arXiv subjects

Ivan Oseledets

Publications and source records attributed to Ivan Oseledets.

At least 91 records · Page 5Linked to original sources

arXiv2025

Faster Language Models with Better Multi-Token Prediction Using Tensor Decomposition

We propose a new model for multi-token prediction in transformers, aiming to enhance sampling efficiency without compromising accuracy. Motivated by recent work that predicts the probabilities of subsequent tokens using multiple heads, we connect this approach to rank-$1$ canonical tensor decomposition. By generalizing it to a rank-$r$ canonical probability decomposition, we develop an improved model that predicts multiple tokens simultaneously. This model can also be interpreted as a mixture of experts, allowing us to leverage successful techniques from that domain for efficient and robust training. Importantly, the overall overhead for training and sampling remains low. Our method demonstrates significant improvements in inference speed for both text and code generation tasks, proving particularly beneficial within the self-speculative decoding paradigm. It maintains its effectiveness across various model sizes and training epochs, highlighting its robustness and scalability.

cs.LG↗

arXiv2025

Learning from Linear Algebra: A Graph Neural Network Approach to Preconditioner Design for Conjugate Gradient Solvers

Large linear systems are ubiquitous in modern computational science and engineering. The main recipe for solving them is the use of Krylov subspace iterative methods with well-designed preconditioners. Recently, GNNs have been shown to be a promising tool for designing preconditioners to reduce the overall computational cost of iterative methods by constructing them more efficiently than with classical linear algebra techniques. Preconditioners designed with these approaches cannot outperform those designed with classical methods in terms of the number of iterations in CG. In our work, we recall well-established preconditioners from linear algebra and use them as a starting point for training the GNN to obtain preconditioners that reduce the condition number of the system more significantly than classical preconditioners. Numerical experiments show that our approach outperforms both classical and neural network-based methods for an important class of parametric partial differential equations. We also provide a heuristic justification for the loss function used and show that preconditioners obtained by learning with this loss function reduce the condition number in a more desirable way for CG.

cs.LG↗

arXiv2025

ConDiff: A Challenging Dataset for Neural Solvers of Partial Differential Equations

We present ConDiff, a novel dataset for scientific machine learning. ConDiff focuses on the parametric diffusion equation with space dependent coefficients, a fundamental problem in many applications of partial differential equations (PDEs). The main novelty of the proposed dataset is that we consider discontinuous coefficients with high contrast. These coefficient functions are sampled from a selected set of distributions. This class of problems is not only of great academic interest, but is also the basis for describing various environmental and industrial problems. In this way, ConDiff shortens the gap with real-world problems while remaining fully synthetic and easy to use. ConDiff consists of a diverse set of diffusion equations with coefficients covering a wide range of contrast levels and heterogeneity with a measurable complexity metric for clearer comparison between different coefficient functions. We baseline ConDiff on standard deep learning models in the field of scientific machine learning. By providing a large number of problem instances, each with its own coefficient function and right-hand side, we hope to encourage the development of novel physics-based deep learning approaches, such as neural operators, ultimately driving progress towards more accurate and efficient solutions of complex PDE problems.

cs.LG↗

arXiv2024

Certification of Speaker Recognition Models to Additive Perturbations

Speaker recognition technology is applied to various tasks, from personal virtual assistants to secure access systems. However, the robustness of these systems against adversarial attacks, particularly to additive perturbations, remains a significant challenge. In this paper, we pioneer applying robustness certification techniques to speaker recognition, initially developed for the image domain. Our work covers this gap by transferring and improving randomized smoothing certification techniques against norm-bounded additive perturbations for classification and few-shot learning tasks to speaker recognition. We demonstrate the effectiveness of these methods on VoxCeleb 1 and 2 datasets for several models. We expect this work to improve the robustness of voice biometrics and accelerate the research of certification methods in the audio domain.

cs.SD↗

arXiv2024

Scalable Cross-Entropy Loss for Sequential Recommendations with Large Item Catalogs

