arXiv subjects

Denise Lanzieri

Publications and source records attributed to Denise Lanzieri.

11 recordsLinked to original sources

arXiv2026

SOM-VQ: Topology-Aware Tokenization for Interactive Generative Models

Vector-quantized representations enable powerful discrete generative models but lack semantic structure in token space, limiting interpretable human control. We introduce SOM-VQ, a tokenization method that combines vector quantization with Self-Organizing Maps to learn discrete codebooks with explicit low-dimensional topology. Unlike standard VQ-VAE, SOM-VQ uses topology-aware updates that preserve neighborhood structure: nearby tokens on a learned grid correspond to semantically similar states, enabling direct geometric manipulation of the latent space. We demonstrate that SOM-VQ produces more learnable token sequences in the evaluated domains while providing an explicit navigable geometry in code space. Critically, the topological organization enables intuitive human-in-the-loop control: users can steer generation by manipulating distances in token space, achieving semantic alignment without frame-level constraints. We focus on human motion generation - a domain where kinematic structure, smooth temporal continuity, and interactive use cases (choreography, rehabilitation, HCI) make topology-aware control especially natural - demonstrating controlled divergence and convergence from reference sequences through simple grid-based sampling. SOM-VQ provides a general framework for interpretable discrete representations applicable to music, gesture, and other interactive generative domains.

cs.LG↗

arXiv2026

Inverting Self-Organizing Maps: A Unified Activation-Based Framework

Self-Organizing Maps (SOMs) provide topology-preserving projections of high-dimensional data, yet their use as generative models remains largely unexplored. We show that the activation pattern of a SOM -- the squared distances to its prototypes -- can be \emph{inverted} to recover the exact input, following from a classical result in Euclidean distance geometry: a point in $D$ dimensions is uniquely determined by its distances to $D{+}1$ affinely independent references. We derive the corresponding linear system and characterize the conditions under which inversion is well-posed. Building on this mechanism, we introduce the \emph{Manifold-Aware Unified SOM Inversion and Control} (MUSIC) update rule, which modifies squared distances to selected prototypes while preserving others, producing controlled, semantically meaningful trajectories aligned with the SOM's piecewise-linear structure. Tikhonov regularization stabilizes the update and ensures smooth motion in high dimensions. Unlike variational or diffusion-based generative models, MUSIC requires no sampling, latent priors, or learned decoders: it operates entirely on prototype geometry. If no perturbation is applied, inversion recovers the exact input; when a target prototype or cluster is specified, MUSIC produces coherent semantic transitions. We validate the framework on synthetic Gaussian mixtures, MNIST digits, and the Labeled Faces in the Wild dataset. Across all settings, MUSIC trajectories maintain high classifier confidence, produce significantly sharper intermediate images than linear interpolation, and reveal an interpretable geometric structure of the learned map.

cs.LG↗

arXiv2025

First-Extinction Law for Resampling Processes

Extinction times in resampling processes are fundamental yet often intractable, as previous formulas scale as $2^M$ with the number of states $M$ present in the initial probability distribution. We solve this by treating multinomial updates as independent square-root diffusions of zero drift, yielding a closed-form law for the first-extinction time. We prove that the mean coincides exactly with the Wright-Fisher result of Baxter et al., thereby replacing exponential-cost evaluations with a linear-cost expression, and we validate this result through extensive simulations. Finally, we demonstrate predictive power for model collapse in a simple self-training setup: the onset of collapse coincides with the resampling-driven first-extinction time computed from the model's initial stationary distribution. These results hint to a unified view of resampling extinction dynamics.

stat.ML↗

arXiv2025

Lyapunov Learning at the Onset of Chaos

Handling regime shifts and non-stationary time series in deep learning systems presents a significant challenge. In the case of online learning, when new information is introduced, it can disrupt previously stored data and alter the model's overall paradigm, especially with non-stationary data sources. Therefore, it is crucial for neural systems to quickly adapt to new paradigms while preserving essential past knowledge relevant to the overall problem. In this paper, we propose a novel training algorithm for neural networks called \textit{Lyapunov Learning}. This approach leverages the properties of nonlinear chaotic dynamical systems to prepare the model for potential regime shifts. Drawing inspiration from Stuart Kauffman's Adjacent Possible theory, we leverage local unexplored regions of the solution space to enable flexible adaptation. The neural network is designed to operate at the edge of chaos, where the maximum Lyapunov exponent, indicative of a system's sensitivity to small perturbations, evolves around zero over time. Our approach demonstrates effective and significant improvements in experiments involving regime shifts in non-stationary systems. In particular, we train a neural network to deal with an abrupt change in Lorenz's chaotic system parameters. The neural network equipped with Lyapunov learning significantly outperforms the regular training, increasing the loss ratio by about $96\%$.

