arXiv subjects

Pawel Polak

Publications and source records attributed to Pawel Polak.

15 recordsLinked to original sources

arXiv2026

PG-MAP: Joint MAP Optimization for Inference-Time Alignment of Diffusion and Flow-Matching Models

Inference-time alignment of pretrained text-to-image models is typically performed along a single control axis, such as classifier-free guidance, attention editing, or reward-based latent perturbations. This limitation prevents modeling joint dependencies between conditioning and latent variables and hinders transfer across generative transports. We propose PG-MAP, a training-free framework that formulates inference-time alignment as a trajectory-level Gibbs-MAP / proximal energy optimization over the conditioning $c$ and latent state $z_t$ via a forward-consistency coupling, optionally guided by a frozen preference reward. This joint formulation enables coordinated updates across modalities while remaining compatible with both diffusion and flow-matching models through transport-specific adaptations. Across diffusion backbones (SD~1.5, SDXL), PG-MAP consistently improves alignment metrics such as PickScore and Aesthetic, and can be effectively combined with tuned classifier-free guidance to achieve the strongest overall performance. On flow-matching models (SD3.5-medium), the framework reduces to a latent-only variant, achieving $\mathbf{91.9\%}$ PickScore and $75.7\%$ HPS win rates against a static baseline, with controlled experiments ruling out noise-related artifacts. Human evaluations further confirm consistent preference over strong baselines, including tuned CFG and compute-matched universal guidance. Finally, an oracle-routing analysis shows that the relative importance of conditioning and latent optimization depends on prompt types, surfacing further headroom that a per-prompt selector could exploit.

cs.LG↗

arXiv2026

When Quotes Crumble: Detecting Transient Mechanical Liquidity Erosion in Limit Order Books

We study the detection of transient liquidity erosion ("crumbling quotes") in electronic limit order books, where observable quote deterioration may reflect either mechanical liquidity withdrawal or informational repricing. Using the ABIDES agent-based simulator, we construct a multi-agent environment in which crumbling emerges from stochastic regime switches in a market maker, providing time-resolved ground truth unavailable in real market data. We develop a detection pipeline that identifies mechanically driven quote erosion using order book features, and train a neural model to produce calibrated crumbling probabilities. Experiments demonstrate that the proposed framework reliably identifies crumbling events against agent-level ground truth, with the neural model achieving +36% AUC improvement over rule-based baselines and robust performance across normal, high-volatility, bull, and bear market conditions. Ablation studies on temporal features and varying the dependence structure of the ground-truth mechanism confirm that the framework generalizes across both independent and autocorrelated liquidity withdrawal dynamics.

cs.LG↗

arXiv2025

Scaling Conditional Autoencoders for Portfolio Optimization via Uncertainty-Aware Factor Selection

Conditional Autoencoders (CAEs) offer a flexible, interpretable approach for estimating latent asset-pricing factors from firm characteristics. However, existing studies usually limit the latent factor dimension to around K=5 due to concerns that larger K can degrade performance. To overcome this challenge, we propose a scalable framework that couples a high-dimensional CAE with an uncertainty-aware factor selection procedure. We employ three models for quantile prediction: zero-shot Chronos, a pretrained time-series foundation model (ZS-Chronos), gradient-boosted quantile regression trees using XGBoost and RAPIDS (Q-Boost), and an I.I.D bootstrap-based sample mean model (IID-BS). For each model, we rank factors by forecast uncertainty and retain the top-k most predictable factors for portfolio construction, where k denotes the selected subset of factors. This pruning strategy delivers substantial gains in risk-adjusted performance across all forecasting models. Furthermore, due to each model's uncorrelated predictions, a performance-weighted ensemble consistently outperforms individual models with higher Sharpe, Sortino, and Omega ratios.

