arXiv subjects

Mikhail Kuznetsov

Publications and source records attributed to Mikhail Kuznetsov.

At least 19 recordsLinked to original sources

arXiv2026

Visual Implicit Geometry Transformer for Autonomous Driving

We introduce the Visual Implicit Geometry Transformer (ViGT), an autonomous driving geometric model that estimates continuous 3D occupancy fields from surround-view camera rigs. ViGT represents a step towards foundational geometric models for autonomous driving, prioritizing scalability, architectural simplicity, and generalization across diverse sensor configurations. Our approach achieves this through a calibration-free architecture, enabling a single model to adapt to different sensor setups. Unlike general-purpose geometric foundational models that focus on pixel-aligned predictions, ViGT estimates a continuous 3D occupancy field in a birds-eye-view (BEV) addressing domain-specific requirements. ViGT naturally infers geometry from multiple camera views into a single metric coordinate frame, providing a common representation for multiple geometric tasks. Unlike most existing occupancy models, we adopt a self-supervised training procedure that leverages synchronized image-LiDAR pairs, eliminating the need for costly manual annotations. We validate the scalability and generalizability of our approach by training our model on a mixture of five large-scale autonomous driving datasets (NuScenes, Waymo, NuPlan, ONCE, and Argoverse) and achieving state-of-the-art performance on the pointmap estimation task, with the best average rank across all evaluated baselines. We further evaluate ViGT on the Occ3D-nuScenes benchmark, where ViGT achieves comparable performance with supervised methods. The source code is publicly available at \href{https://github.com/whesense/ViGT}{https://github.com/whesense/ViGT}.

cs.CV↗

arXiv2025

STARS: Synchronous Token Alignment for Robust Supervision in Large Language Models

Aligning large language models (LLMs) with human values is crucial for safe deployment. Inference-time techniques offer granular control over generation; however, they rely on model uncertainty, meaning an internal estimate of how likely the model believes its next tokens or outputs are correct, for segmentation. We show that this introduces two critical limitations: (a) vulnerability to miscalibrated confident hallucinations and (b) poor hardware utilization due to asynchronous, ragged batch processing. Together, these issues reduce alignment reliability while increasing token and compute costs, which limits their practical scalability. To address these limitations, building on dynamic inference-time alignment methods, we introduce STARS, Synchronous Token Alignment for Robust Supervision, a decoding-time algorithm, which steers generation by enforcing verification at fixed-horizon intervals. By decoupling segmentation from confidence, STARS enables lockstep parallel execution and robustly detects errors that uncertainty metrics miss. On the HH-RLHF benchmark, we demonstrate that STARS achieves competitive alignment quality with that of state-of-the-art dynamic methods, while strictly bounding rejection costs and maximizing system throughput. Furthermore, it outperforms fine-tuning and several state-of-the-art inference-time decoding strategies by good margins, and establishes fixed-horizon sampling as a robust, system-efficient alternative for aligning LLMs at scale. The code is publicly available at https://github.com/purseclab/STARS.

cs.CL↗

arXiv2025

Adaptive Blockwise Search: Inference-Time Alignment for Large Language Models

LLM alignment remains a critical challenge. Inference-time methods provide a flexible alternative to fine-tuning, but their uniform computational effort often yields suboptimal alignment. We hypothesize that for many alignment tasks, the initial tokens of a response are disproportionately more critical. To leverage this principle, we introduce AdaSearch, a novel blockwise search strategy. It adaptively allocates a fixed computational budget using a sampling schedule, focusing search effort on these critical tokens. We apply AdaSearch to sequential decoding and introduce its tree-search counterpart, AdaBeam. Our comprehensive evaluation across eight LLMs demonstrates that AdaSearch outperforms strong Best-of-N and fine-tuning baselines. Specifically, win-rates improve by over 10% for harmlessness generation, controlled sentiment generation, and for mathematical reasoning tasks relative to Best-of-N.

