Dmitry Zhevnenko, Mikhail Kazantsev, Ilya Makarov

The paper deals with the problem of controlling the state of industrial devices according to the readings of their sensors. The current methods rely on one approach to feature extraction in which the prediction occurs. We proposed a technique to build a scalable model that combines multiple different feature extractor blocks. A new model based on sequential sensor space analysis achieves state-of-the-art results on the C-MAPSS benchmark for equipment remaining useful life estimation. The resulting model performance was validated including the prediction changes with scaling.

Daria Pugacheva, Andrei Ermakov, Igor Lyskov et al.

Combinatorial optimization (CO) problems are crucial in various scientific and industrial applications. Recently, researchers have proposed using unsupervised Graph Neural Networks (GNNs) to address NP-hard combinatorial optimization problems, which can be reformulated as Quadratic Unconstrained Binary Optimization (QUBO) problems. GNNs have demonstrated high performance with nearly linear scalability and significantly outperformed classic heuristic-based algorithms in terms of computational efficiency on large-scale problems. However, when utilizing standard node features, GNNs tend to get trapped to suboptimal local minima of the energy landscape, resulting in low quality solutions. We introduce a novel algorithm, denoted hereafter as QRF-GNN, leveraging the power of GNNs to efficiently solve CO problems with QUBO formulation. It relies on unsupervised learning by minimizing the loss function derived from QUBO relaxation. The proposed key components of the architecture include the recurrent use of intermediate GNN predictions, parallel convolutional layers and combination of static node features as input. Altogether, it helps to adapt the intermediate solution candidate to minimize QUBO-based loss function, taking into account not only static graph features, but also intermediate predictions treated as dynamic, i.e. iteratively changing recurrent features. The performance of the proposed algorithm has been evaluated on the canonical benchmark datasets for maximum cut, graph coloring and maximum independent set problems. Results of experiments show that QRF-GNN drastically surpasses existing learning-based approaches and is comparable to the state-of-the-art conventional heuristics, improving their scalability on large instances.

Alexander Gambashidze, Aleksandr Dadukin, Maksim Golyadkin et al.

This paper addresses the critical challenges of sparsity and occlusion in LiDAR-based 3D object detection. Current methods often rely on supplementary modules or specific architectural designs, potentially limiting their applicability to new and evolving architectures. To our knowledge, we are the first to propose a versatile technique that seamlessly integrates into any existing framework for 3D Object Detection, marking the first instance of Weak-to-Strong generalization in 3D computer vision. We introduce a novel framework, X-Ray Distillation with Object-Complete Frames, suitable for both supervised and semi-supervised settings, that leverages the temporal aspect of point cloud sequences. This method extracts crucial information from both previous and subsequent LiDAR frames, creating Object-Complete frames that represent objects from multiple viewpoints, thus addressing occlusion and sparsity. Given the limitation of not being able to generate Object-Complete frames during online inference, we utilize Knowledge Distillation within a Teacher-Student framework. This technique encourages the strong Student model to emulate the behavior of the weaker Teacher, which processes simple and informative Object-Complete frames, effectively offering a comprehensive view of objects as if seen through X-ray vision. Our proposed methods surpass state-of-the-art in semi-supervised learning by 1-1.5 mAP and enhance the performance of five established supervised models by 1-2 mAP on standard autonomous driving datasets, even with default hyperparameters. Code for Object-Complete frames is available here: https://github.com/sakharok13/X-Ray-Teacher-Patching-Tools.

Harry Proshian, Nikita Severin, Sergey Nikolenko et al.

Large-scale digital platforms generate billions of timestamped user-item interactions (events) that are crucial for predicting user attributes in, e.g., fraud prevention and recommendations. While self-supervised learning (SSL) effectively models the temporal order of events, it typically overlooks the global structure of the user-item interaction graph. To bridge this gap, we propose three model-agnostic strategies for integrating this structural information into contrastive SSL: enriching event embeddings, aligning client representations with graph embeddings, and adding a structural pretext task. Experiments on four financial and e-commerce datasets demonstrate that our approach consistently improves the accuracy (up to a 2.3% AUC) and reveals that graph density is a key factor in selecting the optimal integration strategy.

