Forecasts are typically not produced in a vacuum but in a business context, where forecasts are generated on a regular basis and interact with each other. For decisions, it may be important that forecasts do not change arbitrarily, and are stable in some sense. However, this area has received only limited attention in the forecasting literature. In this paper, we explore two types of forecast stability that we call vertical stability and horizontal stability. The existing works in the literature are only applicable to certain base models and extending these frameworks to be compatible with any base model is not straightforward. Furthermore, these frameworks can only stabilise the forecasts vertically. To fill this gap, we propose a simple linear-interpolation-based approach that is applicable to stabilise the forecasts provided by any base model vertically and horizontally. The approach can produce both accurate and stable forecasts. Using N-BEATS, Pooled Regression and LightGBM as the base models, in our evaluation on four publicly available datasets, the proposed framework is able to achieve significantly higher stability and/or accuracy compared to a set of benchmarks including a state-of-the-art forecast stabilisation method across three error metrics and six stability metrics.

The ability to perform offline A/B-testing and off-policy learning using logged contextual bandit feedback is highly desirable in a broad range of applications, including recommender systems, search engines, ad placement, and personalized health care. Both offline A/B-testing and off-policy learning require a counterfactual estimator that evaluates how some new policy would have performed, if it had been used instead of the logging policy. This paper proposes a new counterfactual estimator - called Continuous Adaptive Blending (CAB) - for this policy evaluation problem that combines regression and weighting approaches for an effective bias/variance trade-off. It can be substantially less biased than clipped Inverse Propensity Score weighting and the Direct Method, and it can have less variance compared with Doubly Robust and IPS estimators. Experimental results show that CAB provides excellent and reliable estimation accuracy compared to other blended estimators, and - unlike the SWITCH estimator - is sub-differentiable such that it can be used for learning.

In this note we consider setups in which variational objectives for Bayesian neural networks can be computed in closed form. In particular we focus on single-layer networks in which the activation function is piecewise polynomial (e.g. ReLU). In this case we show that for a Normal likelihood and structured Normal variational distributions one can compute a variational lower bound in closed form. In addition we compute the predictive mean and variance in closed form. Finally, we also show how to compute approximate lower bounds for other likelihoods (e.g. softmax classification). In experiments we show how the resulting variational objectives can help improve training and provide fast test time predictions.

We propose a probabilistic framework to directly insert prior knowledge in reinforcement learning (RL) algorithms by defining the behaviour policy as a Bayesian posterior distribution. Such a posterior combines task specific information with prior knowledge, thus allowing to achieve transfer learning across tasks. The resulting method is flexible and it can be easily incorporated to any standard off-policy and on-policy algorithms, such as those based on temporal differences and policy gradients. We develop a specific instance of this Bayesian transfer RL framework by expressing prior knowledge as general deterministic rules that can be useful in a large variety of tasks, such as navigation tasks. Also, we elaborate more on recent probabilistic and entropy-regularised RL by developing a novel temporal learning algorithm and show how to combine it with Bayesian transfer RL. Finally, we demonstrate our method for solving mazes and show that significant speed ups can be obtained.

We are interested in learning customers' video preferences from their historic viewing patterns and geographical location. We consider a Bayesian latent factor modeling approach for this task. In order to tune the complexity of the model to best represent the data, we make use of Bayesian nonparameteric techniques. We describe an inference technique that can scale to large real-world data sets. Finally we show results obtained by applying the model to a large internal Netflix data set, that illustrates that the model was able to capture interesting relationships between viewing patterns and geographical location.