cs.CL↗

arXiv2024

Distributionally robust self-supervised learning for tabular data

Machine learning (ML) models trained using Empirical Risk Minimization (ERM) often exhibit systematic errors on specific subpopulations of tabular data, known as error slices. Learning robust representation in presence of error slices is challenging, especially in self-supervised settings during the feature reconstruction phase, due to high cardinality features and the complexity of constructing error sets. Traditional robust representation learning methods are largely focused on improving worst group performance in supervised setting in computer vision, leaving a gap in approaches tailored for tabular data. We address this gap by developing a framework to learn robust representation in tabular data during self-supervised pre-training. Our approach utilizes an encoder-decoder model trained with Masked Language Modeling (MLM) loss to learn robust latent representations. This paper applies the Just Train Twice (JTT) and Deep Feature Reweighting (DFR) methods during the pre-training phase for tabular data. These methods fine-tune the ERM pre-trained model by up-weighting error-prone samples or creating balanced datasets for specific categorical features. This results in specialized models for each feature, which are then used in an ensemble approach to enhance downstream classification performance. This methodology improves robustness across slices, thus enhancing overall generalization performance. Extensive experiments across various datasets demonstrate the efficacy of our approach. The code is available: https://github.com/amazon-science/distributionally-robust-self-supervised-learning-for-tabular-data.

cs.LG↗

arXiv2024

HLogformer: A Hierarchical Transformer for Representing Log Data

Transformers have gained widespread acclaim for their versatility in handling diverse data structures, yet their application to log data remains underexplored. Log data, characterized by its hierarchical, dictionary-like structure, poses unique challenges when processed using conventional transformer models. Traditional methods often rely on manually crafted templates for parsing logs, a process that is labor-intensive and lacks generalizability. Additionally, the linear treatment of log sequences by standard transformers neglects the rich, nested relationships within log entries, leading to suboptimal representations and excessive memory usage. To address these issues, we introduce HLogformer, a novel hierarchical transformer framework specifically designed for log data. HLogformer leverages the hierarchical structure of log entries to significantly reduce memory costs and enhance representation learning. Unlike traditional models that treat log data as flat sequences, our framework processes log entries in a manner that respects their inherent hierarchical organization. This approach ensures comprehensive encoding of both fine-grained details and broader contextual relationships. Our contributions are threefold: First, HLogformer is the first framework to design a dynamic hierarchical transformer tailored for dictionary-like log data. Second, it dramatically reduces memory costs associated with processing extensive log sequences. Third, comprehensive experiments demonstrate that HLogformer more effectively encodes hierarchical contextual information, proving to be highly effective for downstream tasks such as synthetic anomaly detection and product recommendation.

cs.LG↗

arXiv2024

Salient Object-Aware Background Generation using Text-Guided Diffusion Models

Generating background scenes for salient objects plays a crucial role across various domains including creative design and e-commerce, as it enhances the presentation and context of subjects by integrating them into tailored environments. Background generation can be framed as a task of text-conditioned outpainting, where the goal is to extend image content beyond a salient object's boundaries on a blank background. Although popular diffusion models for text-guided inpainting can also be used for outpainting by mask inversion, they are trained to fill in missing parts of an image rather than to place an object into a scene. Consequently, when used for background creation, inpainting models frequently extend the salient object's boundaries and thereby change the object's identity, which is a phenomenon we call "object expansion." This paper introduces a model for adapting inpainting diffusion models to the salient object outpainting task using Stable Diffusion and ControlNet architectures. We present a series of qualitative and quantitative results across models and datasets, including a newly proposed metric to measure object expansion that does not require any human labeling. Compared to Stable Diffusion 2.0 Inpainting, our proposed approach reduces object expansion by 3.6x on average with no degradation in standard visual metrics across multiple datasets.

cs.CV↗

arXiv2024

HairFastGAN: Realistic and Robust Hair Transfer with a Fast Encoder-Based Approach

Our paper addresses the complex task of transferring a hairstyle from a reference image to an input photo for virtual hair try-on. This task is challenging due to the need to adapt to various photo poses, the sensitivity of hairstyles, and the lack of objective metrics. The current state of the art hairstyle transfer methods use an optimization process for different parts of the approach, making them inexcusably slow. At the same time, faster encoder-based models are of very low quality because they either operate in StyleGAN's W+ space or use other low-dimensional image generators. Additionally, both approaches have a problem with hairstyle transfer when the source pose is very different from the target pose, because they either don't consider the pose at all or deal with it inefficiently. In our paper, we present the HairFast model, which uniquely solves these problems and achieves high resolution, near real-time performance, and superior reconstruction compared to optimization problem-based methods. Our solution includes a new architecture operating in the FS latent space of StyleGAN, an enhanced inpainting approach, and improved encoders for better alignment, color transfer, and a new encoder for post-processing. The effectiveness of our approach is demonstrated on realism metrics after random hairstyle transfer and reconstruction when the original hairstyle is transferred. In the most difficult scenario of transferring both shape and color of a hairstyle from different images, our method performs in less than a second on the Nvidia V100. Our code is available at https://github.com/AIRI-Institute/HairFastGAN.