Kirill Lukyanov, Georgii Sazonov, Serafim Boyarsky et al.

Graph Neural Networks (GNNs) have become a cornerstone in graph-based data analysis, with applications in diverse domains such as bioinformatics, social networks, and recommendation systems. However, the interplay between model interpretability and robustness remains poorly understood, especially under adversarial scenarios like poisoning and evasion attacks. This paper presents a comprehensive benchmark to systematically analyze the impact of various factors on the interpretability of GNNs, including the influence of robustness-enhancing defense mechanisms. We evaluate six GNN architectures based on GCN, SAGE, GIN, and GAT across five datasets from two distinct domains, employing four interpretability metrics: Fidelity, Stability, Consistency, and Sparsity. Our study examines how defenses against poisoning and evasion attacks, applied before and during model training, affect interpretability and highlights critical trade-offs between robustness and interpretability. The framework will be published as open source. The results reveal significant variations in interpretability depending on the chosen defense methods and model architecture characteristics. By establishing a standardized benchmark, this work provides a foundation for developing GNNs that are both robust to adversarial threats and interpretable, facilitating trust in their deployment in sensitive applications.

Maksim Golyadkin, Alexander Gambashidze, Ildar Nurgaliev et al.

In response to the growing demand for 3D object detection in applications such as autonomous driving, robotics, and augmented reality, this work focuses on the evaluation of semi-supervised learning approaches for point cloud data. The point cloud representation provides reliable and consistent observations regardless of lighting conditions, thanks to advances in LiDAR sensors. Data annotation is of paramount importance in the context of LiDAR applications, and automating 3D data annotation with semi-supervised methods is a pivotal challenge that promises to reduce the associated workload and facilitate the emergence of cost-effective LiDAR solutions. Nevertheless, the task of semi-supervised learning in the context of unordered point cloud data remains formidable due to the inherent sparsity and incomplete shapes that hinder the generation of accurate pseudo-labels. In this study, we consider these challenges by posing the question: "To what extent does unlabelled data contribute to the enhancement of model performance?" We show that improvements from previous semi-supervised methods may not be as profound as previously thought. Our results suggest that simple grid search hyperparameter tuning applied to a supervised model can lead to state-of-the-art performance on the ONCE dataset, while the contribution of unlabelled data appears to be comparatively less exceptional.

Kirill Lukyanov, Mikhail Drobyshevskiy, Georgii Sazonov et al.

The growing need for Trusted AI (TAI) highlights the importance of interpretability and robustness in machine learning models. However, many existing tools overlook graph data and rarely combine these two aspects into a single solution. Graph Neural Networks (GNNs) have become a popular approach, achieving top results across various tasks. We introduce GNN-AID (Graph Neural Network Analysis, Interpretation, and Defense), an open-source framework designed for graph data to address this gap. Built as a Python library, GNN-AID supports advanced trust methods and architectural layers, allowing users to analyze graph datasets and GNN behavior using attacks, defenses, and interpretability methods. GNN-AID is built on PyTorch-Geometric, offering preloaded datasets, models, and support for any GNNs through customizable interfaces. It also includes a web interface with tools for graph visualization and no-code features like an interactive model builder, simplifying the exploration and analysis of GNNs. The framework also supports MLOps techniques, ensuring reproducibility and result versioning to track and revisit analyses efficiently. GNN-AID is a flexible tool for developers and researchers. It helps developers create, analyze, and customize graph models, while also providing access to prebuilt datasets and models for quick experimentation. Researchers can use the framework to explore advanced topics on the relationship between interpretability and robustness, test defense strategies, and combine methods to protect against different types of attacks. We also show how defenses against evasion and poisoning attacks can conflict when applied to graph data, highlighting the complex connections between defense strategies. GNN-AID is available at \href{https://github.com/ispras/GNN-AID}{github.com/ispras/GNN-AID}

Alexander Gambashidze, Anton Kulikov, Yuriy Sosnin et al.

