IMIvan Menshikov

Ivan Menshikov

*This is an algorithmically generated author record based on publicly available information.

Ivan Menshikov has a total of 11 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    From Oscillatory Modes To Source Geometry: Identifiability And Localization In Multichannel Time Series

    by Alexandra Bernadotte, Ivan Menshikov

    Recovering the structure of latent dynamical sources from multichannel observations is a fundamental inverse problem. While spectral methods can identify oscillatory components, determining whether their temporal signatures also contain sufficient information to reconstruct source geometry requires additional identifiability conditions. We develop a framework connecting the spectral structure of multichannel time series to the geometric localization of latent oscillatory sources. Building on multichannel singular spectrum analysis (MSSA) of the block-Hankel embedding, which counts the oscillatory modes of a recording and recovers their temporal poles and spatial loading vectors (the companion paper), we treat the loadings as source-specific spatial fingerprints that can be matched to a known propagation or attenuation model to estimate source coordinates. We characterize the geometric ambiguities arising from normalized attenuation profiles, establish identifiability in three strengths-local, generic and configuration-specific-with the sensor counts that each requires, and investigate the identifiability of sources that share the same temporal mode. For power-law and exponential attenuation, the ambiguity sets admit explicit geometric descriptions, and a rank-based criterion provides a computable sufficient condition for the global identifiability of a sensor array. We derive the first-order localization accuracy under an additive loading-noise model and show that it is the Cramér-Rao bound of that model. Numerical experiments examine source localization, shared-frequency sources, the effects of noise and of model mismatch-errors in the sensor coordinates, spatially correlated noise, propagation delays-and a comparison with blind separation. The complete algorithm is run on real EEG from the UCI database with a spherical head model, as a demonstration of the procedure rather than a validation. The results show how temporal mode separation, combined with a physically specified forward model, enables spatial reconstruction from multichannel observations. The framework connects oscillatory dimensionality, spectral subspace recovery, and geometric inverse problems, providing a mathematical basis for studying latent source structure in multichannel dynamical systems.

  • Open Access

    MULTICHANNEL SINGULAR SPECTRUM ANALYSIS: COUNTING OSCILLATORY MODES AND RECOVERING THEIR SPATIAL LOADINGS

    by Alexandra Bernadotte, Ivan Menshikov

    A multichannel recording is often a mixture of a few latent oscillatory sources, and the first inverse problem is to count them. We write multichannel singular spectrum analysis (MSSA) as a block-Hankel embedding of the recording, prove that the rank of the embedding is the number of distinct exponential atoms, two per oscillatory mode, whatever the number of channels, and recover the spatial loading of every mode from the poles of the signal subspace, exactly without noise and without any orthogonality between modes. In noise the count is governed by the weakest signal eigenvalue and the floor of the embedded noise. An oracle theorem, free of any assumption on the distribution of the noise, gives a threshold that returns the exact number of modes whenever the former exceeds four times the latter; a data-driven counterpart states what an estimated threshold has to satisfy; and the floor converges to the noise variance in an explicit regime. The Hankel dimension, the number of modes, which is the Bernadotte Oscillator Intrinsic Dimension (BOID), and the number of physical sources are kept apart. Simulations with confidence intervals, coloured and spatially correlated noise, and the geometry of the trajectory space on synthetic series and on EEG trials check the statements. Turning the loadings into source positions is the subject of the companion paper.

  • Open Access

    Multi-Agent Verification Games: Shared Sources, Crowding Out, and the Price of Independence

    by Alexandra Bernadotte, Ivan Menshikov

    When several agents confirm the same claim, their agreement is ordinarily taken as stronger evidence of its truth. But what if their independence is only procedural, while their evidence originates from the same source, and what if every agent acts rationally yet no one has an incentive to obtain genuinely independent evidence? We study these questions by connecting the epistemology of verification with the game theory of costly information acquisition in a multi-agent system. The primitive c is the conditional fraction of erroneous claims that remain observationally indistinguishable from correct claims to source-sharing verifiers; it separates the number of verifications from the number of independent evidentiary origins. We show that c imposes an irreducible floor on the false-accept/false-reject frontier of every decision rule based on such verdicts, and that the floor persists on arbitrary copying structures: cross-checking any number of registries that copy one root leaves it unchanged. Verification networks need not be trees: on a cycle of delegation agents confirm one another's records, and we count the architectures that avoid this. Acyclic orientations of a consultation graph are at most n! of 2 g ; unique-sink orientations of the h-cube, in which every sub-community has one responsible agent, number h Θ(2 h) and may still contain a loop; and when each agent goes to the origin with probability θ the architecture is loop-free with probability tending to θe 1-θ. The central economic result concerns a simultaneous procurement game in which n receivers choose between a cheap shared-source verification and a costlier source-independent one. We classify its pure equilibria and identify a nonempty region, pcL < d < npcL with d the independence premium, in which every pure-strategy Nash equilibrium purchases shared-source verification although source-independent verification minimizes aggregate expected cost; we bound the own error the result tolerates, derive a targeted subsidy that implements the efficient equilibrium, and confirm the conditions by exhaustive enumeration. The floor is illustrated on a chain of copies, the paper's epigraph, and on the verification record of a machine-found theorem. The findings separate strategic rationality from epistemic reliability: agents may individually optimize while collectively preserving an avoidable source-dependent error, so that independence is an economic and architectural resource and not only a statistical property of evidence. These are conditional theoretical results rather than empirical claims about deployed AI verifiers.

