Reduced-rank (RR) regression may be interpreted as a dimensionality reduction technique able to reveal complex relationships among the data parsimoniously. However, RR models typically overlook any potential group structure responses by assuming low-rank on coefficient matrix. To address this limitation, Bayesian Partial (BPRR) is exploited, where response vector and matrix are partitioned into low- full-rank sub-groups. As opposed literature, which assumes known rank, novel strategy...

We tackle the challenges of modeling high-dimensional data sets, particularly those with latent low-dimensional structures hidden within complex, non-linear, and noisy relationships. Our approach enables a seamless integration concepts from non-parametric regression, factor models, neural networks for regression. introduces PCA Soft layers, which can be embedded at any stage network architecture, allowing model to alternate between non-linear transformations. This flexibility makes our...

Verbal autopsies (VA) are increasingly used in low- and middle-income countries where most causes of death (COD) occur at home without medical attention, deaths differ substantially from hospital deaths. Hence, there is no plausible "standard" against which VAs for may be validated. Previous studies have shown contradictory performance automated methods compared to physician-based classification CODs. We sought compare the classic naive Bayes classifier (NBC) versus existing classifiers,...

Abstract Not all graphs are clusterable. have a clustered structure and can be meaningfully summarized through vertex clustering. Clusterable characterized by pockets of densely connected vertices that only sparsely to the remaining graph. In this article, we re-introduce very simple intuitive, yet highly informative, statistical hypothesis test for graph clusterability is based on neighborhood samples. The goal determine if meets necessary structural conditions clusters. Our clusterable...

Abstract We introduce graph clustering quality measures based on comparisons of global, intra- and inter-cluster densities, an accompanying statistical significance test a step-by-step routine for assessment. Our work is centred the idea that well-clustered graphs will display mean intra-cluster density higher than global density. do not rely any generative model null graph. are shown to meet axioms good function. They have intuitive graph-theoretic interpretation, formal interpretation can...

In multivariate time series systems, key insights can be obtained by discovering lead-lag relationships inherent in the data, which refer to dependence between two shifted relative one another, and leveraged for purposes of control, forecasting or clustering. We develop a clustering-driven methodology robust detection lagged multi-factor models. Within our framework, envisioned pipeline takes as input set series, creates an enlarged universe extracted subsequence from each using sliding...

inference problem that has no straightforward solution. We take a Bayesian approach to the of unknown parameters non-linear state model; this, in turn, requires availability ecient Markov Chain Monte Carlo (MCMC) sampling methods for latent (hidden) variables and model parameters. Using ensemble technique Neal (2010) embedded HMM (2003), we introduce new method space models. The key idea is perform parameter updates conditional on an enormously large sequences, as opposed single sequence,...

We propose a new scheme for selecting pool states the embedded Hidden Markov Model (HMM) Chain Monte Carlo (MCMC) method. This allows HMM method to be used efficient sampling in state space models where can high-dimensional. Previously, methods were only applicable low-dimensional state-space models. demonstrate that using our proposed selection scheme, an sampler have similar performance well-tuned uses combination of Particle Gibbs with Backward Sampling (PGBS) and Metropolis updates. The...

We propose an efficient online approximate Bayesian inference algorithm for estimating the parameters of a nonlinear function from potentially non-stationary data stream. The method is based on extended Kalman filter (EKF), but uses novel low-rank plus diagonal decomposition posterior precision matrix, which gives cost per step linear in number model parameters. In contrast to methods stochastic variational inference, our fully deterministic, and does not require step-size tuning. show...

In the light of micro-scale inefficiencies due to highly fragmented bitcoin trading landscape, we use a granular data set comprising orderbook and trades from most liquid markets, understand price formation process at sub-1-second time scales. To this end, construct features that encapsulate relevant microstructural information over short lookback windows. These are subsequently leveraged, first generate leader–lagger network quantifies how markets impact one another, then train linear...

Measuring graph clustering quality remains an open problem. To address it, we introduce measures based on comparisons of intra- and inter-cluster densities, accompanying statistical test the significance their differences a step-by-step routine for assessment. Our null hypothesis does not rely any generative model graph, unlike modularity which uses configuration as model. are shown to meet axioms good function, very commonly used measure. They also have intuitive graph-theoretic...

Background: Verbal autopsies (VA) are increasingly used in low- and middle-income countries where most causes of death (COD) occur at home without medical attention, deaths differ substantially from hospital deaths. Hence, there is no plausible “standard” against which VAs for may be validated. Previous studies have shown contradictory performance automated methods compared to physician-based classification CODs. We sought compare the classic naive Bayes classifier (NBC) versus existing...

We present the findings of a large-scale live trading experiment involving placement millions market orders sent at high frequency on two cryptocurrency exchanges, Bybit and Binance. analyze execution outcomes these in comparison to expected outcome based most recent snapshot Limit Order Book (LOB) time order submission for modes: one using second marketable limit aiming best price. Discrepancies between actual are due intermittent LOB updates during span resulting from delays exchange,...

We derive a novel, provably robust, and closed-form Bayesian update rule for online filtering in state-space models the presence of outliers misspecified measurement models. Our method combines generalised inference with methods such as extended ensemble Kalman filter. use former to show robustness latter ensure computational efficiency case nonlinear matches or outperforms other robust (such those based on variational Bayes) at much lower cost. this empirically range problems outlier...

Reduced-rank (RR) regression may be interpreted as a dimensionality reduction technique able to reveal complex relationships among the data parsimoniously. However, RR models typically overlook any potential group structure responses by assuming low-rank on coefficient matrix. To address this limitation, Bayesian Partial (BPRR) is exploited, where response vector and matrix are partitioned into low- full-rank sub-groups. As opposed literature, which assumes known rank, novel strategy...

We demonstrate the use of Conditional Variational Encoder (CVAE) to improve forecasts daily stock volume time series in both short and long term forecasting tasks, with advanced information input variables such as rebalancing dates. CVAE generates non-linear out-of-sample forecasts, which have better accuracy closer fit correlation actual data, compared traditional linear models. These generative can also be used for scenario generation, aids interpretation. further discuss correlations...

We employ a Bayesian modelling technique for high dimensional cointegration estimation to construct low volatility portfolios from large number of stocks. The proposed framework effectively identifies sparse and important relationships amongst baskets stocks across various asset spaces, resulting in with reduced volatility. Such persist well over the out-of-sample testing time, providing practical benefits portfolio construction optimization. Further studies on drawdown minimization also...