Machine Learning manuscript No. (will be inserted by the editor)Asymptotic consistency and order specification for logistic classifier chains in multi-label learning Paweł T eisseyre Received: date / Accepted: date Abstract Classifier chains are popular and effective method to tackle a multi-label classification problem. The aim of this paper is to study the asymptotic properties of the chain model in which the conditional probabilities are of the logistic form. In particular we find conditions on the number of labels and the distribution of feature vector under which the estimated mode of the joint distribution of labels converges to the true mode. Best of our knowledge, this important issue has not yet been studied in the context of multi-label learning. We also investigate how the order of model building in a chain influences the estimation of the joint distribution of labels. We establish the link between the problem of incorrect ordering in the chain and incorrect model specification. We propose a procedure of determining the optimal ordering of labels in the chain, which is based on using measures of correct specification and allows to find the ordering such that the consecutive logistic models are best possibly specified. The other important question raised in this paper is how accurately can we estimate the joint posterior probability when the ordering of labels is wrong or the logistic models in the chain are incorrectly specified. The numerical experiments illustrate the theoretical results. Keywords classifier chains· logistic regression· joint mode estimation· label ordering· asymptotic consistency 1 Introduction In multi-label classification the task is to automatically assign an object to multiple categories based on its characteristics. Each object of our interest is described by a feature vector x belonging to p-dimensional space and vector of K labels y ( y 1,..., y K)′ . In this paper we consider binary labels such thaty k 1 indicates that the considered object belongs to k-th category or has the k-th property. The issue has recently attracted significant attention, motivated by an increasing number of applications such as image and video annotationPaweł Teisseyre Institute of Computer Science, Polish Academy of Sciences Jana Kazimierza 5 01-248 Warsaw, Poland Tel.: 48-22-380-05-55 Email: teisseyrep@ipipan.waw.pl

As part of daily monitoring of human activities, wearable sensors and devices are becoming increasingly popular sources of data. With the advent of smartphones equipped with acceloremeter, gyroscope and camera; it is now possible to develop activity classification platforms everyone can use conveniently. In this paper, we propose a fast inference method for an unsupervised non-parametric time series model namely variational inference for sticky HDP-SLDS(Hierarchical Dirichlet Process Switching Linear Dynamical System). We show that the proposed algorithm can differentiate various indoor activities such as sitting, walking, turning, going up/down the stairs and taking the elevator using only the acceloremeter of an Android smartphone Samsung Galaxy S4. We used the front camera of the smartphone to annotate activity types precisely. We compared the proposed method with Hidden Markov Models with Gaussian emission probabilities on a dataset of 10 subjects. We showed that the efficacy of the stickiness property. We further compared the variational inference to the Gibbs sampler on the same model and show that variational inference is faster in one order of magnitude.

Hidden Markov models have successfully been applied as models of discrete time series in many fields. Often, when applied in practice, the parameters of these models have to be estimated. The currently predominating identification methods, such as maximum-likelihood estimation and especially expectation-maximization, are iterative and prone to have problems with local minima. A non-iterative method employing a spectral subspace-like approach has recently been proposed in the machine learning literature. This paper evaluates the performance of this algorithm, and compares it to the performance of the expectation-maximization algorithm, on a number of numerical examples. We find that the performance is mixed; it successfully identifies some systems with relatively few available observations, but fails completely for some systems even when a large amount of observations is available. An open question is how this discrepancy can be explained. We provide some indications that it could be related to how well-conditioned some system parameters are.

Pattern recognition is a central topic in Learning Theory with numerous applications such as voice and text recognition, image analysis, computer diagnosis. The statistical set-up in classification is the following: we are given an i.i.d. training set $(X_{1},Y_{1}),... (X_{n},Y_{n})$ where $X_{i}$ represents a feature and $Y_{i}\in \{0,1\}$ is a label attached to that feature. The underlying joint distribution of $(X,Y)$ is unknown, but we can learn about it from the training set and we aim at devising low error classifiers $f:X\to Y$ used to predict the label of new incoming features. Here we solve a quantum analogue of this problem, namely the classification of two arbitrary unknown qubit states. Given a number of `training' copies from each of the states, we would like to `learn' about them by performing a measurement on the training set. The outcome is then used to design mesurements for the classification of future systems with unknown labels. We find the asymptotically optimal classification strategy and show that typically, it performs strictly better than a plug-in strategy based on state estimation. The figure of merit is the excess risk which is the difference between the probability of error and the probability of error of the optimal measurement when the states are known, that is the Helstrom measurement. We show that the excess risk has rate $n^{-1}$ and compute the exact constant of the rate.