A. Savchenko

Mikhail Shirokikh, Ilya Shenbin, Anton Alekseev et al.

We develop and evaluate neural architectures to model the user behavior in recommender systems (RS) inspired by click models for Web search but going beyond standard click models. Proposed architectures include recurrent networks, Transformer-based models that alleviate the quadratic complexity of self-attention, adversarial and hierarchical architectures. Our models outperform baselines on the ContentWise and RL4RS datasets and can be used in RS simulators to model user response for RS evaluation and pretraining.

Andrey V. Savchenko

In this paper, we describe the results of the HSEmotion team in two tasks of the seventh Affective Behavior Analysis in-the-wild (ABAW) competition, namely, multi-task learning for simultaneous prediction of facial expression, valence, arousal, and detection of action units, and compound expression recognition. We propose an efficient pipeline based on frame-level facial feature extractors pre-trained in multi-task settings to estimate valence-arousal and basic facial expressions given a facial photo. We ensure the privacy-awareness of our techniques by using the lightweight architectures of neural networks, such as MT-EmotiDDAMFN, MT-EmotiEffNet, and MT-EmotiMobileFaceNet, that can run even on a mobile device without the need to send facial video to a remote server. It was demonstrated that a significant step in improving the overall accuracy is the smoothing of neural network output scores using Gaussian or box filters. It was experimentally demonstrated that such a simple post-processing of predictions from simple blending of two top visual models improves the F1-score of facial expression recognition up to 7%. At the same time, the mean Concordance Correlation Coefficient (CCC) of valence and arousal is increased by up to 1.25 times compared to each model's frame-level predictions. As a result, our final performance score on the validation set from the multi-task learning challenge is 4.5 times higher than the baseline (1.494 vs 0.32).

Alina Kostromina, Kseniia Kuvshinova, Aleksandr Yugay et al.

While current time series research focuses on developing new models, crucial questions of selecting an optimal approach for training such models are underexplored. Tsururu, a Python library introduced in this paper, bridges SoTA research and industry by enabling flexible combinations of global and multivariate approaches and multi-step-ahead forecasting strategies. It also enables seamless integration with various forecasting models. Available at https://github.com/sb-ai-lab/tsururu .

Daria Tikhonovich, Nikita Zelinskiy, Aleksandr V. Petrov et al.

Since their introduction, Transformer-based models, such as SASRec and BERT4Rec, have become common baselines for sequential recommendations, surpassing earlier neural and non-neural methods. A number of following publications have shown that the effectiveness of these models can be improved by, for example, slightly updating the architecture of the Transformer layers, using better training objectives, and employing improved loss functions. However, the additivity of these modular improvements has not been systematically benchmarked - this is the gap we aim to close in this paper. Through our experiments, we identify a very strong model that uses SASRec's training objective, LiGR Transformer layers, and Sampled Softmax Loss. We call this combination eSASRec (Enhanced SASRec). While we primarily focus on realistic, production-like evaluation, in our preliminarily study we find that common academic benchmarks show eSASRec to be 23% more effective compared to the most recent state-of-the-art models, such as ActionPiece. In our main production-like benchmark, eSASRec resides on the Pareto frontier in terms of the accuracy-coverage tradeoff (alongside the recent industrial models HSTU and FuXi. As the modifications compared to the original SASRec are relatively straightforward and no extra features are needed (such as timestamps in HSTU), we believe that eSASRec can be easily integrated into existing recommendation pipelines and can can serve as a strong yet very simple baseline for emerging complicated algorithms. To facilitate this, we provide the open-source implementations for our models and benchmarks in repository https://github.com/blondered/transformer_benchmark

Aleksey Lapin, Igor Hromov, Stanislav Chumakov et al.