cs.CV↗

arXiv2023

Staging E-Commerce Products for Online Advertising using Retrieval Assisted Image Generation

Online ads showing e-commerce products typically rely on the product images in a catalog sent to the advertising platform by an e-commerce platform. In the broader ads industry such ads are called dynamic product ads (DPA). It is common for DPA catalogs to be in the scale of millions (corresponding to the scale of products which can be bought from the e-commerce platform). However, not all product images in the catalog may be appealing when directly re-purposed as an ad image, and this may lead to lower click-through rates (CTRs). In particular, products just placed against a solid background may not be as enticing and realistic as a product staged in a natural environment. To address such shortcomings of DPA images at scale, we propose a generative adversarial network (GAN) based approach to generate staged backgrounds for un-staged product images. Generating the entire staged background is a challenging task susceptible to hallucinations. To get around this, we introduce a simpler approach called copy-paste staging using retrieval assisted GANs. In copy paste staging, we first retrieve (from the catalog) staged products similar to the un-staged input product, and then copy-paste the background of the retrieved product in the input image. A GAN based in-painting model is used to fill the holes left after this copy-paste operation. We show the efficacy of our copy-paste staging method via offline metrics, and human evaluation. In addition, we show how our staging approach can enable animations of moving products leading to a video ad from a product image.

cs.CV↗

arXiv2023

Revising deep learning methods in parking lot occupancy detection

Parking guidance systems have recently become a popular trend as a part of the smart cities' paradigm of development. The crucial part of such systems is the algorithm allowing drivers to search for available parking lots across regions of interest. The classic approach to this task is based on the application of neural network classifiers to camera records. However, existing systems demonstrate a lack of generalization ability and appropriate testing regarding specific visual conditions. In this study, we extensively evaluate state-of-the-art parking lot occupancy detection algorithms, compare their prediction quality with the recently emerged vision transformers, and propose a new pipeline based on EfficientNet architecture. Performed computational experiments have demonstrated the performance increase in the case of our model, which was evaluated on 5 different datasets.

cs.LG↗

arXiv2021

The UHECR dipole and quadrupole in the latest data from the original Auger and TA surface detectors

The sources of ultra-high-energy cosmic rays are still unknown, but assuming standard physics, they are expected to lie within a few hundred megaparsecs from us. Indeed, over cosmological distances cosmic rays lose energy to interactions with background photons, at a rate depending on their mass number and energy and properties of photonuclear interactions and photon backgrounds. The universe is not homogeneous at such scales, hence the distribution of the arrival directions of cosmic rays is expected to reflect the inhomogeneities in the distribution of galaxies; the shorter the energy loss lengths, the stronger the expected anisotropies. Galactic and intergalactic magnetic fields can blur and distort the picture, but the magnitudes of the largest-scale anisotropies, namely the dipole and quadrupole moments, are the most robust to their effects. Measuring them with no bias regardless of any higher-order multipoles is not possible except with full-sky coverage. In this work, we achieve this in three energy ranges (approximately 8--16 EeV, 16--32 EeV, and 32--$\infty$ EeV) by combining surface-detector data collected at the Pierre Auger Observatory until 2020 and at the Telescope Array (TA) until 2019, before the completion of the upgrades of the arrays with new scintillator detectors. We find that the full-sky coverage achieved by combining Auger and TA data reduces the uncertainties on the north-south components of the dipole and quadrupole in half compared to Auger-only results.

astro-ph.HE↗

arXiv2021

UHECR arrival directions in the latest data from the original Auger and TA surface detectors and nearby galaxies