q-fin.PM↗

arXiv2025

Online Ensemble Learning for Sector Rotation: A Gradient-Free Framework

We propose a gradient-free online ensemble learning algorithm that dynamically combines forecasts from a heterogeneous set of machine learning models based on their recent predictive performance, measured by out-of-sample R-squared. The ensemble is model-agnostic, requires no gradient access, and is designed for sequential forecasting under nonstationarity. It adaptively reweights 16 constituent models-three linear benchmarks (OLS, PCR, LASSO) and thirteen nonlinear learners including Random Forests, Gradient-Boosted Trees, and a hierarchy of neural networks (NN1-NN12). We apply the framework to sector rotation, using sector-level features aggregated from firm characteristics. Empirically, sector returns are more predictable and stable than individual asset returns, making them suitable for cross-sectional forecasting. The algorithm constructs sector-specific ensembles that assign adaptive weights in a rolling-window fashion, guided by forecast accuracy. Our key theoretical result bounds the online forecast regret directly in terms of realized out-of-sample R-squared, providing an interpretable guarantee that the ensemble performs nearly as well as the best model in hindsight. Empirically, the ensemble consistently outperforms individual models, equal-weighted averages, and traditional offline ensembles, delivering higher predictive accuracy, stronger risk-adjusted returns, and robustness across macroeconomic regimes, including during the COVID-19 crisis.

q-fin.ST↗

arXiv2025

Mix- and MoE-DPO: A Variational Inference Approach to Direct Preference Optimization

Direct Preference Optimization (DPO) has recently emerged as a simple and effective alternative to reinforcement learning from human feedback (RLHF) for aligning large language models (LLMs) with user preferences. However, existing DPO formulations rely on a single monolithic model, which limits their expressivity in multi-task settings and their adaptability to heterogeneous or diverse preference distributions. In this work, we propose Mix- and MoE-DPO, a framework that extends DPO with both soft mixture models and mixture-of-experts (MoE) architectures, using a stochastic variational inference approach. Our method introduces a latent-variable model over expert assignments and optimizes a variational evidence lower bound (ELBO), enabling stable and efficient learning of specialized expert policies from preference data. Mix- and MoE-DPO provides three key advantages over standard DPO: (i) generalization via universal function approximation through mixtures; (ii) reward and policy specialization through expert components tailored to distinct preference modes; and (iii) contextual alignment through input-dependent soft gating that enables user-specific mixture policies. Our framework supports both shared base architectures with expert-specific policy heads and fully independent expert models, allowing flexible trade-offs between parameter efficiency and specialization. We validate our approach on a variety of model sizes and multi-preference datasets, demonstrating that Mix- and MoE-DPO offers a powerful and scalable method for preference-based LLM alignment.

cs.LG↗

arXiv2025

Every Image Listens, Every Image Dances: Music-Driven Image Animation

Image animation has become a promising area in multimodal research, with a focus on generating videos from reference images. While prior work has largely emphasized generic video generation guided by text, music-driven dance video generation remains underexplored. In this paper, we introduce MuseDance, an innovative end-to-end model that animates reference images using both music and text inputs. This dual input enables MuseDance to generate personalized videos that follow text descriptions and synchronize character movements with the music. Unlike existing approaches, MuseDance eliminates the need for complex motion guidance inputs, such as pose or depth sequences, making flexible and creative video generation accessible to users of all expertise levels. To advance research in this field, we present a new multimodal dataset comprising 2,904 dance videos with corresponding background music and text descriptions. Our approach leverages diffusion-based methods to achieve robust generalization, precise control, and temporal consistency, setting a new baseline for the music-driven image animation task.

cs.CV↗

arXiv2024

Smoothing the Conditional Value-at-Risk based Pickands Estimators

We incorporate the conditional value-at-risk (CVaR) quantity into a generalized class of Pickands estimators. By introducing CVaR, the newly developed estimators not only retain the desirable properties of consistency, location, and scale invariance inherent to Pickands estimators, but also achieve a reduction in mean squared error (MSE). To address the issue of sensitivity to the choice of the number of top order statistics used for the estimation, and ensure robust estimation, which are crucial in practice, we first propose a beta measure, which is a modified beta density function, to smooth the estimator. Then, we develop an algorithm to approximate the asymptotic mean squared error (AMSE) and determine the optimal beta measure that minimizes AMSE. A simulation study involving a wide range of distributions shows that our estimators have good and highly stable finite-sample performance and compare favorably with the other estimators.