Recent advancements in human preference optimization, initially developed for Language Models (LMs), have shown promise for text-to-image Diffusion Models, enhancing prompt alignment, visual appeal, and user preference. Unlike LMs, Diffusion Models typically optimize in pixel or VAE space, which does not align well with human perception, leading to slower and less efficient training during the preference alignment stage. We propose using a perceptual objective in the U-Net embedding space of the diffusion model to address these issues. Our approach involves fine-tuning Stable Diffusion 1.5 and XL using Direct Preference Optimization (DPO), Contrastive Preference Optimization (CPO), and supervised fine-tuning (SFT) within this embedding space. This method significantly outperforms standard latent-space implementations across various metrics, including quality and computational cost. For SDXL, our approach provides 60.8\% general preference, 62.2\% visual appeal, and 52.1\% prompt following against original open-sourced SDXL-DPO on the PartiPrompts dataset, while significantly reducing compute. Our approach not only improves the efficiency and quality of human preference alignment for diffusion models but is also easily integrable with other optimization techniques. The training code and LoRA weights will be available here: https://huggingface.co/alexgambashidze/SDXL\_NCP-DPO\_v0.1

Boris Tseytlin, Ilya Makarov

During a long-running pandemic a pathogen can mutate, producing new strains with different epidemiological parameters. Existing approaches to epidemic modelling only consider one virus strain. We have developed a modified SEIR model to simulate multiple virus strains within the same population. As a case study, we investigate the potential effects of SARS-CoV-2 strain B.1.1.7 on the city of Moscow. Our analysis indicates a high risk of a new wave of infections in September-October 2021 with up to 35 000 daily infections at peak. We open-source our code and data.

Boris Tseytlin, Ilya Makarov

We approach the problem of hotel recognition with deep metric learning. We overview the existing approaches and propose a modification to Contrastive loss called Contrastive-Triplet loss. We construct a robust pipeline for benchmarking metric learning models and perform experiments on Hotels-50K and CUB200 datasets. Contrastive-Triplet loss is shown to achieve better retrieval on Hotels-50k. We open-source our code.

Daniil Storonkin, Ilia Dziub, Maksim Golyadkin et al.

Scrap quality directly affects energy use, emissions, and safety in steelmaking. Today, the share of non-metallic inclusions (contamination) is judged visually by inspectors - an approach that is subjective and hazardous due to dust and moving machinery. We present an assistive computer vision pipeline that estimates contamination (per percent) from images captured during railcar unloading and also classifies scrap type. The method formulates contamination assessment as a regression task at the railcar level and leverages sequential data through multi-instance learning (MIL) and multi-task learning (MTL). Best results include MAE 0.27 and R2 0.83 by MIL; and an MTL setup reaches MAE 0.36 with F1 0.79 for scrap class. Also we present the system in near real time within the acceptance workflow: magnet/railcar detection segments temporal layers, a versioned inference service produces railcar-level estimates with confidence scores, and results are reviewed by operators with structured overrides; corrections and uncertain cases feed an active-learning loop for continual improvement. The pipeline reduces subjective variability, improves human safety, and enables integration into acceptance and melt-planning workflows.

Dmitry Zhevnenko, Ilya Makarov, Aleksandr Kovalenko et al.

Anomaly detection (AD) for safety-critical IoT time series should be judged at the event level: reliability and earliness under realistic perturbations. Yet many studies still emphasize point-level results on curated base datasets, limiting value for model selection in practice. We introduce an evaluation protocol with unified event-level augmentations that simulate real-world issues: calibrated sensor dropout, linear and log drift, additive noise, and window shifts. We also perform sensor-level probing via mask-as-missing zeroing with per-channel influence estimation to support root-cause analysis. We evaluate 14 representative models on five public anomaly datasets (SWaT, WADI, SMD, SKAB, TEP) and two industrial datasets (steam turbine, nuclear turbogenerator) using unified splits and event aggregation. There is no universal winner: graph-structured models transfer best under dropout and long events (e.g., on SWaT under additive noise F1 drops 0.804->0.677 for a graph autoencoder, 0.759->0.680 for a graph-attention variant, and 0.762->0.756 for a hybrid graph attention model); density/flow models work well on clean stationary plants but can be fragile to monotone drift; spectral CNNs lead when periodicity is strong; reconstruction autoencoders become competitive after basic sensor vetting; predictive/hybrid dynamics help when faults break temporal dependencies but remain window-sensitive. The protocol also informs design choices: on SWaT under log drift, replacing normalizing flows with Gaussian density reduces high-stress F1 from ~0.75 to ~0.57, and fixing a learned DAG gives a small clean-set gain (~0.5-1.0 points) but increases drift sensitivity by ~8x.