AutoML has advanced in handling complex tasks using the integration of LLMs, yet its efficiency remains limited by dependence on specific underlying tools. In this paper, we introduce LightAutoDS-Tab, a multi-AutoML agentic system for tasks with tabular data, which combines an LLM-based code generation with several AutoML tools. Our approach improves the flexibility and robustness of pipeline design, outperforming state-of-the-art open-source solutions on several data science tasks from Kaggle. The code of LightAutoDS-Tab is available in the open repository https://github.com/sb-ai-lab/LADS

Andrey V. Savchenko

In this paper, the multi-task learning of lightweight convolutional neural networks is studied for face identification and classification of facial attributes (age, gender, ethnicity) trained on cropped faces without margins. The necessity to fine-tune these networks to predict facial expressions is highlighted. Several models are presented based on MobileNet, EfficientNet and RexNet architectures. It was experimentally demonstrated that they lead to near state-of-the-art results in age, gender and race recognition on the UTKFace dataset and emotion classification on the AffectNet dataset. Moreover, it is shown that the usage of the trained models as feature extractors of facial regions in video frames leads to 4.5% higher accuracy than the previously known state-of-the-art single models for the AFEW and the VGAF datasets from the EmotiW challenges. The models and source code are publicly available at https://github.com/HSE-asavchenko/face-emotion-recognition.

A. V. Savchenko, K. V. Demochkin, I. S. Grechikhin

In this paper user modeling task is examined by processing a gallery of photos and videos on a mobile device. We propose novel engine for user preference prediction based on scene recognition, object detection and facial analysis. At first, all faces in a gallery are clustered and all private photos and videos with faces from large clusters are processed on the embedded system in offline mode. Other photos may be sent to the remote server to be analyzed by very deep models. The visual features of each photo are obtained from scene recognition and object detection models. These features are aggregated into a single user descriptor in the neural attention block. The proposed pipeline is implemented for the Android mobile platform. Experimental results with a subset of Photo Event Collection, Web Image Dataset for Event Recognition and Amazon Fashion datasets demonstrate the possibility to process images very efficiently without significant accuracy degradation. The source code of Android mobile application is publicly available at https://github.com/HSE-asavchenko/mobile-visual-preferences.

Valeriy Shevchenko, Nikita Belousov, Alexey Vasilev et al.

In the rapidly evolving domain of Recommender Systems (RecSys), new algorithms frequently claim state-of-the-art performance based on evaluations over a limited set of arbitrarily selected datasets. However, this approach may fail to holistically reflect their effectiveness due to the significant impact of dataset characteristics on algorithm performance. Addressing this deficiency, this paper introduces a novel benchmarking methodology to facilitate a fair and robust comparison of RecSys algorithms, thereby advancing evaluation practices. By utilizing a diverse set of $30$ open datasets, including two introduced in this work, and evaluating $11$ collaborative filtering algorithms across $9$ metrics, we critically examine the influence of dataset characteristics on algorithm performance. We further investigate the feasibility of aggregating outcomes from multiple datasets into a unified ranking. Through rigorous experimental analysis, we validate the reliability of our methodology under the variability of datasets, offering a benchmarking strategy that balances quality and computational demands. This methodology enables a fair yet effective means of evaluating RecSys algorithms, providing valuable guidance for future research endeavors.

Alexandra Bazarova, Maria Kovaleva, Ilya Kuleshov et al.

In today's world, banks use artificial intelligence to optimize diverse business processes, aiming to improve customer experience. Most of the customer-related tasks can be categorized into two groups: 1) local ones, which focus on a client's current state, such as transaction forecasting, and 2) global ones, which consider the general customer behaviour, e.g., predicting successful loan repayment. Unfortunately, maintaining separate models for each task is costly. Therefore, to better facilitate information management, we compared eight state-of-the-art unsupervised methods on 11 tasks in search for a one-size-fits-all solution. Contrastive self-supervised learning methods were demonstrated to excel at global problems, while generative techniques were superior at local tasks. We also introduced a novel approach, which enriches the client's representation by incorporating external information gathered from other clients. Our method outperforms classical models, boosting accuracy by up to 20\%.