math.ST↗

arXiv2024

Online Nonconvex Bilevel Optimization with Bregman Divergences

Bilevel optimization methods are increasingly relevant within machine learning, especially for tasks such as hyperparameter optimization and meta-learning. Compared to the offline setting, online bilevel optimization (OBO) offers a more dynamic framework by accommodating time-varying functions and sequentially arriving data. This study addresses the online nonconvex-strongly convex bilevel optimization problem. In deterministic settings, we introduce a novel online Bregman bilevel optimizer (OBBO) that utilizes adaptive Bregman divergences. We demonstrate that OBBO enhances the known sublinear rates for bilevel local regret through a novel hypergradient error decomposition that adapts to the underlying geometry of the problem. In stochastic contexts, we introduce the first stochastic online bilevel optimizer (SOBBO), which employs a window averaging method for updating outer-level variables using a weighted average of recent stochastic approximations of hypergradients. This approach not only achieves sublinear rates of bilevel local regret but also serves as an effective variance reduction strategy, obviating the need for additional stochastic gradient samples at each timestep. Experiments on online hyperparameter optimization and online meta-learning highlight the superior performance, efficiency, and adaptability of our Bregman-based algorithms compared to established online and offline bilevel benchmarks.

math.OC↗

arXiv2024

Asymptotic Normality of the Conditional Value-at-Risk based Pickands Estimator

The Pickands estimator for the extreme value index is beneficial due to its universal consistency, location, and scale invariance, which sets it apart from other types of estimators. However, similar to many extreme value index estimators, it is marked by poor asymptotic efficiency. Chen (2021) introduces a Conditional Value-at-Risk (CVaR)-based Pickands estimator, establishes its consistency, and demonstrates through simulations that this estimator significantly reduces mean squared error while preserving its location and scale invariance. The initial focus of this paper is on demonstrating the weak convergence of the empirical CVaR in functional space. Subsequently, based on the established weak convergence, the paper presents the asymptotic normality of the CVaR-based Pickands estimator. It further supports these theoretical findings with empirical evidence obtained through simulation studies.

math.ST↗

arXiv2024

High-Frequency Volatility Estimation with Fast Multiple Change Points Detection

We propose a method for constructing sparse high-frequency volatility estimators that are robust against change points in the spot volatility process. The estimators we propose are $\ell_1$-regularized versions of existing volatility estimators. We focus on power variation estimators as they represent a fundamental class of volatility estimators. We establish consistency of these estimators for the true unobserved volatility and the change points locations, showing that minimax rates can be achieved for particular volatility estimators. The new estimators utilize the computationally efficient least angle regression algorithm for estimation purposes, followed by a reduced dynamic programming step to refine the final number of change points. In terms of numerical performance, these estimators are not only computationally fast but also accurately identify breakpoints near the end of the sample, both features highly desirable in today's electronic trading environment. In terms of out-of-sample volatility prediction, our new estimators provide more realistic and smoother volatility forecasts, outperforming a broad range of classical and recent volatility estimators across various frequencies and forecasting horizons.

q-fin.ST↗

arXiv2024

CP-PINNs: Data-Driven Changepoints Detection in PDEs Using Online Optimized Physics-Informed Neural Networks

We investigate the inverse problem for Partial Differential Equations (PDEs) in scenarios where the parameters of the given PDE dynamics may exhibit changepoints at random time. We employ Physics-Informed Neural Networks (PINNs) - universal approximators capable of estimating the solution of any physical law described by a system of PDEs, which serves as a regularization during neural network training, restricting the space of admissible solutions and enhancing function approximation accuracy. We demonstrate that when the system exhibits sudden changes in the PDE dynamics, this regularization is either insufficient to accurately estimate the true dynamics, or it may result in model miscalibration and failure. Consequently, we propose a PINNs extension using a Total-Variation penalty, which allows to accommodate multiple changepoints in the PDE dynamics and significantly improves function approximation. These changepoints can occur at random locations over time and are estimated concurrently with the solutions. Additionally, we introduce an online learning method for re-weighting loss function terms dynamically. Through empirical analysis using examples of various equations with parameter changes, we showcase the advantages of our proposed model. In the absence of changepoints, the model reverts to the original PINNs model. However, when changepoints are present, our approach yields superior parameter estimation, improved model fitting, and reduced training error compared to the original PINNs model.