Scalability issue plays a crucial role in productionizing modern recommender systems. Even lightweight architectures may suffer from high computational overload due to intermediate calculations, limiting their practicality in real-world applications. Specifically, applying full Cross-Entropy (CE) loss often yields state-of-the-art performance in terms of recommendations quality. Still, it suffers from excessive GPU memory utilization when dealing with large item catalogs. This paper introduces a novel Scalable Cross-Entropy (SCE) loss function in the sequential learning setup. It approximates the CE loss for datasets with large-size catalogs, enhancing both time efficiency and memory usage without compromising recommendations quality. Unlike traditional negative sampling methods, our approach utilizes a selective GPU-efficient computation strategy, focusing on the most informative elements of the catalog, particularly those most likely to be false positives. This is achieved by approximating the softmax distribution over a subset of the model outputs through the maximum inner product search. Experimental results on multiple datasets demonstrate the effectiveness of SCE in reducing peak memory usage by a factor of up to 100 compared to the alternatives, retaining or even exceeding their metrics values. The proposed approach also opens new perspectives for large-scale developments in different domains, such as large language models.

cs.IR↗

arXiv2024

SparseGrad: A Selective Method for Efficient Fine-tuning of MLP Layers

The performance of Transformer models has been enhanced by increasing the number of parameters and the length of the processed text. Consequently, fine-tuning the entire model becomes a memory-intensive process. High-performance methods for parameter-efficient fine-tuning (PEFT) typically work with Attention blocks and often overlook MLP blocks, which contain about half of the model parameters. We propose a new selective PEFT method, namely SparseGrad, that performs well on MLP blocks. We transfer layer gradients to a space where only about 1\% of the layer's elements remain significant. By converting gradients into a sparse structure, we reduce the number of updated parameters. We apply SparseGrad to fine-tune BERT and RoBERTa for the NLU task and LLaMa-2 for the Question-Answering task. In these experiments, with identical memory requirements, our method outperforms LoRA and MeProp, robust popular state-of-the-art PEFT approaches.

cs.CL↗

arXiv2024

Spectral Informed Neural Network: An Efficient and Low-Memory PINN

With growing investigations into solving partial differential equations by physics-informed neural networks (PINNs), more accurate and efficient PINNs are required to meet the practical demands of scientific computing. One bottleneck of current PINNs is computing the high-order derivatives via automatic differentiation which often necessitates substantial computing resources. In this paper, we focus on removing the automatic differentiation of the spatial derivatives and propose a spectral-based neural network that substitutes the differential operator with a multiplication. Compared to the PINNs, our approach requires lower memory and shorter training time. Thanks to the exponential convergence of the spectral basis, our approach is more accurate. Moreover, to handle the different situations between physics domain and spectral domain, we provide two strategies to train networks by their spectral information. Through a series of comprehensive experiments, We validate the aforementioned merits of our proposed network.

cs.LG↗

arXiv2024

Inverted Activations: Reducing Memory Footprint in Neural Network Training

The scaling of neural networks with increasing data and model sizes necessitates the development of more efficient deep learning algorithms. A significant challenge in neural network training is the memory footprint associated with activation tensors, particularly in pointwise nonlinearity layers that traditionally save the entire input tensor for the backward pass, leading to substantial memory consumption. In this paper, we propose a modification to the handling of activation tensors in pointwise nonlinearity layers. Our method involves saving the output tensor instead of the input tensor during the forward pass. Since the subsequent layer typically also saves its input tensor, this approach reduces the total memory required by storing only one tensor between layers instead of two. This optimization is especially beneficial for transformer-based architectures like GPT, BERT, Mistral, and Llama. To enable this approach, we utilize the inverse function of the nonlinearity during the backward pass. As the inverse cannot be computed analytically for most nonlinearities, we construct accurate approximations using simpler functions. Experimental results demonstrate that our method significantly reduces memory usage without affecting training accuracy or computational performance. Our implementation is provided as a drop-in replacement for standard nonlinearity layers in the PyTorch framework, facilitating easy adoption without requiring architectural modifications.

cs.LG↗

arXiv2024

Tensor-Train Point Cloud Compression and Efficient Approximate Nearest-Neighbor Search

Nearest-neighbor search in large vector databases is crucial for various machine learning applications. This paper introduces a novel method using tensor-train (TT) low-rank tensor decomposition to efficiently represent point clouds and enable fast approximate nearest-neighbor searches. We propose a probabilistic interpretation and utilize density estimation losses like Sliced Wasserstein to train TT decompositions, resulting in robust point cloud compression. We reveal an inherent hierarchical structure within TT point clouds, facilitating efficient approximate nearest-neighbor searches. In our paper, we provide detailed insights into the methodology and conduct comprehensive comparisons with existing methods. We demonstrate its effectiveness in various scenarios, including out-of-distribution (OOD) detection problems and approximate nearest-neighbor (ANN) search tasks.