Andrey Savchenko, Oleg Kachan

Accurate forecasting of multivariate time series data remains a formidable challenge, particularly due to the growing complexity of temporal dependencies in real-world scenarios. While neural network-based models have achieved notable success in this domain, complex channel-dependent models often suffer from performance degradation compared to channel-independent models that do not consider the relationship between components but provide high robustness due to small capacity. In this work, we propose HN-MVTS, a novel architecture that integrates a hypernetwork-based generative prior with an arbitrary neural network forecasting model. The input of this hypernetwork is a learnable embedding matrix of time series components. To restrict the number of new parameters, the hypernetwork learns to generate the weights of the last layer of the target forecasting networks, serving as a data-adaptive regularizer that improves generalization and long-range predictive accuracy. The hypernetwork is used only during the training, so it does not increase the inference time compared to the base forecasting model. Extensive experiments on eight benchmark datasets demonstrate that application of HN-MVTS to the state-of-the-art models (DLinear, PatchTST, TSMixer, etc.) typically improves their performance. Our findings suggest that hypernetwork-driven parameterization offers a promising direction for enhancing existing forecasting techniques in complex scenarios.

Oleg Kachan, Andrey Savchenko, Gleb Gusev

SMOTE (Synthetic Minority Oversampling Technique) is the established geometric approach to random oversampling to balance classes in the imbalanced learning problem, followed by many extensions. Its idea is to introduce synthetic data points of the minor class, with each new point being the convex combination of an existing data point and one of its k-nearest neighbors. In this paper, by viewing SMOTE as sampling from the edges of a geometric neighborhood graph and borrowing tools from the topological data analysis, we propose a novel technique, Simplicial SMOTE, that samples from the simplices of a geometric neighborhood simplicial complex. A new synthetic point is defined by the barycentric coordinates w.r.t. a simplex spanned by an arbitrary number of data points being sufficiently close rather than a pair. Such a replacement of the geometric data model results in better coverage of the underlying data distribution compared to existing geometric sampling methods and allows the generation of synthetic points of the minority class closer to the majority class on the decision boundary. We experimentally demonstrate that our Simplicial SMOTE outperforms several popular geometric sampling methods, including the original SMOTE. Moreover, we show that simplicial sampling can be easily integrated into existing SMOTE extensions. We generalize and evaluate simplicial extensions of the classic Borderline SMOTE, Safe-level SMOTE, and ADASYN algorithms, all of which outperform their graph-based counterparts.

Ivan Karpukhin, Andrey Savchenko

Deep learning has achieved remarkable success in modeling sequential data, including event sequences, temporal point processes, and irregular time series. Recently, transformers have largely replaced recurrent networks in these tasks. However, transformers often underperform RNNs in classification tasks where the objective is to predict future targets. The reason behind this performance gap remains largely unexplored. In this paper, we identify a key limitation of transformers: the absence of a single state vector that provides a compact and effective representation of the entire sequence. Additionally, we show that contrastive pretraining of embedding vectors fails to capture local context, which is crucial for accurate prediction. To address these challenges, we introduce history tokens, a novel concept that facilitates the accumulation of historical information during next-token prediction pretraining. Our approach significantly improves transformer-based models, achieving impressive results in finance, e-commerce, and healthcare tasks. The code is publicly available on GitHub.

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.

Ilia Revin, Leon Strelkov, Vadim A. Potemkin et al.

There are many critical challenges in optimizing neural network models, including distributed computing, compression techniques, and efficient training, regardless of their application to specific tasks. Solving such problems is crucial because the need for scalable and resource-efficient models is increasing. To address these challenges, we have developed a new automated machine learning (AutoML) framework, Parameter Efficient Training with Robust Automation (PETRA). It applies evolutionary optimization to model architecture and training strategy. PETRA includes pruning, quantization, and loss regularization. Experimental studies on real-world data with financial event sequences, as well as image and time-series -- benchmarks, demonstrate PETRA's ability to improve neural model performance and scalability -- namely, a significant decrease in model size (up to 75%) and latency (up to 33%), and an increase in throughput (by 13%) without noticeable degradation in the target metric.

Konstantin Egorov, Stepan Botman, Pavel Blinov et al.