cs.LG↗

arXiv2024

Dreaming Learning

Incorporating novelties into deep learning systems remains a challenging problem. Introducing new information to a machine learning system can interfere with previously stored data and potentially alter the global model paradigm, especially when dealing with non-stationary sources. In such cases, traditional approaches based on validation error minimization offer limited advantages. To address this, we propose a training algorithm inspired by Stuart Kauffman's notion of the Adjacent Possible. This novel training methodology explores new data spaces during the learning phase. It predisposes the neural network to smoothly accept and integrate data sequences with different statistical characteristics than expected. The maximum distance compatible with such inclusion depends on a specific parameter: the sampling temperature used in the explorative phase of the present method. This algorithm, called Dreaming Learning, anticipates potential regime shifts over time, enhancing the neural network's responsiveness to non-stationary events that alter statistical properties. To assess the advantages of this approach, we apply this methodology to unexpected statistical changes in Markov chains and non-stationary dynamics in textual sequences. We demonstrated its ability to improve the auto-correlation of generated textual sequences by $\sim 29\%$ and enhance the velocity of loss convergence by $\sim 100\%$ in the case of a paradigm shift in Markov chains.

cs.LG↗

arXiv2024

Simulation-Based Inference Benchmark for Weak Lensing Cosmology

Standard cosmological analysis, which relies on two-point statistics, fails to extract the full information of the data. This limits our ability to constrain with precision cosmological parameters. Thus, recent years have seen a paradigm shift from analytical likelihood-based to simulation-based inference. However, such methods require a large number of costly simulations. We focus on full-field inference, considered the optimal form of inference. Our objective is to benchmark several ways of conducting full-field inference to gain insight into the number of simulations required for each method. We make a distinction between explicit and implicit full-field inference. Moreover, as it is crucial for explicit full-field inference to use a differentiable forward model, we aim to discuss the advantages of having this property for the implicit approach. We use the sbi_lens package which provides a fast and differentiable log-normal forward model. This forward model enables us to compare explicit and implicit full-field inference with and without gradient. The former is achieved by sampling the forward model through the No U-Turns sampler. The latter starts by compressing the data into sufficient statistics and uses the Neural Likelihood Estimation algorithm and the one augmented with gradient. We perform a full-field analysis on LSST Y10 like weak lensing simulated mass maps. We show that explicit and implicit full-field inference yield consistent constraints. Explicit inference requires 630 000 simulations with our particular sampler corresponding to 400 independent samples. Implicit inference requires a maximum of 101 000 simulations split into 100 000 simulations to build sufficient statistics (this number is not fine tuned) and 1 000 simulations to perform inference. Additionally, we show that our way of exploiting the gradients does not significantly help implicit inference.

astro-ph.CO↗

arXiv2024

Optimal Neural Summarisation for Full-Field Weak Lensing Cosmological Implicit Inference

Traditionally, weak lensing cosmological surveys have been analyzed using summary statistics motivated by their analytically tractable likelihoods, or by their ability to access higher-order information, at the cost of requiring Simulation-Based Inference (SBI) approaches. While informative, these statistics are neither designed nor guaranteed to be statistically sufficient. With the rise of deep learning, it becomes possible to create summary statistics optimized to extract the full data information. We compare different neural summarization strategies proposed in the weak lensing literature, to assess which loss functions lead to theoretically optimal summary statistics to perform full-field inference. In doing so, we aim to provide guidelines and insights to the community to help guide future neural-based inference analyses. We design an experimental setup to isolate the impact of the loss function used to train neural networks. We have developed the sbi_lens JAX package, which implements an automatically differentiable lognormal wCDM LSST-Y10 weak lensing simulator. The explicit full-field posterior obtained using the Hamiltonian Monte Carlo sampler gives us a ground truth to which to compare different compression strategies. We provide theoretical insight into the loss functions used in the literature and show that some do not necessarily lead to sufficient statistics (e.g. Mean Square Error (MSE)), while those motivated by information theory (e.g. Variational Mutual Information Maximization (VMIM)) can. Our numerical experiments confirm these insights and show, in our simulated wCDM scenario, that the Figure of Merit (FoM) of an analysis using neural summaries optimized under VMIM achieves 100% of the reference Omega_c - sigma_8 full-field FoM, while an analysis using neural summaries trained under MSE achieves only 81% of the same reference FoM.

astro-ph.CO↗

arXiv2023

Forecasting the power of Higher Order Weak Lensing Statistics with automatically differentiable simulations