Vitaliy Pozdnyakov, Aleksandr Kovalenko, Ilya Makarov et al.

Integrating machine learning into Automated Control Systems (ACS) enhances decision-making in industrial process management. One of the limitations to the widespread adoption of these technologies in industry is the vulnerability of neural networks to adversarial attacks. This study explores the threats in deploying deep learning models for fault diagnosis in ACS using the Tennessee Eastman Process dataset. By evaluating three neural networks with different architectures, we subject them to six types of adversarial attacks and explore five different defense methods. Our results highlight the strong vulnerability of models to adversarial samples and the varying effectiveness of defense strategies. We also propose a novel protection approach by combining multiple defense methods and demonstrate it's efficacy. This research contributes several insights into securing machine learning within ACS, ensuring robust fault diagnosis in industrial processes.

Elizabeth G. Campolongo, Yuan-Tang Chou, Ekaterina Govorkova et al.

Scientific discoveries are often made by finding a pattern or object that was not predicted by the known rules of science. Oftentimes, these anomalous events or objects that do not conform to the norms are an indication that the rules of science governing the data are incomplete, and something new needs to be present to explain these unexpected outliers. The challenge of finding anomalies can be confounding since it requires codifying a complete knowledge of the known scientific behaviors and then projecting these known behaviors on the data to look for deviations. When utilizing machine learning, this presents a particular challenge since we require that the model not only understands scientific data perfectly but also recognizes when the data is inconsistent and out of the scope of its trained behavior. In this paper, we present three datasets aimed at developing machine learning-based anomaly detection for disparate scientific domains covering astrophysics, genomics, and polar science. We present the different datasets along with a scheme to make machine learning challenges around the three datasets findable, accessible, interoperable, and reusable (FAIR). Furthermore, we present an approach that generalizes to future machine learning challenges, enabling the possibility of large, more compute-intensive challenges that can ultimately lead to scientific discovery.

Dzianis Piatrashyn, Nikita Kotelevskii, Kirill Grishchenkov et al.

Recently, LLM-based agents have become increasingly popular across many applications, including complex sequential decision-making problems. However, they inherit the tendency of LLMs to hallucinate, leading to incorrect decisions. In sequential settings, even a single mistake can irreversibly degrade the trajectory, making hallucinations an even bigger problem. Although larger LLMs hallucinate less, they incur a significantly higher per-token cost. In this paper, we address this tradeoff by proposing ReDAct (Reason-Defer-Act). In ReDAct, an agent is equipped with two LLMs: a small, cheap model used by default, and a large, more reliable but expensive model. When the predictive uncertainty of the small model exceeds a calibrated threshold, the decision is deferred to the large model. We evaluate our approach in text-based embodied environments such as ALFWorld and MiniGrid and show that deferring only about 15% of decisions to the large model can match the quality of using it exclusively, while significantly reducing inference costs.

Nikita Severin, Aleksei Ziablitsev, Yulia Savelyeva et al.

We present LLM-KT, a flexible framework designed to enhance collaborative filtering (CF) models by seamlessly integrating LLM (Large Language Model)-generated features. Unlike existing methods that rely on passing LLM-generated features as direct inputs, our framework injects these features into an intermediate layer of any CF model, allowing the model to reconstruct and leverage the embeddings internally. This model-agnostic approach works with a wide range of CF models without requiring architectural changes, making it adaptable to various recommendation scenarios. Our framework is built for easy integration and modification, providing researchers and developers with a powerful tool for extending CF model capabilities through efficient knowledge transfer. We demonstrate its effectiveness through experiments on the MovieLens and Amazon datasets, where it consistently improves baseline CF models. Experimental studies showed that LLM-KT is competitive with the state-of-the-art methods in context-aware settings but can be applied to a broader range of CF models than current approaches.