Progress in remote PhotoPlethysmoGraphy (rPPG) is limited by the critical issues of existing publicly available datasets: small size, privacy concerns with facial videos, and lack of diversity in conditions. The paper introduces a novel comprehensive large-scale multi-view video dataset for rPPG and health biomarkers estimation. Our dataset comprises 3600 synchronized video recordings from 600 subjects, captured under varied conditions (resting and post-exercise) using multiple consumer-grade cameras at different angles. To enable multimodal analysis of physiological states, each recording is paired with a 100 Hz PPG signal and extended health metrics, such as electrocardiogram, arterial blood pressure, biomarkers, temperature, oxygen saturation, respiratory rate, and stress level. Using this data, we train an efficient rPPG model and compare its quality with existing approaches in cross-dataset scenarios. The public release of our dataset and model should significantly speed up the progress in the development of AI medical assistants.

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.

Sokolov Artem, Andrey V. Savchenko

This paper is focused on the finetuning of acoustic models for speaker adaptation goals on a given gender. We pretrained the Transformer baseline model on Librispeech-960 and conduct experiments with finetuning on the gender-specific test subsets and. In general, we do not obtain essential WER reduction by finetuning techniques by this approach. We achieved up to ~5% lower word error rate on the male subset and 3% on the female subset if the layers in the encoder and decoder are not frozen, but the tuning is started from the last checkpoints. Moreover, we adapted our base model on the full L2 Arctic dataset of accented speech and fine-tuned it for particular speakers and male and female genders separately. The models trained on the gender subsets obtained 1-2% higher accuracy when compared to the model tuned on the whole L2 Arctic dataset. Finally, we tested the concatenation of the pretrained x-vector voice embeddings and embeddings from a conventional encoder, but its gain in accuracy is not significant.

Andrey V. Savchenko

In this paper a new formulation of event recognition task is examined: it is required to predict event categories in a gallery of images, for which albums (groups of photos corresponding to a single event) are unknown. We propose the novel two-stage approach. At first, features are extracted in each photo using the pre-trained convolutional neural network. These features are classified individually. The scores of the classifier are used to group sequential photos into several clusters. Finally, the features of photos in each group are aggregated into a single descriptor using neural attention mechanism. This algorithm is optionally extended to improve the accuracy for classification of each image in an album. In contrast to conventional fine-tuning of convolutional neural networks (CNN) we proposed to use image captioning, i.e., generative model that converts images to textual descriptions. They are one-hot encoded and summarized into sparse feature vector suitable for learning of arbitrary classifier. Experimental study with Photo Event Collection and Multi-Label Curation of Flickr Events Dataset demonstrates that our approach is 9-20% more accurate than event recognition on single photos. Moreover, proposed method has 13-16% lower error rate than classification of groups of photos obtained with hierarchical clustering. It is experimentally shown that the image captions trained on Conceptual Captions dataset can be classified more accurately than the features from object detector, though they both are obviously not as rich as the CNN-based features. However, it is possible to combine our approach with conventional CNNs in an ensemble to provide the state-of-the-art results for several event datasets.

Artem M. Grachev, Dmitry I. Ignatov, Andrey V. Savchenko

Recurrent neural networks have proved to be an effective method for statistical language modeling. However, in practice their memory and run-time complexity are usually too large to be implemented in real-time offline mobile applications. In this paper we consider several compression techniques for recurrent neural networks including Long-Short Term Memory models. We make particular attention to the high-dimensional output problem caused by the very large vocabulary size. We focus on effective compression methods in the context of their exploitation on devices: pruning, quantization, and matrix decomposition approaches (low-rank factorization and tensor train decomposition, in particular). For each model we investigate the trade-off between its size, suitability for fast inference and perplexity. We propose a general pipeline for applying the most suitable methods to compress recurrent neural networks for language modeling. It has been shown in the experimental study with the Penn Treebank (PTB) dataset that the most efficient results in terms of speed and compression-perplexity balance are obtained by matrix decomposition techniques.