cs.CV↗

arXiv2024

Probabilistically Robust Watermarking of Neural Networks

As deep learning (DL) models are widely and effectively used in Machine Learning as a Service (MLaaS) platforms, there is a rapidly growing interest in DL watermarking techniques that can be used to confirm the ownership of a particular model. Unfortunately, these methods usually produce watermarks susceptible to model stealing attacks. In our research, we introduce a novel trigger set-based watermarking approach that demonstrates resilience against functionality stealing attacks, particularly those involving extraction and distillation. Our approach does not require additional model training and can be applied to any model architecture. The key idea of our method is to compute the trigger set, which is transferable between the source model and the set of proxy models with a high probability. In our experimental study, we show that if the probability of the set being transferable is reasonably high, it can be effectively used for ownership verification of the stolen model. We evaluate our method on multiple benchmarks and show that our approach outperforms current state-of-the-art watermarking techniques in all considered experimental setups.

cs.CR↗

arXiv2024

Anatomical Positional Embeddings

We propose a self-supervised model producing 3D anatomical positional embeddings (APE) of individual medical image voxels. APE encodes voxels' anatomical closeness, i.e., voxels of the same organ or nearby organs always have closer positional embeddings than the voxels of more distant body parts. In contrast to the existing models of anatomical positional embeddings, our method is able to efficiently produce a map of voxel-wise embeddings for a whole volumetric input image, which makes it an optimal choice for different downstream applications. We train our APE model on 8400 publicly available CT images of abdomen and chest regions. We demonstrate its superior performance compared with the existing models on anatomical landmark retrieval and weakly-supervised few-shot localization of 13 abdominal organs. As a practical application, we show how to cheaply train APE to crop raw CT images to different anatomical regions of interest with 0.99 recall, while reducing the image volume by 10-100 times. The code and the pre-trained APE model are available at https://github.com/mishgon/ape .

cs.CV↗

arXiv2024

RECE: Reduced Cross-Entropy Loss for Large-Catalogue Sequential Recommenders

Scalability is a major challenge in modern recommender systems. In sequential recommendations, full Cross-Entropy (CE) loss achieves state-of-the-art recommendation quality but consumes excessive GPU memory with large item catalogs, limiting its practicality. Using a GPU-efficient locality-sensitive hashing-like algorithm for approximating large tensor of logits, this paper introduces a novel RECE (REduced Cross-Entropy) loss. RECE significantly reduces memory consumption while allowing one to enjoy the state-of-the-art performance of full CE loss. Experimental results on various datasets show that RECE cuts training peak memory usage by up to 12 times compared to existing methods while retaining or exceeding performance metrics of CE loss. The approach also opens up new possibilities for large-scale applications in other domains.

cs.IR↗

arXiv2024

General Lipschitz: Certified Robustness Against Resolvable Semantic Transformations via Transformation-Dependent Randomized Smoothing

Randomized smoothing is the state-of-the-art approach to construct image classifiers that are provably robust against additive adversarial perturbations of bounded magnitude. However, it is more complicated to construct reasonable certificates against semantic transformation (e.g., image blurring, translation, gamma correction) and their compositions. In this work, we propose \emph{General Lipschitz (GL),} a new framework to certify neural networks against composable resolvable semantic perturbations. Within the framework, we analyze transformation-dependent Lipschitz-continuity of smoothed classifiers w.r.t. transformation parameters and derive corresponding robustness certificates. Our method performs comparably to state-of-the-art approaches on the ImageNet dataset.

cs.CV↗

arXiv2024

Explicit Flow Matching: On The Theory of Flow Matching Algorithms with Applications

This paper proposes a novel method, Explicit Flow Matching (ExFM), for training and analyzing flow-based generative models. ExFM leverages a theoretically grounded loss function, ExFM loss (a tractable form of Flow Matching (FM) loss), to demonstrably reduce variance during training, leading to faster convergence and more stable learning. Based on theoretical analysis of these formulas, we derived exact expressions for the vector field (and score in stochastic cases) for model examples (in particular, for separating multiple exponents), and in some simple cases, exact solutions for trajectories. In addition, we also investigated simple cases of diffusion generative models by adding a stochastic term and obtained an explicit form of the expression for score. While the paper emphasizes the theoretical underpinnings of ExFM, it also showcases its effectiveness through numerical experiments on various datasets, including high-dimensional ones. Compared to traditional FM methods, ExFM achieves superior performance in terms of both learning speed and final outcomes.

cs.LG↗

arXiv2024

Fast gradient-free activation maximization for neurons in spiking neural networks