The distribution of ultra-high-energy cosmic-ray arrival directions appears to be nearly isotropic except for a dipole moment of order $6 \times (E/10~\mathrm{EeV})$ per cent. Nonetheless, at the highest energies, as the number of possible candidate sources within the propagation horizon and the magnetic deflections both shrink, smaller-scale anisotropies might be expected to emerge. On the other hand, the flux suppression reduces the statistics available for searching for such anisotropies. In this work, we consider two different lists of candidate sources: a sample of nearby starburst galaxies and the 2MRS catalog tracing stellar mass within $250~\mathrm{Mpc}$. We combine surface-detector data collected at the Pierre Auger Observatory until 2020 and the Telescope Array until 2019, and use them to test models in which UHECRs comprise an isotropic background and a foreground originating from the candidate sources and randomly deflected by magnetic fields. The free parameters of these models are the energy threshold, the signal fraction, and the search angular scale. We find a correlation between the arrival directions of $11.8\%_{-3.1\%}^{+5.0\%}$ of cosmic rays detected with $E \ge 38~\mathrm{EeV}$ by Auger or with $E \gtrsim 49~\mathrm{EeV}$ by TA and the position of nearby starburst galaxies on a ${15.5^\circ}_{-3.2^\circ}^{+5.3^\circ}$ angular scale, with a $4.2\sigma$ post-trial significance, as well as a weaker correlation with the overall galaxy distribution.

astro-ph.HE↗

arXiv2021

Hunting super-heavy dark matter with ultra-high energy photons

At any epoch, particle physics must be open to completely unexpected discoveries, and that is reason enough to extend the reach of searches for ultra-high energy (UHE) photons. The observation of a population of photons with energies $E \gtrsim 100$ EeV would for example imply the existence of either a completely new physical phenomena, or particle acceleration mechanisms heretofore never seen or imagined. But as we outline in this Letter of Interest, there are also good arguments for super-heavy dark matter (SHDM) in a parameter range such that it could be discovered via its decays to, in particular, UHE photons. Only ultra-high energy cosmic ray observatories have capabilities to detect UHE photons. We first investigate how current and future observations can probe and constrain SHDM models in important directions, and then outline some of the scenarios that motivate such searches. We also discuss connections between constraints on SHDM and on the parameter values of cosmological models.

hep-ph↗

arXiv2021

Fast Parallel Hypertree Decompositions in Logarithmic Recursion Depth

Modern trends in data collection are bringing current mainstream techniques for database query processing to their limits. Consequently, various novel approaches for efficient query processing are being actively studied. One such approach is based on hypertree decompositions (HDs), which have been shown to carry great potential to process complex queries more efficiently and with stronger theoretical guarantees. However, using HDs for query execution relies on the difficult task of computing decompositions of the query structure, which guides the efficient execution of the query. From theoretical results we know that the performance of purely sequential methods is inherently limited, yet the problem is susceptible to parallelisation. In this paper we propose the first algorithm for computing hypertree decompositions that is well-suited for parallelisation. The proposed algorithm log-k-decomp requires only a logarithmic number of recursion levels and additionally allows for highly parallelised pruning of the search space by restriction to balanced separators. We provide detailed experimental evaluation over the HyperBench benchmark and demonstrate that our approach is highly effective especially for complex queries.

cs.DB↗

arXiv2020

Probabilistic Label Trees for Extreme Multi-label Classification

Extreme multi-label classification (XMLC) is a learning task of tagging instances with a small subset of relevant labels chosen from an extremely large pool of possible labels. Problems of this scale can be efficiently handled by organizing labels as a tree, like in hierarchical softmax used for multi-class problems. In this paper, we thoroughly investigate probabilistic label trees (PLTs) which can be treated as a generalization of hierarchical softmax for multi-label problems. We first introduce the PLT model and discuss training and inference procedures and their computational costs. Next, we prove the consistency of PLTs for a wide spectrum of performance metrics. To this end, we upperbound their regret by a function of surrogate-loss regrets of node classifiers. Furthermore, we consider a problem of training PLTs in a fully online setting, without any prior knowledge of training instances, their features, or labels. In this case, both node classifiers and the tree structure are trained online. We prove a specific equivalence between the fully online algorithm and an algorithm with a tree structure given in advance. Finally, we discuss several implementations of PLTs and introduce a new one, napkinXC, which we empirically evaluate and compare with state-of-the-art algorithms.

cs.LG↗

arXiv2020

Constraining superheavy decaying dark matter with directional ultra-high energy gamma-ray limits

