Nicolas Chapados

Cost functions and model combination for VaR-based asset allocation using neural networks

We introduce an asset-allocation framework based on the active control of the value-at-risk of the portfolio. Within this framework, we compare two paradigms for making the allocation using neural networks. The first one uses the network to make a forecast of asset behavior, in conjunction with a traditional mean-variance allocator for constructing the portfolio. The second paradigm uses the network to directly make the portfolio allocation decisions. We consider a method for performing soft input variable selection, and show its considerable utility. We use model combination (committee) methods to systematize the choice of hyperparameters during training. We show that committees using both paradigms are significantly outperforming the benchmark market performance.

HeMIS: Hetero-Modal Image Segmentation

We introduce a deep learning image segmentation framework that is extremely robust to missing imaging modalities. Instead of attempting to impute or synthesize missing data, the proposed approach learns, for each modality, an embedding of the input image into a single latent vector space for which arithmetic operations (such as taking the mean) are well defined. Points in that space, which are averaged over modalities available at inference time, can then be further processed to yield the desired segmentation. As such, any combinatorial subset of available modalities can be provided as input, without having to learn a combinatorial number of imputation models. Evaluated on two neurological MRI datasets (brain tumors and MS lesions), the approach yields state-of-the-art segmentation results when provided with all modalities; moreover, its performance degrades remarkably gracefully when modalities are removed, significantly more so than alternative mean-filling or other synthesis approaches.

Retail store scheduling for profit

In spite of its tremendous economic significance, the problem of sales staff schedule optimization for retail stores has received relatively scant attention. Current approaches typically attempt to minimize payroll costs by closely fitting a staffing curve derived from exogenous sales forecasts, oblivious to the ability of additional staff to (sometimes) positively impact sales. In contrast, this paper frames the retail scheduling problem in terms of operating profit maximization, explicitly recognizing the dual role of sales employees as sources of revenues as well as generators of operating costs. We introduce a flexible stochastic model of retail store sales, estimated from store-specific historical data, that can account for the impact of all known sales drivers, including the number of scheduled staff, and provide an accurate sales forecast at a high intra-day resolution. We also present solution techniques based on mixed-integer (MIP) and constraint programming (CP) to efficiently solve the complex mixed integer non-linear scheduling (MINLP) problem with a profit-maximization objective. The proposed approach allows solving full weekly schedules to optimality, or near-optimality with a very small gap. On a case-study with a medium-sized retail chain, this integrated forecasting–scheduling methodology yields significant projected net profit increases on the order of 2–3% compared to baseline schedules.

Retail store workforce scheduling by expected operating income maximization

We address the problem of retail store sales personnel scheduling by casting it in terms of an expected operating income maximization. In this framework, salespeople are no longer only responsible for operating costs, but also contribute to operating revenue. We model the marginal impact of an additional staff by making use of historical sales and payroll data, conditioned on a store-, date- and time-dependent traffic forecast. The expected revenue and its uncertainty are then fed into a constraint program which builds an operational schedule maximizing the expected operating income. A case study with a medium-sized retailer suggests that revenue increases of 7% and operating income increases of 3% are possible with the approach.

Detonation Classification from Acoustic Signature with the Restricted Boltzmann Machine

We compare the recently proposed Discriminative Restricted Boltzmann Machine (DRBM) to the classical Support Vector Machine (SVM) on a challenging classification task consisting in identifying weapon classes from audio signals. The three weapon classes considered in this work (mortar, rocket, and rocket‐propelled grenade), are difficult to reliably classify with standard techniques because they tend to have similar acoustic signatures. In addition, specificities of the data available in this study make it challenging to rigorously compare classifiers, and we address methodological issues arising from this situation. Experiments show good classification accuracy that could make these techniques suitable for fielding on autonomous devices. DRBMs appear to yield better accuracy than SVMs, and are less sensitive to the choice of signal preprocessing and model hyperparameters. This last property is especially appealing in such a task where the lack of data makes model validation difficult.