  • Open Access

    Brain Oscillator Intrinsic Dimension Marks Consciousness, Recovery Mode, and Failed Downshifting in Coma

    by Alexandra Bernadotte, Ivan Menshikov

    Brain function may depend less on maximizing complexity than on controlling how many dynamical degrees of freedom are active at a given time. Here we introduce brain oscillator intrinsic dimension (BOID), a geometrically grounded measure of the number of resolved oscillatory degrees of freedom in multichannel neural time series, linking finite-mode Hankel geometry to explicitly resolved dynamical modes. Across human and mouse sleep, propofol anesthesia, expert meditation, and post-anoxic coma, BOID revealed a common principle of regulated oscillator dimensionality. In intact brains, sleep and anesthesia were accompanied by an organized contraction of the active oscillator repertoire. Human slow-wave sleep showed the deepest dimensional downshift, together with reduced estimator dispersion, lower spectral entropy, and simplified trajectory topology; the sleep-related contraction was reproducible across individual recordings and across species. Propofol produced a similar reduction despite prominent organized alpha activity, indicating that oscillator dimensionality is distinct from spectral power alone. Expert meditation further showed that reduced dimensionality can coexist with preserved waking control, demonstrating that low BOID is not intrinsically a marker of impaired consciousness. Severe post-anoxic injury revealed the opposite regime. In 575 patients from the PhysioNet I-CARE cohort, poorer outcome was associated with increased oscillator dimensionality and pathological mode proliferation. A pathological mode-proliferation index reached an AUC of 0.77, while a multimodal model combining oscillator, intrinsic-dimensional, spectral, burst-suppression, and clinical features reached an internally cross-validated AUC of 0.81. Because post-cardiac-arrest outcomes can be confounded by EEG-informed withdrawal of life-sustaining therapy, these associations should not be interpreted as direct measures of biological recoverability. Together, these results support regulated oscillator dimensionality as a structural principle of brain-state organization: intact neural systems expand and contract their active dynamical repertoire according to state, whereas severe injury can disrupt this regulation and produce pathological mode proliferation. The critical variable may therefore be not complexity itself, but control over dynamical dimensionality. BOID is defined for any multivariate time series with finite oscillatory structure, not only for brain recordings. We outline how the same estimator applies to economic and financial time series as a regime read-out, to multivariate data analysis as a model-order criterion, and to the structural assessment of artificial intelligence, where it supplies the dynamical component of the intrinsic-dimension criterion for strong intelligence.

  • Open Access

    When Do Language Models Reach Nash? Heterogeneous Cross-Family LLM Agents in Classic Economic Games.