Dwarf spheroidal galaxies are compact stellar objects with small or negligible astrophysical backgrounds, widely considered as promising targets to search for a signal from the dark matter decay and annihilation. We present constraints on the lifetime of the superheavy decaying dark matter branching to the $q\bar{q}$ channel in the mass range $10^{19} - 10^{25}$ eV based on the directional limits on the ultra-high-energy (UHE) gamma rays obtained by the Pierre Auger Observatory and the Telescope Array experiment. Attenuation effects during the propagation of UHE photons towards Earth are taken into account, with the strongest constraints derived for the Ursa Major II, Coma Berenices and Segue I galaxies.

astro-ph.HE↗

arXiv2019

Resolving isotope splitting of boron-related intracenter transitions in diamond by infrared absorption spectroscopy

Isotopic enrichment offers cutting-edge properties of materials opening exciting research and development opportunities. In semiconductors, reached progress of ultimate control in growth and doping techniques follows nowadays the high level isotopic purification. This requires deep understanding of isotopic disorder effects and techniques of their effective determination. Isotopic content of both crystal lattice and impurity centers cause the effects, which can be examined by different optical techniques. While disorder in the host lattice can be straight forward evaluated by inelastic light scattering or by SIMS measurements, determination of isotopic contributions of many orders less presented impurities remains challenging and usually observed in high-resolution photoluminescence or optical absorption spectra. Boron-doped diamonds exhibit complex infrared absorption spectra while boron-related luminescence remains unobserved. Boron, as a most light element acting as a hydrogen-like dopant in elemental semiconductors, has a largest relative difference in its isotope masses, and by this, cause the largest isotopic disorder in semiconductors, including diamond, an elemental semiconductor with the lightest atomic mass of a host lattice. This enables an access to the isotopic constitution of boron in diamond by infrared absorption spectroscopy. By comparison of low-temperature absorption spectra of a natural (20% of 10B and 80% of 11B isotopes) and 11B enriched (up to 99%) doped diamonds we differentiate the intracenter transitions related to 10B and to 11B isotopes. We have found that the isotopic spectral lines of the same boron intracenter transition are separated with the energy of about 0.7 meV. This is the largest impurity isotopic shift ever observed in semiconductors doped by hydrogen-like impurity centers.

cond-mat.mtrl-sci↗

arXiv2019

On the computational complexity of the probabilistic label tree algorithms

Label tree-based algorithms are widely used to tackle multi-class and multi-label problems with a large number of labels. We focus on a particular subclass of these algorithms that use probabilistic classifiers in the tree nodes. Examples of such algorithms are hierarchical softmax (HSM), designed for multi-class classification, and probabilistic label trees (PLTs) that generalize HSM to multi-label problems. If the tree structure is given, learning of PLT can be solved with provable regret guaranties [Wydmuch et.al. 2018]. However, to find a tree structure that results in a PLT with a low training and prediction computational costs as well as low statistical error seems to be a very challenging problem, not well-understood yet. In this paper, we address the problem of finding a tree structure that has low computational cost. First, we show that finding a tree with optimal training cost is NP-complete, nevertheless there are some tractable special cases with either perfect approximation or exact solution that can be obtained in linear time in terms of the number of labels $m$. For the general case, we obtain $O(\log m)$ approximation in linear time too. Moreover, we prove an upper bound on the expected prediction cost expressed in terms of the expected training cost. We also show that under additional assumptions the prediction cost of a PLT is $O(\log m)$.

cs.LG↗

arXiv2018

A no-regret generalization of hierarchical softmax to extreme multi-label classification

Extreme multi-label classification (XMLC) is a problem of tagging an instance with a small subset of relevant labels chosen from an extremely large pool of possible labels. Large label spaces can be efficiently handled by organizing labels as a tree, like in the hierarchical softmax (HSM) approach commonly used for multi-class problems. In this paper, we investigate probabilistic label trees (PLTs) that have been recently devised for tackling XMLC problems. We show that PLTs are a no-regret multi-label generalization of HSM when precision@k is used as a model evaluation metric. Critically, we prove that pick-one-label heuristic - a reduction technique from multi-label to multi-class that is routinely used along with HSM - is not consistent in general. We also show that our implementation of PLTs, referred to as extremeText (XT), obtains significantly better results than HSM with the pick-one-label heuristic and XML-CNN, a deep network specifically designed for XMLC problems. Moreover, XT is competitive to many state-of-the-art approaches in terms of statistical performance, model size and prediction time which makes it amenable to deploy in an online system.

cs.LG↗