A high-order feature synthesis and selection algorithm applied to insurance risk modelling

In many jurisdictions, automobile insurers have access to risk-sharing pools to which they can transfer some risks. We consider different feature selection and modelling approaches to maximise profitability of these transfers through better risk selection. For that purpose, we introduce a flexible scoring model and devise a robust feature synthesis and selection method. We show what should be the most suitable sorting criterion depending on pool regulations. We use a technique, similar to cross validation, but that is coherent with the sequential structure of insurance data. We explain how software maturity level impacts profitability.

Noisy K Best-Paths for Approximate Dynamic Programming with Application to Portfolio Optimization

We describe a general method to transform a non-Markovian sequential decision problem into a supervised learning problem using a K-best-paths algorithm. We consider an application in financial portfolio management where we can train a controller to directly optimize a Sharpe Ratio (or other risk-averse non-additive) utility function. We illustrate the approach by demonstrating experimental results using a kernel-based controller architecture that would not normally be considered in traditional reinforcement learning or approximate dynamic programming. We further show that using a non-additive criterion (incremental Sharpe Ratio) yields a noisy K-best-paths extraction problem, that can give substantially improved performance.

CASED: Curriculum Adaptive Sampling for Extreme Data Imbalance

We introduce CASED, a novel curriculum sampling algorithm that facilitates the optimization of deep learning segmentation or detection models on data sets with extreme class imbalance. We evaluate the CASED learning framework on the task of lung nodule detection in chest CT. In contrast to two-stage solutions, wherein nodule candidates are first proposed by a segmentation model and refined by a second detection stage, CASED improves the training of deep nodule segmentation models (e.g. UNet) to the point where state of the art results are achieved using only a trivial detection stage. CASED improves the optimization of deep segmentation models by allowing them to first learn how to distinguish nodules from their immediate surroundings, while continuously adding a greater proportion of difficult-to-classify global context, until uniformly sampling from the empirical data distribution. Using CASED during training yields a minimalist proposal to the lung nodule detection problem that tops the LUNA16 nodule detection benchmark with an average sensitivity score of 88.35%. Furthermore, we find that models trained using CASED are robust to nodule annotation quality by showing that comparable results can be achieved when only a point and radius for each ground truth nodule are provided during training. Finally, the CASED learning framework makes no assumptions with regard to imaging modality or segmentation target and should generalize to other medical imaging problems where class imbalance is a persistent problem.

Volatility Forecasting and Explanatory Variables: A Tractable Bayesian Approach to Stochastic Volatility

We provide a formulation of stochastic volatility (SV) based on Gaussian process regression (GPR). Forecasting volatility out-of-sample, both simulation and empirical analyses show that our GPR-based stochastic volatility (GPSV) model clearly outperforms SV and GARCH benchmarks, especially at long horizons. Most importantly, our approach enables the straightforward incorporation of arbitrary covariates without requiring the specification of functional forms a priori. Augmenting the GPSV model with exogenous variables increases its performance even further. In particular, a simple set of covariates reduces the error rate on one-year out-of-sample forecasting during the 2007-09 recession by 26% relative to a benchmark range-based SV model.

DETONATION CLASSIFICATION FROM ACOUSTIC SIGNATURE WITH THE RESTRICTED BOLTZMANN MACHINE: Detonation Classification from Acoustic Signature with the Restricted Boltzmann Machine

We compare the recently proposed Discriminative Restricted Boltzmann Machine (DRBM) to the classical Support Vector Machine (SVM) on a challenging classification task consisting in identifying weapon classes from audio signals. The three weapon classes considered in this work (mortar, rocket, and rocket‐propelled grenade), are difficult to reliably classify with standard techniques because they tend to have similar acoustic signatures. In addition, specificities of the data available in this study make it challenging to rigorously compare classifiers, and we address methodological issues arising from this situation. Experiments show good classification accuracy that could make these techniques suitable for fielding on autonomous devices. DRBMs appear to yield better accuracy than SVMs, and are less sensitive to the choice of signal preprocessing and model hyperparameters. This last property is especially appealing in such a task where the lack of data makes model validation difficult.