Recently, I was wondering how I could escape the web development bubble for a while. The last year was all about those subjects, so I needed a side project to escape it for a while and to get into a zen mode of learning again. How did I get to machine learning? A couple of months ago, I started to listen to the Machine Learning Guide podcast. I found out about it by chance and highly recommend it to get you an introduction for machine learning. Tyler Renelle is doing an amazing job to get you excited about the topic. I almost feel like I am following him on the same path to learn about machine learning now. Even though I didn't actively plan about learning ML, it was interesting to hear about all those foreign concepts. There it was again; this excitement when everything is unexplored. I felt like a whole new world opened up in front of me. It was the same feeling when I finally got the foot into web development. As I read about a couple of machine learning articles, the course on Machine Learning by Andrew Ng was the by far most recommended to get started in machine learning. I have never taken an online course from start to end before, even though I actively give these online courses myself, but I decided to give it a shot this time. Fortunately, the course had started one week ago. So I enrolled in it and by now finished it.

Unsolicited bulk emails, also known as Spam, make up for approximately 60% of the global email traffic. Despite the fact that technology has advanced in the field of Spam detection since the first unsolicited bulk email was sent in 1978 spamming remains a time consuming and expensive problem. This report compares the performance of three machine learning techniques for spam detection including Random Forest (RF), k-Nearest Neighbours (kNN) and Support Vector Machines (SVM). Despite the rising popularity of instant messaging technologies in recent years, email continues to be the dominant medium for digital communications for both consumer and business use. Following industry estimations (Symantec Corporation, 2016, pp 31 1), approximately 200 billion emails were sent each day in 2015.

Linear least-squares regression with a "design" matrix A approximates a given matrix B via minimization of the spectral- or Frobenius-norm discrepancy ||AX-B|| over every conformingly sized matrix X. Another popular approximation is low-rank approximation via principal component analysis (PCA) -- which is essentially singular value decomposition (SVD) -- or interpolative decomposition (ID). Classically, PCA/SVD and ID operate solely with the matrix B being approximated, not supervised by any auxiliary matrix A. However, linear least-squares regression models can inform the ID, yielding regression-aware ID. As a bonus, this provides an interpretation as regression-aware PCA for a kind of canonical correlation analysis between A and B. The regression-aware decompositions effectively enable supervision to inform classical dimensionality reduction, which classically has been totally unsupervised. The regression-aware decompositions reveal the structure inherent in B that is relevant to regression against A.

A problem facing many services - from search engines and news feeds to machine learning - is data summarization: how can one select a small but representative, i.e., diverse, subset from a large dataset. For instance, Google Images outputs a small subset of images from its enormous dataset given a user query. Similarly, in training a learning algorithm one may be required to choose a subset of data points to train on as training on the entire dataset may be costly. However, data summarization algorithms prevalent in the online world have been recently shown to be biased with respect to sensitive attributes such as gender, race and ethnicity. For instance, a recent study found evidence of systematic under-representation of women in search results [14]. Concretely, the above work studied the output of Google Images for various search terms involving occupations and found, e.g., that for the search term "CEO", the percentage of women in top 100 results was 11%, significantly lower than the ground truth of 27%. Through studies on human subjects, they also found that such misrepresentations have the power to influence people's perception about reality. Beyond humans, since data summaries are used to train algorithms, there is a danger that these biases in the data might be passed on to the algorithms that use them; a phenomena that is being revealed more and more in automated data-driven processes in education, recruitment, banking, and judiciary systems, see [22]. A robust and widely deployed method for data summarization is to associate a diversity score to each subset and select a subset with probability proportional to this score; see [13].

An interactive image retrieval system learns which images in the database belong to a user's query concept, by analyzing the example images and feedback provided by the user. The challenge is to retrieve the relevant images with minimal user interaction. In this work, we propose to solve this problem by posing it as a binary classification task of classifying all images in the database as being relevant or irrelevant to the user's query concept. Our method combines active learning with graph-based semi-supervised learning (GSSL) to tackle this problem. Active learning reduces the number of user interactions by querying the labels of the most informative points and GSSL allows to use abundant unlabeled data along with the limited labeled data provided by the user. To efficiently find the most informative point, we use an uncertainty sampling based method that queries the label of the point nearest to the decision boundary of the classifier. We estimate this decision boundary using our heuristic of adaptive threshold. To utilize huge volumes of unlabeled data we use an efficient approximation based method that reduces the complexity of GSSL from $O(n^3)$ to $O(n)$, making GSSL scalable. We make the classifier robust to the diversity and noisy labels associated with images in large databases by incorporating information from multiple modalities such as visual information extracted from deep learning based models and semantic information extracted from the WordNet. High F1 scores within few relevance feedback rounds in our experiments with concepts defined on AnimalWithAttributes and Imagenet (1.2 million images) datasets indicate the effectiveness and scalability of our approach.