Andrey V. Savchenko

This paper is focused on the automatic extraction of persons and their attributes (gender, year of born) from album of photos and videos. We propose the two-stage approach, in which, firstly, the convolutional neural network simultaneously predicts age/gender from all photos and additionally extracts facial representations suitable for face identification. We modified the MobileNet, which is preliminarily trained to perform face recognition, in order to additionally recognize age and gender. In the second stage of our approach, extracted faces are grouped using hierarchical agglomerative clustering techniques. The born year and gender of a person in each cluster are estimated using aggregation of predictions for individual photos. We experimentally demonstrated that our facial clustering quality is competitive with the state-of-the-art neural networks, though our implementation is much computationally cheaper. Moreover, our approach is characterized by more accurate video-based age/gender recognition when compared to the publicly available models.

Andrey V. Savchenko, Natalya S. Belova

The paper deals with the still-to-video face recognition for the small sample size problem based on computation of distances between high-dimensional deep bottleneck features. We present the novel statistical recognition method, in which the still-to-video recognition task is casted into Maximum A Posteriori estimation. In this method we maximize the joint probabilistic density of the distances to all reference still images. It is shown that this likelihood can be estimated with the known asymptotically normal distribution of the Kullback-Leibler discriminations between nonnegative features. The experimental study with the LFW (Labeled Faces in the Wild), YTF (YouTube Faces) and IJB-A (IARPA Janus Benchmark A) datasets has been provided. We demonstrated, that the proposed approach can be applied with the state-of-the-art deep features and dissimilarity measures. Our algorithm achieves 3-5% higher accuracy when compared with conventional aggregation of decisions obtained for all frames.

Artem M. Grachev, Dmitry I. Ignatov, Andrey V. Savchenko

In this paper, we consider several compression techniques for the language modeling problem based on recurrent neural networks (RNNs). It is known that conventional RNNs, e.g, LSTM-based networks in language modeling, are characterized with either high space complexity or substantial inference time. This problem is especially crucial for mobile applications, in which the constant interaction with the remote server is inappropriate. By using the Penn Treebank (PTB) dataset we compare pruning, quantization, low-rank factorization, tensor train decomposition for LSTM networks in terms of model size and suitability for fast inference.

Andrey Savchenko

If the training dataset is not very large, image recognition is usually implemented with the transfer learning methods. In these methods the features are extracted using a deep convolutional neural network, which was preliminarily trained with an external very-large dataset. In this paper we consider the nonparametric classification of extracted feature vectors with the probabilistic neural network (PNN). The number of neurons at the pattern layer of the PNN is equal to the database size, which causes the low recognition performance and high memory space complexity of this network. We propose to overcome these drawbacks by replacing the exponential activation function in the Gaussian Parzen kernel to the complex exponential functions in the Fejér kernel. We demonstrate that in this case it is possible to implement the network with the number of neurons in the pattern layer proportional to the cubic root of the database size. Thus, the proposed modification of the PNN makes it possible to significantly decrease runtime and memory complexities without loosing its main advantages, namely, extremely fast training procedure and the convergence to the optimal Bayesian decision. An experimental study in visual object category classification and unconstrained face recognition with contemporary deep neural networks have shown, that our approach obtains very efficient and rather accurate decisions for the small training sample in comparison with the well-known classifiers.

Andrey Savchenko

We explore the problems of classification of composite object (images, speech signals) with low number of models per class. We study the question of improving recognition performance for medium-sized database (thousands of classes). The key issue of fast approximate nearest-neighbor methods widely applied in this task is their heuristic nature. It is possible to strongly prove their efficiency by using the theory of algorithms only for simple similarity measures and artificially generated tasks. On the contrary, in this paper we propose an alternative, statistically optimal greedy algorithm. At each step of this algorithm joint density (likelihood) of distances to previously checked models is estimated for each class. The next model to check is selected from the class with the maximal likelihood. The latter is estimated based on the asymptotic properties of the Kullback-Leibler information discrimination and mathematical model of piecewise-regular object with distribution of each regular segment of exponential type. Experimental results in face recognition for FERET dataset prove that the proposed method is much more effective than not only brute force and the baseline (directed enumeration method) but also approximate nearest neighbor methods from FLANN and NonMetricSpaceLib libraries (randomized kd-tree, composite index, perm-sort).