We present the Differentiable Lensing Lightcone (DLL), a fully differentiable physical model designed for being used as a forward model in Bayesian inference algorithms requiring access to derivatives of lensing observables with respect to cosmological parameters. We extend the public FlowPM N-body code, a particle-mesh N-body solver, simulating lensing lightcones and implementing the Born approximation in the Tensorflow framework. Furthermore, DLL is aimed at achieving high accuracy with low computational costs. As such, it integrates a novel Hybrid Physical-Neural parameterisation able to compensate for the small-scale approximations resulting from particle-mesh schemes for cosmological N-body simulations. We validate our simulations in an LSST setting against high-resolution $κ$TNG simulations by comparing both the lensing angular power spectrum and multiscale peak counts. We demonstrate an ability to recover lensing $C_\ell$ up to a 10% accuracy at $\ell=1000$ for sources at redshift 1, with as few as $\sim 0.6$ particles per Mpc/h. As a first use case, we use this tool to investigate the relative constraining power of the angular power spectrum and peak counts statistic in an LSST setting. Such comparisons are typically very costly as they require a large number of simulations, and do not scale well with the increasing number of cosmological parameters. As opposed to forecasts based on finite differences, these statistics can be analytically differentiated with respect to cosmology, or any systematics included in the simulations at the same computational cost of the forward simulation. We find that the peak counts outperform the power spectrum on the cold dark matter parameter $Ω_c$, on the amplitude of density fluctuations $σ_8$, and on the amplitude of the intrinsic alignment signal $A_{IA}$.

astro-ph.IM↗

arXiv2023

JAX-COSMO: An End-to-End Differentiable and GPU Accelerated Cosmology Library

We present jax-cosmo, a library for automatically differentiable cosmological theory calculations. It uses the JAX library, which has created a new coding ecosystem, especially in probabilistic programming. As well as batch acceleration, just-in-time compilation, and automatic optimization of code for different hardware modalities (CPU, GPU, TPU), JAX exposes an automatic differentiation (autodiff) mechanism. Thanks to autodiff, jax-cosmo gives access to the derivatives of cosmological likelihoods with respect to any of their parameters, and thus enables a range of powerful Bayesian inference algorithms, otherwise impractical in cosmology, such as Hamiltonian Monte Carlo and Variational Inference. In its initial release, jax-cosmo implements background evolution, linear and non-linear power spectra (using halofit or the Eisenstein and Hu transfer function), as well as angular power spectra with the Limber approximation for galaxy and weak lensing probes, all differentiable with respect to the cosmological parameters and their other inputs. We illustrate how autodiff can be a game-changer for common tasks involving Fisher matrix computations, or full posterior inference with gradient-based techniques. In particular, we show how Fisher matrices are now fast, exact, no longer require any fine tuning, and are themselves differentiable. Finally, using a Dark Energy Survey Year 1 3x2pt analysis as a benchmark, we demonstrate how jax-cosmo can be combined with Probabilistic Programming Languages to perform posterior inference with state-of-the-art algorithms including a No U-Turn Sampler, Automatic Differentiation Variational Inference,and Neural Transport HMC. We further demonstrate that Normalizing Flows using Neural Transport are a promising methodology for model validation in the early stages of analysis.

astro-ph.CO↗

arXiv2022

Hybrid Physical-Neural ODEs for Fast N-body Simulations

We present a new scheme to compensate for the small-scales approximations resulting from Particle-Mesh (PM) schemes for cosmological N-body simulations. This kind of simulations are fast and low computational cost realizations of the large scale structures, but lack resolution on small scales. To improve their accuracy, we introduce an additional effective force within the differential equations of the simulation, parameterized by a Fourier-space Neural Network acting on the PM-estimated gravitational potential. We compare the results for the matter power spectrum obtained to the ones obtained by the PGD scheme (Potential gradient descent scheme). We notice a similar improvement in term of power spectrum, but we find that our approach outperforms PGD for the cross-correlation coefficients, and is more robust to changes in simulation settings (different resolutions, different cosmologies).

astro-ph.CO↗

arXiv2022

AMICO galaxy clusters in KiDS-DR3: Measurement of the halo bias and power spectrum normalization from a stacked weak lensing analysis

Galaxy clusters are biased tracers of the underlying matter density field. At very large radii beyond about 10 Mpc/\textit{h}, the shear profile shows evidence of a second-halo term. This is related to the correlated matter distribution around galaxy clusters and proportional to the so-called halo bias. We present an observational analysis of the halo bias-mass relation based on the AMICO galaxy cluster catalog, comprising around 7000 candidates detected in the third release of the KiDS survey. We split the cluster sample into 14 redshift-richness bins and derive the halo bias and the virial mass in each bin by means of a stacked weak lensing analysis. The observed halo bias-mass relation and the theoretical predictions based on the $Λ$CDM standard cosmological model show an agreement within $2σ$. The mean measurements of bias and mass over the full catalog give $M_{200c} = (4.9 \pm 0.3) \times 10^{13} M_{\odot}/\textit{h}$ and $b_h σ_8^2 = 1.2 \pm 0.1$. With the additional prior of a bias-mass relation from numerical simulations, we constrain the normalization of the power spectrum with a fixed matter density $Ω_m = 0.3$, finding $σ_8 = 0.63 \pm 0.10$.

astro-ph.CO↗