    by Alexandra Bernadotte, Ivan Menshikov

    Whether artificial agents reach the Nash equilibrium of an economic game has become a load-bearing question for algorithmic markets and multi-agent AI. The existing literature establishes a robust yes for one behaviour, tacit collusion above the competitive benchmark, while painting a mixed picture for equilibrium play in general. We argue the more useful question is not whether but when language models reach Nash, and what instrument detects it. Using a heterogeneous, cross-machine testbed in which agents drawn from three distinct model families (Qwen 3.6, Gemma 4, Nemotron 3) interact across two physical servers, we run three classic games with analytically known equilibria: the p-beauty contest, the iterated Prisoner’s Dilemma, and a Braess-paradox routing game. We find a consistent pattern: agents reproduce the exact Nash equilibrium only when the decision reduces to selecting the lowest of a set of given payoffs (the Braess game with explicit option costs, where the paradoxical equilibrium is recovered to the unit), whereas every task demanding genuine strategic reasoning, level-k depth in the beauty contest, backward induction in the Prisoner’s Dilemma, or computing congestion from cost formulas, is met instead with a memorized or social heuristic that diverges from equilibrium, sometimes catastrophically. A simple contamination probe, inverting the beauty-contest parameter so the equilibrium moves from 0 to 100, exposes the apparent equilibrium “success” of the standard game as template recall: agents move in the wrong direction. Crucially, heterogeneous cross-family interaction surfaces strategic asymmetries that are invisible under self-play: Gemma 4 performs textbook end-game defection and exploits a trusting Qwen in the Prisoner’s Dilemma, while Qwen out-routes Gemma and Nemotron in the congestion game. We adopt normalized exploitability (ε-Nash) as the primary metric, and we identify four under-explored measurement instruments, surrogate (non-interacting) controls, output-space convergence as an equilibrium signature, true cross-family heterogeneity, and human+AI neural co-measurement, as the open ground this testbed is built to claim. All results are pilot-scale and intended as a proof of concept and a methodological scaffold rather than a definitive benchmark.

  • A Lightweight Hankel-Embedded Pipeline for Real-Time EEG Filtering and Classification

    by Ivan Menshikov, Nikita Elfimov, Alexandra Bernadotte

    EEG-based BCIs face noise and computational constraints for on-device deployment. We present a lightweight pipeline using Hankel embeddings: projecting EEG blocks onto paired eigenvectors yields oscillators-closed-orbit components. Our approach provides three contributions: (1) jPCA-based oscillator dynamics with mechanics-inspired descriptors for stable, class-separable features; (2) adaptive band-pass filtering by reconstructing signals from selected oscillators, suppressing noise without preset frequencies; (3) consistent spatiotemporal oscillator localization across channels. The pipeline uses only matrix operations, enabling GPU-free streaming. On motor-imagery benchmarks, our Hankel-based filter improves classifier accuracy by 6%, providing a resource-efficient EEG representation for edge BCI systems.


  • Open Access

    Semi-automatic annotation of brain vessels in magnetic resonance angiography images

    by Alexandra Bernadotte, Nikita Elfimov, Ivan Menshikov

    Accurate segmentation of brain vessels in magnetic resonance angiography (MRA) is essential for surgical procedures. Neural networks are powerful tools for medical image segmentation, but their development requires well-annotated datasets. However, publicly available MRA datasets with detailed vessel annotations are scarce. We present a dataset of 100 manually annotated brain MRA images from the IXI Dataset, representing one of the largest publicly available collections with detailed vessel segmentation. We focused on the Circle of Willis and associated vessels, critical for neurovascular surgery planning. The annotation pipeline involved automated segmentation using the Frangi vesselness filter, followed by manual refinement by three annotators under supervision of three neurovascular surgeons. Images were acquired using 1.5T and 3T MRI scanners. The dataset includes demographic metadata, with clustering analysis revealing four distinct morphological patterns. This resource enables development of automated segmentation algorithms, investigation of cerebrovascular morphology variations, and advancement of AI-driven diagnostic tools.


  • Open Access

    Silent EEG-Speech Recognition Using Convolutional and Recurrent Neural Network with 85% Accuracy of 9 Words Classification

    by Darya Vorontsova, Ivan Menshikov, Aleksandr Zubov, Kirill Orlov, Peter Rikunov, Ekaterina Zvereva, Lev Flitman, Anton Lanikin, Anna Sokolova, Sergey Markov, et al.

    In this work, we focus on silent speech recognition in electroencephalography (EEG) data of healthy individuals to advance brain–computer interface (BCI) development to include people with neurodegeneration and movement and communication difficulties in society. Our dataset was recorded from 270 healthy subjects during silent speech of eight different Russia words (commands): ‘forward’, ‘backward’, ‘up’, ‘down’, ‘help’, ‘take’, ‘stop’, and ‘release’, and one pseudoword. We began by demonstrating that silent word distributions can be very close statistically and that there are words describing directed movements that share similar patterns of brain activity. However, after training one individual, we achieved 85% accuracy performing 9 words (including pseudoword) classification and 88% accuracy on binary classification on average. We show that a smaller dataset collected on one participant allows for building a more accurate classifier for a given subject than a larger dataset collected on a group of people. At the same time, we show that the learning outcomes on a limited sample of EEG-data are transferable to the general population. Thus, we demonstrate the possibility of using selected command-words to create an EEG-based input device for people on whom the neural network classifier has not been trained, which is particularly important for people with disabilities.