These new applications require a new way of thinking about the development process. Traditional application development has been enhanced by the idea of DevOps, which forces operational considerations into development time, execution, and process. In this tutorial, we outline a "cognitive DevOps" process that refines and adapts the best parts of DevOps for new cognitive applications. Specifically, we cover applying DevOps to the training process of cognitive systems including training data, modeling, and performance evaluation. A cognitive or artificial intelligence (AI) system fundamentally exhibits capabilities such as understanding, reasoning, and learning from data. At a deeper level, the system is built upon a combination of various types of cognitive tasks, which, when combined, make up a part of the overall cognitive application. The science upon which a cognitive system is built includes, but is not limited to, machine learning (ML) including deep learning and natural language processing.

This paper discusses predictive inference and feature selection for generalized linear models with scarce but high-dimensional data. We argue that in many cases one can benefit from a decision theoretically justified two-stage approach: first, construct a possibly non-sparse model that predicts well, and then find a minimal subset of features that characterize the predictions. The model built in the first step is referred to as the \emph{reference model} and the operation during the latter step as predictive \emph{projection}. The key characteristic of this approach is that it finds an excellent tradeoff between sparsity and predictive accuracy, and the gain comes from utilizing all available information including prior and that coming from the left out features. We review several methods that follow this principle and provide novel methodological contributions. We present a new projection technique that unifies two existing techniques and is both accurate and fast to compute. We also propose a way of evaluating the feature selection process using fast leave-one-out cross-validation that allows for easy and intuitive model size selection. Furthermore, we prove a theorem that helps to understand the conditions under which the projective approach could be beneficial. The benefits are illustrated via several simulated and real world examples.

We propose shifted inner-product similarity (SIPS), which is a novel yet very simple extension of the ordinary inner-product similarity (IPS) for neural-network based graph embedding (GE). In contrast to IPS, that is limited to approximating positive-definite (PD) similarities, SIPS goes beyond the limitation by introducing bias terms in IPS; we theoretically prove that SIPS is capable of approximating not only PD but also conditionally PD (CPD) similarities with many examples such as cosine similarity, negative Poincare distance and negative Wasserstein distance. Since SIPS with sufficiently large neural networks learns a variety of similarities, SIPS alleviates the need for configuring the similarity function of GE. Approximation error rate is also evaluated, and experiments on two real-world datasets demonstrate that graph embedding using SIPS indeed outperforms existing methods.

In this paper, we try to explore the evolution of language through case calculations. First, we chose the novels of eleven British writers from 1400 to 2005 and found the corresponding works; Then, we use the natural language processing tool to construct the corresponding eleven corpora, and calculate the respective word vectors of 100 high-frequency words in eleven corpora; Next, for each corpus, we concatenate the 100 word vectors from beginning to end into one; Finally, we use the similarity comparison and hierarchical clustering method to generate the relationship tree between the combined eleven word vectors. This tree represents the relationship between eleven corpora. We found that in the tree generated by clustering, the distance between the corpus and the year corresponding to the corpus are basically the same. This means that we have discovered a specific language evolution tree. To verify the stability and versatility of this method, we add three other themes: Dickens's eight works, the 19th century poets' works, and art criticism of recent 60 years. For these four themes, we tested different parameters such as the time span of the corpus, the time interval between the corpora, the dimension of the word vector, and the number of high-frequency public words. The results show that this is fairly stable and versatile.

Neural networks trained by first order methods have achieved a remarkable impact on many applications, but their theoretical properties are still mysteries. One of the empirical observation is even though the optimization objective function is non-convex and non-smooth, randomly initialized first order methods like stochastic gradient descent can still find a global minimum. Surprisingly, this property is not correlated with labels. In Zhang et al. [2016], authors replaced the true labels with randomly generated labels, but still found randomly initialized first order methods can always achieve zero training loss. A widely believed explanation on why a neural network can fit all training labels is because the neural network is over-parameterized. For example, Wide ResNet [Zagoruyko and Komodakis] uses 100x parameters than the number of training data and thus there must exist one such neural network of this architecture that can fit all training data. However, the existence does not imply why the network found by a randomly initialized first order method can fit all the data.