Elements of neural networks, both biological and artificial, can be described by their selectivity for specific cognitive features. Understanding these features is important for understanding the inner workings of neural networks. For a living system, such as a neuron, whose response to a stimulus is unknown and not differentiable, the only way to reveal these features is through a feedback loop that exposes it to a large set of different stimuli. The properties of these stimuli should be varied iteratively in order to maximize the neuronal response. To utilize this feedback loop for a biological neural network, it is important to run it quickly and efficiently in order to reach the stimuli that maximizes certain neurons' activation with the least number of iterations possible. Here we present a framework with an efficient design for such a loop. We successfully tested it on an artificial spiking neural network (SNN), which is a model that simulates the asynchronous spiking activity of neurons in living brains. Our optimization method for activation maximization is based on the low-rank Tensor Train decomposition of the discrete activation function. The optimization space is the latent parameter space of images generated by SN-GAN or VQ-VAE generative models. To our knowledge, this is the first time that effective AM has been applied to SNNs. We track changes in the optimal stimuli for artificial neurons during training and show that highly selective neurons can form already in the early epochs of training and in the early layers of a convolutional spiking network. This formation of refined optimal stimuli is associated with an increase in classification accuracy. Some neurons, especially in the deeper layers, may gradually change the concepts they are selective for during learning, potentially explaining their importance for model performance.

cs.NE↗

arXiv2024

Run LoRA Run: Faster and Lighter LoRA Implementations

LoRA is a technique that reduces the number of trainable parameters in a neural network by introducing low-rank adapters to linear layers. This technique is used both for fine-tuning and full training of large language models. This paper presents the RunLoRA framework for efficient implementations of LoRA that significantly improves the speed of neural network training and fine-tuning using low-rank adapters. The proposed implementation optimizes the computation of LoRA operations based on dimensions of corresponding linear layer, layer input dimensions and lora rank by choosing best forward and backward computation graph based on FLOPs and time estimations, resulting in faster training without sacrificing accuracy. The experimental results show up to 28\% speedup on language modeling networks.

cs.LG↗

arXiv2024

A case study of spatiotemporal forecasting techniques for weather forecasting

The majority of real-world processes are spatiotemporal, and the data generated by them exhibits both spatial and temporal evolution. Weather is one of the most essential processes in this domain, and weather forecasting has become a crucial part of our daily routine. Weather data analysis is considered the most complex and challenging task. Although numerical weather prediction models are currently state-of-the-art, they are resource-intensive and time-consuming. Numerous studies have proposed time series-based models as a viable alternative to numerical forecasts. Recent research in the area of time series analysis indicates significant advancements, particularly regarding the use of state-space-based models (white box) and, more recently, the integration of machine learning and deep neural network-based models (black box). The most famous examples of such models are RNNs and transformers. These models have demonstrated remarkable results in the field of time-series analysis and have demonstrated effectiveness in modelling temporal correlations. It is crucial to capture both temporal and spatial correlations for a spatiotemporal process, as the values at nearby locations and time affect the values of a spatiotemporal process at a specific point. This self-contained paper explores various regional data-driven weather forecasting methods, i.e., forecasting over multiple latitude-longitude points (matrix-shaped spatial grid) to capture spatiotemporal correlations. The results showed that spatiotemporal prediction models reduced computational costs while improving accuracy. In particular, the proposed tensor train dynamic mode decomposition-based forecasting model has comparable accuracy to the state-of-the-art models without the need for training. We provide convincing numerical experiments to show that the proposed approach is practical.

cs.LG↗

arXiv2024

Data-driven optimal prediction with control

This study presents the extension of the data-driven optimal prediction approach to the dynamical system with control. The optimal prediction is used to analyze dynamical systems in which the states consist of resolved and unresolved variables. The latter variables can not be measured explicitly. They may have smaller amplitudes and affect the resolved variables that can be measured. The optimal prediction approach recovers the averaged trajectories of the resolved variables by computing conditional expectations, while the distribution of the unresolved variables is assumed to be known. We consider such dynamical systems and introduce their additional control functions. To predict the targeted trajectories numerically, we develop a data-driven method based on the dynamic mode decomposition. The proposed approach takes the $\mathit{measured}$ trajectories of the resolved variables, constructs an approximate linear operator from the Mori-Zwanzig decomposition, and reconstructs the $\mathit{averaged}$ trajectories of the same variables. It is demonstrated that the method is much faster than the Monte Carlo simulations and it provides a reliable prediction. We experimentally confirm the efficacy of the proposed method for two Hamiltonian dynamical systems.

math.DS↗