Mark Kashirskiy, Ilya Makarov

We propose Strategy-aware Surprise (SuS), a novel intrinsic motivation framework that uses pre-post prediction mismatch as a novelty signal for exploration in reinforcement learning. Unlike traditional curiosity-driven methods that rely solely on state prediction error, SuS introduces two complementary components: Strategy Stability (SS) and Strategy Surprise (SuS). SS measures consistency in behavioral strategy across temporal steps, while SuS captures unexpected outcomes relative to the agent's current strategy representation. Our combined reward formulation leverages both signals through learned weighting coefficients. We evaluate SuS on mathematical reasoning tasks using large language models, demonstrating significant improvements in both accuracy and solution diversity. Ablation studies confirm that removing either component results in at least 10% performance degradation, validating the synergistic nature of our approach. SuS achieves 17.4% improvement in Pass@1 and 26.4% improvement in Pass@5 compared to baseline methods, while maintaining higher strategy diversity throughout training.

Aleksandr Razin, Danil Kazantsev, Ilya Makarov

Diffusion models struggle to scale beyond their training resolutions, as direct high-resolution sampling is slow and costly, while post-hoc image super-resolution (ISR) introduces artifacts and additional latency by operating after decoding. We present the Latent Upscaler Adapter (LUA), a lightweight module that performs super-resolution directly on the generator's latent code before the final VAE decoding step. LUA integrates as a drop-in component, requiring no modifications to the base model or additional diffusion stages, and enables high-resolution synthesis through a single feed-forward pass in latent space. A shared Swin-style backbone with scale-specific pixel-shuffle heads supports 2x and 4x factors and remains compatible with image-space SR baselines, achieving comparable perceptual quality with nearly 3x lower decoding and upscaling time (adding only +0.42 s for 1024 px generation from 512 px, compared to 1.87 s for pixel-space SR using the same SwinIR architecture). Furthermore, LUA shows strong generalization across the latent spaces of different VAEs, making it easy to deploy without retraining from scratch for each new decoder. Extensive experiments demonstrate that LUA closely matches the fidelity of native high-resolution generation while offering a practical and efficient path to scalable, high-fidelity image synthesis in modern diffusion pipelines.

Daniil Sukhorukov, Andrei Zakharov, Nikita Glazkov et al.

We present the Hierarchical AI-Meteorologist, an LLM-agent system that generates explainable weather reports using a hierarchical forecast reasoning and weather keyword generation. Unlike standard approaches that treat forecasts as flat time series, our framework performs multi-scale reasoning across hourly, 6-hour, and daily aggregations to capture both short-term dynamics and long-term trends. Its core reasoning agent converts structured meteorological inputs into coherent narratives while simultaneously extracting a few keywords effectively summarizing the dominant meteorological events. These keywords serve as semantic anchors for validating consistency, temporal coherence and factual alignment of the generated reports. Using OpenWeather and Meteostat data, we demonstrate that hierarchical context and keyword-based validation substantially improve interpretability and robustness of LLM-generated weather narratives, offering a reproducible framework for semantic evaluation of automated meteorological reporting and advancing agent-based scientific reasoning.

Igor Abramov, Ilya Makarov

Existing EEG-driven image reconstruction methods often overlook spatial attention mechanisms, limiting fidelity and semantic coherence. To address this, we propose a dual-conditioning framework that combines EEG embeddings with spatial saliency maps to enhance image generation. Our approach leverages the Adaptive Thinking Mapper (ATM) for EEG feature extraction and fine-tunes Stable Diffusion 2.1 via Low-Rank Adaptation (LoRA) to align neural signals with visual semantics, while a ControlNet branch conditions generation on saliency maps for spatial control. Evaluated on THINGS-EEG, our method achieves a significant improvement in the quality of low- and high-level image features over existing approaches. Simultaneously, strongly aligning with human visual attention. The results demonstrate that attentional priors resolve EEG ambiguities, enabling high-fidelity reconstructions with applications in medical diagnostics and neuroadaptive interfaces, advancing neural decoding through efficient adaptation of pre-trained diffusion models.

Akim Tsvigun, Daniil Vasilev, Ivan Tsvigun et al.