Gaussian processes (GPs) offer a flexible class of priors for nonparametric Bayesian regression, but popular GP posterior inference methods are typically prohibitively slow or lack desirable finite-data guarantees on quality. We develop an approach to scalable approximate GP regression with finite-data guarantees on the accuracy of pointwise posterior mean and variance estimates. Our main contribution is a novel objective for approximate inference in the nonparametric setting: the preconditioned Fisher (pF) divergence. We show that unlike the Kullback--Leibler divergence (used in variational inference), the pF divergence bounds the 2-Wasserstein distance, which in turn provides tight bounds the pointwise difference of the mean and variance functions. We demonstrate that, for sparse GP likelihood approximations, we can minimize the pF divergence efficiently. Our experiments show that optimizing the pF divergence has the same computational requirements as variational sparse GPs while providing comparable empirical performance--in addition to our novel finite-data quality guarantees.

Many real-world visual recognition use-cases can not directly benefit from state-of-the-art CNN-based approaches because of the lack of many annotated data. The usual approach to deal with this is to transfer a representation pre-learned on a large annotated source-task onto a target-task of interest. This raises the question of how well the original representation is "universal", that is to say directly adapted to many different target-tasks. To improve such universality, the state-of-the-art consists in training networks on a diversified source problem, that is modified either by adding generic or specific categories to the initial set of categories. In this vein, we proposed a method that exploits finer-classes than the most specific ones existing, for which no annotation is available. We rely on unsupervised learning and a bottom-up split and merge strategy. We show that our method learns more universal representations than state-of-the-art, leading to significantly better results on 10 target-tasks from multiple domains, using several network architectures, either alone or combined with networks learned at a coarser semantic level.

However, some seasoned techniques are here to stay. At the top of the list are regression techniques. As long as this number is as high, you will encounter regressions during your machine learning career. Even if you don't use them yourself, it is essential to be aware of the different flavors and which problems they tackle. In this post, I provide you with a quick overview of five different (groups of) regressions.

We present a new method for uncertainty estimation and out-of-distribution detection in neural networks with softmax output. We extend softmax layer with an additional constant input. The corresponding additional output is able to represent the uncertainty of the network. The proposed method requires neither additional parameters nor multiple forward passes nor input preprocessing nor out-of-distribution datasets. We show that our method performs comparably to more computationally expensive methods and outperforms baselines on our experiments from image recognition and sentiment analysis domains. The applications of computational learning systems might cause intrusive effects if we assume that predictions are always as accurate as during the experimental phase. Examples include misclassified traffic signs (Evtimov et al., 2018) and an image tagger that classified two African Americans as gorillas (Curtis, 2015). This is often caused by overconfidence of models that has been observed in the case of deep neural networks (Guo et al., 2017). Such malfunctions can be prevented if we estimate correctly the uncertainty of the machine learning system.

In this paper, we consider linear prediction models in the form of a sparse linear combination of rules, where a rule is an indicator function defined over a hyperrectangle in the input space. Since the number of all possible rules generated from the training dataset becomes extremely large, it has been difficult to consider all of them when fitting a sparse model. In this paper, we propose Safe Optimal Rule Fit (SORF) as an approach to resolve this problem, which is formulated as a convex optimization problem with sparse regularization. The proposed SORF method utilizes the fact that the set of all possible rules can be represented as a tree. By extending a recently popularized convex optimization technique called safe screening, we develop a novel method for pruning the tree such that pruned nodes are guaranteed to be irrelevant to the prediction model. This approach allows us to efficiently learn a prediction model constructed from an exponentially large number of all possible rules. We demonstrate the usefulness of the proposed method by numerical experiments using several benchmark datasets.