Since the objective is an expectation, we can use the Stochastic Gradient Descent (SGD) algorithm to minimize it, by iteratively minimizing the loss on sampled triplets of (user, positive item, negative item). Our implementation uses a form of parallel and asynchronous SGD updates (called Hogwild! [3]), which can give near-linear speedups in terms of the number of processors, especially when the data is sparse, so that concurrent updates are unlikely to be on the same parameters and overwrite each other. In the case of recommendation problems, the data is usually very sparse (unless, e.g., there are few items and most users have seen the same items), and this approach works very well.

Recognizing emotions in facial expressions is relatively straightforward for humans, and in recent times machines are getting better at it too. The applications of emotion-detecting computers are numerous, from improving advertising to treating depression, the possibilities are limitless. Motivated mainly by the impact in mental health that such technology can have, I started building my own emotion recognition technology. In a previous post I described a quick test in which I used ideas drawn from research on how facial expressions are decomposed. In this simplified scenario a computer distinguished between sad and happy faces by detecting facial landmarks (points of eyes, mouth, etc …) and using one simple geometric feature of the mouth (representing a Lip Corner Puller).

The final homework assignment for CS545 Machine Learning was to implement a K-means clustering algorithm to cluster and classify the OptDigits data. The raw data looks something like the figures to the left. So these instances are fields of 0's whereby some 0's have been flipped to be 1's such that the image is recognizable (to humans) as a handwritten digit. For the K-means classifier, we ran 2 different experiments. The first expeiment used 10 centroids (one per digit), the second used 30 centroids to see if it could find clusters where the handwritten digits were different enough to notice differences.

Effective training of deep neural networks suffers from two main issues. The first is that the parameter space of these models exhibit pathological curvature. Recent methods address this problem by using adaptive preconditioning for Stochastic Gradient Descent (SGD). These methods improve convergence by adapting to the local geometry of parameter space. A second issue is overfitting, which is typically addressed by early stopping. However, recent work has demonstrated that Bayesian model averaging mitigates this problem. The posterior can be sampled by using Stochastic Gradient Langevin Dynamics (SGLD). However, the rapidly changing curvature renders default SGLD methods inefficient. Here, we propose combining adaptive preconditioners with SGLD. In support of this idea, we give theoretical properties on asymptotic convergence and predictive risk. We also provide empirical results for Logistic Regression, Feedforward Neural Nets, and Convolutional Neural Nets, demonstrating that our preconditioned SGLD method gives state-of-the-art performance on these models.

We demonstrate a system for predicting gaming related properties from Twitter accounts. Our system predicts various traits of users based on the tweets publicly available in their profiles. Such inferred traits include degrees of tech-savviness and knowledge on computer games, actual gaming performance, preferred platform, degree of originality, humor and influence on others. Our system is based on machine learning models trained on crowd-sourced data. It allows people to select Twitter accounts of their fellow gamers, examine the trait predictions made by our system, and the main drivers of these predictions. We present empirical results on the performance of our system based on its accuracy on our crowd-sourced dataset.

In a wide range of statistical learning problems such as ranking, clustering or metric learning among others, the risk is accurately estimated by $U$-statistics of degree $d\geq 1$, i.e. functionals of the training data with low variance that take the form of averages over $k$-tuples. From a computational perspective, the calculation of such statistics is highly expensive even for a moderate sample size $n$, as it requires averaging $O(n^d)$ terms. This makes learning procedures relying on the optimization of such data functionals hardly feasible in practice. It is the major goal of this paper to show that, strikingly, such empirical risks can be replaced by drastically computationally simpler Monte-Carlo estimates based on $O(n)$ terms only, usually referred to as incomplete $U$-statistics, without damaging the $O_{\mathbb{P}}(1/\sqrt{n})$ learning rate of Empirical Risk Minimization (ERM) procedures. For this purpose, we establish uniform deviation results describing the error made when approximating a $U$-process by its incomplete version under appropriate complexity assumptions. Extensions to model selection, fast rate situations and various sampling techniques are also considered, as well as an application to stochastic gradient descent for ERM. Finally, numerical examples are displayed in order to provide strong empirical evidence that the approach we promote largely surpasses more naive subsampling techniques.

This paper considers survey prediction from social media. We use topic models to correlate social media messages with survey outcomes and to provide an interpretable representation of the data. Rather than rely on fully unsupervised topic models, we use existing aggregated survey data to inform the inferred topics, a class of topic model supervision referred to as collective supervision. We introduce and explore a variety of topic model variants and provide an empirical analysis, with conclusions of the most effective models for this task.

This paper presents a robust multi-view method for tracking people in 3D scene. Our method distinguishes itself from previous works in two aspects. Firstly, we define a set of binary spatial relationships for individual subjects or pairs of subjects that appear at the same time, e.g. being left or right, being closer or further to the camera, etc. These binary relationships directly reflect relative positions of subjects in 3D scene and thus should be persisted during inference. Secondly, we introduce an unified probabilistic framework to exploit binary spatial constraints for simultaneous 3D localization and cross-view human tracking. We develop a cluster Markov Chain Monte Carlo method to search the optimal solution. We evaluate our method on both public video benchmarks and newly built multi-view video dataset. Results with comparisons showed that our method could achieve state-of-the-art tracking results and meter-level 3D localization on challenging videos.

Unlike human learning, machine learning often fails to handle changes between training (source) and test (target) input distributions. Such domain shifts, common in practical scenarios, severely damage the performance of conventional machine learning methods. Supervised domain adaptation methods have been proposed for the case when the target data have labels, including some that perform very well despite being ``frustratingly easy'' to implement. However, in practice, the target domain is often unlabeled, requiring unsupervised adaptation. We propose a simple, effective, and efficient method for unsupervised domain adaptation called CORrelation ALignment (CORAL). CORAL minimizes domain shift by aligning the second-order statistics of source and target distributions, without requiring any target labels. Even though it is extraordinarily simple--it can be implemented in four lines of Matlab code--CORAL performs remarkably well in extensive evaluations on standard benchmark datasets.

We have designed an open and modular course for data science and big data analytics using a workflow paradigm that allows students to easily experience big data through a sophisticated yet easy to use instrument that is an intelligent workflow system. A key aspect of this work is the use of semantic workflows to capture and reuse end-to-end analytic methods that experts would use to analyze big data, and the use of an intelligent workflow system to elaborate the workflow and manage its execution and resulting datasets. Through the exposure of big data analytics in a workflow framework, students will be able to get first-hand experiences with a breadth of big data topics, including multi-step data analytic and statistical methods, software reuse and composition, parallel distributed programming, high-end computing. In addition, students learn about a range of topics in AI, including semantic representations and ontologies, machine learning, natural language processing, and image analysis.