Active learning (AL) has demonstrated remarkable potential in reducing the annotation effort required for training machine learning models. However, despite the surging popularity of natural language generation (NLG) tasks in recent years, the application of AL to NLG has been limited. In this paper, we introduce Active Text Generation (ATGen) - a comprehensive framework that bridges AL with text generation tasks, enabling the application of state-of-the-art AL strategies to NLG. Our framework simplifies AL-empowered annotation in NLG tasks using both human annotators and automatic annotation agents based on large language models (LLMs). The framework supports LLMs deployed as services, such as ChatGPT and Claude, or operated on-premises. Furthermore, ATGen provides a unified platform for smooth implementation and benchmarking of novel AL strategies tailored to NLG tasks. Finally, we present evaluation results for state-of-the-art AL strategies across diverse settings and multiple text generation tasks. We show that ATGen reduces both the effort of human annotators and costs associated with API calls to LLM-based annotation agents. The code of the framework is available on GitHub under the MIT license. The video presentation is available at http://atgen-video.nlpresearch.group

Andrei Chernov, Vitaliy Pozdnyakov, Ilya Makarov

Recent work in time series forecasting has explored reformulating regression as a classification task. By discretizing the continuous target space into bins and predicting over a fixed set of classes, these approaches benefit from more stable training, improved uncertainty modeling, and compatibility with modern deep learning architectures. However, most existing methods rely on one-hot encoding, which ignores the inherent ordinal structure of the target values. As a result, they fail to convey information about the relative distance between predicted and true values during training. In this paper, we address this limitation by applying \textbf{Cumulative Binary Encoding} (CBE), a monotonic binary representation that transforms both model inputs and outputs. CBE implicitly preserves ordinal and magnitude information, allowing models to learn distance aware representations while operating within a classification framework. To leverage CBE effectively, we propose \textbf{BinConv}, a fully convolutional neural network architecture designed for probabilistic forecasting. We demonstrate that standard fully connected layers are not only less computationally efficient than convolutional layers when used with CBE, but also degrade forecasting performance. Our experiments on standard benchmark datasets show that BinConv achieves superior performance compared to widely used baselines in both point and probabilistic forecasting, while requiring fewer parameters and enabling faster training.

Mikhail Mozikov, Nikita Severin, Valeria Bodishtianu et al.

Behavior study experiments are an important part of society modeling and understanding human interactions. In practice, many behavioral experiments encounter challenges related to internal and external validity, reproducibility, and social bias due to the complexity of social interactions and cooperation in human user studies. Recent advances in Large Language Models (LLMs) have provided researchers with a new promising tool for the simulation of human behavior. However, existing LLM-based simulations operate under the unproven hypothesis that LLM agents behave similarly to humans as well as ignore a crucial factor in human decision-making: emotions. In this paper, we introduce a novel methodology and the framework to study both, the decision-making of LLMs and their alignment with human behavior under emotional states. Experiments with GPT-3.5 and GPT-4 on four games from two different classes of behavioral game theory showed that emotions profoundly impact the performance of LLMs, leading to the development of more optimal strategies. While there is a strong alignment between the behavioral responses of GPT-3.5 and human participants, particularly evident in bargaining games, GPT-4 exhibits consistent behavior, ignoring induced emotions for rationality decisions. Surprisingly, emotional prompting, particularly with `anger' emotion, can disrupt the "superhuman" alignment of GPT-4, resembling human emotional responses.

Ilya Makarov, Andrey Savchenko, Arseny Korovko et al.

Many tasks in graph machine learning, such as link prediction and node classification, are typically solved by using representation learning, in which each node or edge in the network is encoded via an embedding. Though there exists a lot of network embeddings for static graphs, the task becomes much more complicated when the dynamic (i.e. temporal) network is analyzed. In this paper, we propose a novel approach for dynamic network representation learning based on Temporal Graph Network by using a highly custom message generating function by extracting Causal Anonymous Walks. For evaluation, we provide a benchmark pipeline for the evaluation of temporal network embeddings. This work provides the first comprehensive comparison framework for temporal network representation learning in every available setting for graph machine learning problems involving node classification and link prediction. The proposed model outperforms state-of-the-art baseline models. The work also justifies the difference between them based on evaluation in various transductive/inductive edge/node classification tasks. In addition, we show the applicability and superior performance of our model in the real-world downstream graph machine learning task provided by one of the top European banks, involving credit scoring based on transaction data.