The workflow of data scientists normally involves potentially inefficient processes such as data mining, feature engineering and model selection. Recent research has focused on automating this workflow, partly or in its entirety, to improve productivity. We choose the former approach and in this paper share our experience in designing the client2vec: an internal library to rapidly build baselines for banking applications. Client2vec uses marginalized stacked denoising autoencoders on current account transactions data to create vector embeddings which represent the behaviors of our clients. These representations can then be used in, and optimized against, a variety of tasks such as client segmentation, profiling and targeting. Here we detail how we selected the algorithmic machinery of client2vec and the data it works on and present experimental results on several business cases.

Representation learning algorithms are designed to learn abstract features that characterize data. State representation learning (SRL) focuses on a particular kind of representation learning where learned features are in low dimension, evolve through time, and are influenced by actions of an agent. As the representation learned captures the variation in the environment generated by agents, this kind of representation is particularly suitable for robotics and control scenarios. In particular, the low dimension helps to overcome the curse of dimensionality, provides easier interpretation and utilization by humans and can help improve performance and speed in policy learning algorithms such as reinforcement learning. This survey aims at covering the state-of-the-art on state representation learning in the most recent years. It reviews different SRL methods that involve interaction with the environment, their implementations and their applications in robotics control tasks (simulated or real). In particular, it highlights how generic learning objectives are differently exploited in the reviewed algorithms. Finally, it discusses evaluation methods to assess the representation learned and summarizes current and future lines of research.

Healthcare companies must submit pharmaceutical drugs or medical devices to regulatory bodies before marketing new technology. Regulatory bodies frequently require transparent and interpretable computational modelling to justify a new healthcare technology, but researchers may have several competing models for a biological system and too little data to discriminate between the models. In design of experiments for model discrimination, the goal is to design maximally informative physical experiments in order to discriminate between rival predictive models. Prior work has focused either on analytical approaches, which cannot manage all functions, or on data-driven approaches, which may have computational difficulties or lack interpretable marginal predictive distributions. We develop a methodology introducing Gaussian process surrogates in lieu of the original mechanistic models. We thereby extend existing design and model discrimination methods developed for analytical models to cases of non-analytical models in a computationally efficient manner.

Bitcoin is the first decentralized digital cryptocurrency, which has showed significant market capitalization growth in last few years. It is important to understand what drives the fluctuations of the Bitcoin exchange price and to what extent they are predictable. In this paper, we study the ability to make short-term prediction of the exchange price fluctuations (measured with volatility) towards the United States dollar. We use the data of buy and sell orders collected from one of the largest Bitcoin digital trading offices in 2016 and 2017. We construct a generative temporal mixture model of the volatility and trade order book data, which is able to out-perform the current state-of-the-art machine learning and time-series statistical models. With the gate weighting function of our generative temporal mixture model, we are able to detect regimes when the features of buy and sell orders significantly affects the future high volatility periods. Furthermore, we provide insights into dynamical importance of specific features from order book such as market spread, depth, volume and ask/bid slope to explain future short-term price fluctuations.

We study and empirically optimize contextual bandit learning, exploration, and problem encodings across 500+ datasets, creating a reference for practitioners and discovering or reinforcing a number of natural open problems for researchers. Across these experiments we show that minimizing the amount of exploration is a key design goal for practical performance. Remarkably, many problems can be solved purely via the implicit exploration imposed by the diversity of contexts. For practitioners, we introduce a number of practical improvements to common exploration algorithms including Bootstrap Thompson sampling, Online Cover, and $\epsilon$-greedy. We also detail a new form of reduction to regression for learning from exploration data. Overall, this is a thorough study and review of contextual bandit methodology.