We present WatsonPaths, a novel system that can answer scenario-based questions. These include medical questions that present a patient summary and ask for the most likely diagnosis or most appropriate treatment. WatsonPaths builds on the IBM Watson question answering system. WatsonPaths breaks down the input scenario into individual pieces of information, asks relevant subquestions of Watson to conclude new information, and represents these results in a graphical model. Probabilistic inference is performed over the graph to conclude the answer. On a set of medical test preparation questions, WatsonPaths shows a significant improvement in accuracy over multiple baselines.

In fall 2014, we launched a foundational course in artificial intelligence (CS7637: Knowledge-Based AI) as part of the Georgia Institute of Technology's Online Master of Science in Computer Science program. We incorporated principles and practices from the cognitive and learning sciences into the development of the online AI course. We also integrated AI techniques into the instruction of the course, including embedding 100 highly focused intelligent tutoring agents in the video lessons. By now, more than 2000 students have taken the course. Evaluations have indicated that OMSCS students enjoy the course compared to traditional courses, and more importantly, that online students have matched residential students' performance on the same assessments. In this article, we present the design, delivery, and evaluation of the course, focusing on the use of AI for teaching AI. We also discuss lessons we learned for scaling the teaching and learning of AI.

Different subfields of AI (such as vision, learning, reasoning, planning, and others) are often studied in isolation, both in individual courses and in the research literature. This promulgates the idea that these different AI capabilities can easily be integrated later, whereas, in practice, developing integrated AI systems remains an open challenge for both research and industry. Interdisciplinary project-driven courses can fill this gap in AI education, providing challenging problems that require the integration of multiple AI methods. This article explores teaching integrated AI through two project-driven courses: a capstone-style graduate course in advanced robotics, and an undergraduate course on computational sustainability and assistive computing. In addition to studying the integration of AI techniques, these courses provide students with practical applications experience and exposure to social issues of AI and computing. My hope is that other instructors find these courses as useful examples for constructing their own project-driven courses to teach integrated AI.

Additional landmark events in the past 20 or so years that looked at the challenges of AI education have included the AI Education Workshop held at the 2008 AAAI conference and the Improving Instruction of Introductory Artificial Intelligence symposium held at the 1994 AAAI Fall Symposium. To quote Marti Hearst, the organizer of the 1994 symposium (Hearst 1994): "This symposium was motivated by the desire to address an oft-voiced complaint that introductory artificial intelligence is a notoriously difficult course to teach well." With the regular progression of the field and recent successes such as autonomous cars, deep learning, and IBM's Watson system, this situation has not become easier. At the same time, recent innovations in pedagogical technologies, such as massive open online courses (MOOCs), smartphones, and smart classrooms, have revolutionized how we view the art of teaching. We believe that now is a good time to take stock of state-of-the-art practices in the teaching of AI, as well as propose a vision for AI education in the future. This issue of AI Magazine includes five articles at the cutting edge of AI education. Each covers a subject of current concern to the AI education community. We note that the subject area expertise of the authors covers a wide range including robotics, knowledge-based systems, ethics, machine learning, and game theory. The article Ask Me Anything About MOOCs by Douglas Fisher, Charles Isbell, and Michael Littman was a unique project.

Correlated topic modeling has been limited to small model and problem sizes due to their high computational cost and poor scaling. In this paper, we propose a new model which learns compact topic embeddings and captures topic correlations through the closeness between the topic vectors. Our method enables efficient inference in the low-dimensional embedding space, reducing previous cubic or quadratic time complexity to linear w.r.t the topic size. We further speedup variational inference with a fast sampler to exploit sparsity of topic occurrence. Extensive experiments show that our approach is capable of handling model and data scales which are several orders of magnitude larger than existing correlation results, without sacrificing modeling quality by providing competitive or superior performance in document classification and retrieval.

In many applications involving large dataset or online updating, stochastic gradient descent (SGD) provides a scalable way to compute parameter estimates and has gained increasing popularity due to its numerical convenience and memory efficiency. While the asymptotic properties of SGD-based estimators have been established decades ago, statistical inference such as interval estimation remains much unexplored. The traditional resampling method such as the bootstrap is not computationally feasible since it requires to repeatedly draw independent samples from the entire dataset. The plug-in method is not applicable when there are no explicit formulas for the covariance matrix of the estimator. In this paper, we propose a scalable inferential procedure for stochastic gradient descent, which, upon the arrival of each observation, updates the SGD estimate as well as a large number of randomly perturbed SGD estimates. The proposed method is easy to implement in practice. We establish its theoretical properties for a general class of models that includes generalized linear models and quantile regression models as special cases. The finite-sample performance and numerical utility is evaluated by simulation studies and two real data applications.

When devising a course of treatment for a patient, doctors often have little quantitative evidence on which to base their decisions, beyond their medical education and published clinical trials. Stanford Health Care alone has millions of electronic medical records (EMRs) that are only just recently being leveraged to inform better treatment recommendations. These data present a unique challenge because they are high-dimensional and observational. Our goal is to make personalized treatment recommendations based on the outcomes for past patients similar to a new patient. We propose and analyze three methods for estimating heterogeneous treatment effects using observational data. Our methods perform well in simulations using a wide variety of treatment effect functions, and we present results of applying the two most promising methods to data from The SPRINT Data Analysis Challenge, from a large randomized trial of a treatment for high blood pressure.

We examine the role of memorization in deep learning, drawing connections to capacity, generalization, and adversarial robustness. While deep networks are capable of memorizing noise data, our results suggest that they tend to prioritize learning simple patterns first. In our experiments, we expose qualitative differences in gradient-based optimization of deep neural networks (DNNs) on noise vs. real data. We also demonstrate that for appropriately tuned explicit regularization (e.g., dropout) we can degrade DNN training performance on noise datasets without compromising generalization on real data. Our analysis suggests that the notions of effective capacity which are dataset independent are unlikely to explain the generalization performance of deep networks when trained with gradient based methods because training data itself plays an important role in determining the degree of memorization.

We consider the minimization of non-convex functions that typically arise in machine learning. Specifically, we focus our attention on a variant of trust region methods known as cubic regularization. This approach is particularly attractive because it escapes strict saddle points and it provides stronger convergence guarantees than first- and second-order as well as classical trust region methods. However, it suffers from a high computational complexity that makes it impractical for large-scale learning. Here, we propose a novel method that uses sub-sampling to lower this computational cost. By the use of concentration inequalities we provide a sampling scheme that gives sufficiently accurate gradient and Hessian approximations to retain the strong global and local convergence guarantees of cubically regularized methods. To the best of our knowledge this is the first work that gives global convergence guarantees for a sub-sampled variant of cubic regularization on non-convex functions. Furthermore, we provide experimental results supporting our theory.

Colorization of grayscale images has been a hot topic in computer vision. Previous research mainly focuses on producing a colored image to match the original one. However, since many colors share the same gray value, an input grayscale image could be diversely colored while maintaining its reality. In this paper, we design a novel solution for unsupervised diverse colorization. Specifically, we leverage conditional generative adversarial networks to model the distribution of real-world item colors, in which we develop a fully convolutional generator with multi-layer noise to enhance diversity, with multi-layer condition concatenation to maintain reality, and with stride 1 to keep spatial information. With such a novel network architecture, the model yields highly competitive performance on the open LSUN bedroom dataset. The Turing test of 80 humans further indicates our generated color schemes are highly convincible.