stat.ML↗

arXiv2024

MuseChat: A Conversational Music Recommendation System for Videos

Music recommendation for videos attracts growing interest in multi-modal research. However, existing systems focus primarily on content compatibility, often ignoring the users' preferences. Their inability to interact with users for further refinements or to provide explanations leads to a less satisfying experience. We address these issues with MuseChat, a first-of-its-kind dialogue-based recommendation system that personalizes music suggestions for videos. Our system consists of two key functionalities with associated modules: recommendation and reasoning. The recommendation module takes a video along with optional information including previous suggested music and user's preference as inputs and retrieves an appropriate music matching the context. The reasoning module, equipped with the power of Large Language Model (Vicuna-7B) and extended to multi-modal inputs, is able to provide reasonable explanation for the recommended music. To evaluate the effectiveness of MuseChat, we build a large-scale dataset, conversational music recommendation for videos, that simulates a two-turn interaction between a user and a recommender based on accurate music track information. Experiment results show that MuseChat achieves significant improvements over existing video-based music retrieval methods as well as offers strong interpretability and interactability.

cs.LG↗

arXiv2024

Face-GPS: A Comprehensive Technique for Quantifying Facial Muscle Dynamics in Videos

We introduce a novel method that combines differential geometry, kernels smoothing, and spectral analysis to quantify facial muscle activity from widely accessible video recordings, such as those captured on personal smartphones. Our approach emphasizes practicality and accessibility. It has significant potential for applications in national security and plastic surgery. Additionally, it offers remote diagnosis and monitoring for medical conditions such as stroke, Bell's palsy, and acoustic neuroma. Moreover, it is adept at detecting and classifying emotions, from the overt to the subtle. The proposed face muscle analysis technique is an explainable alternative to deep learning methods and a non-invasive substitute to facial electromyography (fEMG).

cs.CV↗

arXiv2023

A Unified Framework for Fast Large-Scale Portfolio Optimization

We introduce a unified framework for rapid, large-scale portfolio optimization that incorporates both shrinkage and regularization techniques. This framework addresses multiple objectives, including minimum variance, mean-variance, and the maximum Sharpe ratio, and also adapts to various portfolio weight constraints. For each optimization scenario, we detail the translation into the corresponding quadratic programming (QP) problem and then integrate these solutions into a new open-source Python library. Using 50 years of return data from US mid to large-sized companies, and 33 distinct firm-specific characteristics, we utilize our framework to assess the out-of-sample monthly rebalanced portfolio performance of widely-adopted covariance matrix estimators and factor models, examining both daily and monthly returns. These estimators include the sample covariance matrix, linear and nonlinear shrinkage estimators, and factor portfolios based on Asset Pricing (AP) Trees, Principal Component Analysis (PCA), Risk Premium PCA (RP-PCA), and Instrumented PCA (IPCA). Our findings emphasize that AP-Trees and PCA-based factor models consistently outperform all other approaches in out-of-sample portfolio performance. Finally, we develop new l1 and l2 regularizations of factor portfolio norms which not only elevate the portfolio performance of AP-Trees and PCA-based factor models but they have a potential to reduce an excessive turnover and transaction costs often associated with these models.

q-fin.PM↗

arXiv2022

Detection of (Hidden) Emotions from Videos using Muscles Movements and Face Manifold Embedding

We provide a new non-invasive, easy-to-scale for large amounts of subjects and a remotely accessible method for (hidden) emotion detection from videos of human faces. Our approach combines face manifold detection for accurate location of the face in the video with local face manifold embedding to create a common domain for the measurements of muscle micro-movements that is invariant to the movement of the subject in the video. In the next step, we employ the Digital Image Speckle Correlation (DISC) and the optical flow algorithm to compute the pattern of micro-movements in the face. The corresponding vector field is mapped back to the original space and superimposed on the original frames of the videos. Hence, the resulting videos include additional information about the direction of the movement of the muscles in the face. We take the publicly available CK++ dataset of visible emotions and add to it videos of the same format but with hidden emotions. We process all the videos using our micro-movement detection and use the results to train a state-of-the-art network for emotions classification from videos -- Frame Attention Network (FAN). Although the original FAN model achieves very high out-of-sample performance on the original CK++ videos, it does not perform so well on hidden emotions videos. The performance improves significantly when the model is trained and tested on videos with the vector fields of muscle movements. Intuitively, the corresponding arrows serve as edges in the image that are easily captured by the convolutions filters in the FAN network.

cs.CV↗