IBM's Watson Education, an artificial intelligence platform that uses data trends to provide insights to teachers and students, is partnering with Edmodo and Scholastic in an effort meant to personalize learning. With Edmodo, a K-12 network for students, teachers, administrators and parents, IBM is collaborating to develop a personalized content recommendation engine that can be integrated within Edmodo's existing social education platform. For Scholastic, a children's publishing, education and media company, the plan is to use the Watson platform to recommend nonfiction content that aligns with curriculum standards and has multiple articles and media for students' skill and interest levels. "Our goal is to use AI to improve learning outcomes," and to personalize content for learners, said Chalapathy Neti, vice president of IBM Watson Education, in an interview. He explained that he refers to AI as "augmented intelligence" rather than the more typical "artificial intelligence," because the way people are thinking about the abbreviated "AI" has produced "a little bit of angst in terms of machines replacing humans."

Many applications in machine learning require optimizing a function whose true gradient is unknown, but where surrogate gradient information (directions that may be correlated with, but not necessarily identical to, the true gradient) is available instead. This arises when an approximate gradient is easier to compute than the full gradient (e.g. in meta-learning or unrolled optimization), or when a true gradient is intractable and is replaced with a surrogate (e.g. in certain reinforcement learning applications, or when using synthetic gradients). We propose Guided Evolutionary Strategies, a method for optimally using surrogate gradient directions along with random search. We define a search distribution for evolutionary strategies that is elongated along a guiding subspace spanned by the surrogate gradients. This allows us to estimate a descent direction which can then be passed to a first-order optimizer. We analytically and numerically characterize the tradeoffs that result from tuning how strongly the search distribution is stretched along the guiding subspace, and we use this to derive a setting of the hyperparameters that works well across problems. Finally, we apply our method to example problems including truncated unrolled optimization and a synthetic gradient problem, demonstrating improvement over both standard evolutionary strategies and first-order methods that directly follow the surrogate gradient.

Target encoding plays a central role when learning Convolutional Neural Networks. In this realm, One-hot encoding is the most prevalent strategy due to its simplicity. However, this so widespread encoding schema assumes a flat label space, thus ignoring rich relationships existing among labels that can be exploited during training. In large-scale datasets, data does not span the full label space, but instead lies in a low-dimensional output manifold. Following this observation, we embed the targets into a low-dimensional space, drastically improving convergence speed while preserving accuracy. Our contribution is two fold: (i) We show that random projections of the label space are a valid tool to find such lower dimensional embeddings, boosting dramatically convergence rates at zero computational cost; and (ii) we propose a normalized eigenrepresentation of the class manifold that encodes the targets with minimal information loss, improving the accuracy of random projections encoding while enjoying the same convergence rates. Experiments on CIFAR-100, CUB200-2011, Imagenet, and MIT Places demonstrate that the proposed approach drastically improves convergence speed while reaching very competitive accuracy rates.

In this work we aim to solve a large collection of tasks using a single reinforcement learning agent with a single set of parameters. A key challenge is to handle the increased amount of data and extended training time. We have developed a new distributed agent IMPALA (Importance Weighted Actor-Learner Architecture) that not only uses resources more efficiently in singlemachine training but also scales to thousands of machines without sacrificing data efficiency or resource utilisation. We achieve stable learning at high throughput by combining decoupled acting and learning with a novel off-policy correction method called V-trace. We demonstrate the effectiveness of IMPALA for multi-task reinforcement learning on DMLab-30 (a set of 30 tasks from the DeepMind Lab environment (Beattie et al., 2016)) and Atari-57 (all available Atari games in Arcade Learning Environment (Bellemare et al., 2013a)). Our results show that IMPALA is able to achieve better performance than previous agents with less data, and crucially exhibits positive transfer between tasks as a result of its multi-task approach. The source code is publicly available at github.com/deepmind/scalable

Transfer learning can greatly speed up reinforcement learning for a new task by leveraging policies of relevant tasks. Existing works of policy reuse either focus on only selecting a single best source policy for transfer without considering contexts, or cannot guarantee to learn an optimal policy for a target task. To improve transfer efficiency and guarantee optimality, we develop a novel policy reuse method, called Context-Aware Policy reuSe (CAPS), that enables multi-policy transfer. Our method learns when and which source policy is best for reuse, as well as when to terminate its reuse. CAPS provides theoretical guarantees in convergence and optimality for both source policy selection and target task learning. Empirical results on a grid-based navigation domain and the Pygame Learning Environment demonstrate that CAPS significantly outperforms other state-of-the-art policy reuse methods.

One important assumption underlying common classification models is the stationarity of the data. However, in real-world streaming applications, the data concept indicated by the joint distribution of feature and label is not stationary but drifting over time. Concept drift detection aims to detect such drifts and adapt the model so as to mitigate any deterioration in the model's predictive performance. Unfortunately, most existing concept drift detection methods rely on a strong and over-optimistic condition that the true labels are available immediately for all already classified instances. In this paper, a novel Hierarchical Hypothesis Testing framework with Request-and-Reverify strategy is developed to detect concept drifts by requesting labels only when necessary. Two methods, namely Hierarchical Hypothesis Testing with Classification Uncertainty (HHT-CU) and Hierarchical Hypothesis Testing with Attribute-wise "Goodness-of-fit" (HHT-AG), are proposed respectively under the novel framework. In experiments with benchmark datasets, our methods demonstrate overwhelming advantages over state-of-the-art unsupervised drift detectors. More importantly, our methods even outperform DDM (the widely used supervised drift detector) when we use significantly fewer labels.

The web is full of good explanations of machine learning algorithms. And every second applicant for a data science position has finished the Coursera course on machine learning. Theory will not help you choose good values for the 16 parameters a standard implementation of a random forest takes. The default values are good to get started, but which parameters should you modify depending on your data? Choosing the right features, algorithms and parameters is an art.

Often when we hear about artificial intelligence in legal it's addressed from a high-level, philosophical perspective that sometimes ignores the immediate use cases for practicing lawyers. Flying in the face of this, AI in legal took a major step forward on June 20 at the University of Chicago's Gleacher Center where leaders from law firms, legal technology providers, law schools and in-house legal departments gathered to examine AI's convergence within specific areas of legal, namely: e-discovery, contract review, contract analysis, litigation, and of course, legal research. There are roughly 900 legal tech startups in the legal ecosystem all attempting to improve how law is practice from solo shops to the biggest firms in the world. Among these tech providers, AI-based tools are becoming more widely accepted and better understood among lawyers. In an encouraging sign, more corporate clients are demanding their outside counsel use these technologies to be more accurate, more innovative and more efficient.

Garbage In/Garbage Out (GIGO) Many projects fail because project managers forget to check data quality or do not have the right approach to identify and resolve these issues. When we analyze incomplete or "dirty" data sets, our AI ends up making decisions and recommendations based on a poor foundation. Apples and Oranges Comparing unrelated data sets and/or data points will result in inferring relationships or similarities that do not exist. Overly Narrow Focus Some projects are designed to consider one data set without considering other data points that might be crucial for the analysis. For example, a project set up to analyze learner pass/fail rates while ignoring the course completion rate may inflate performance results. Cool but Useless Some AI projects are quick to deliver but fail to make a significant impact on the learner's everyday experience. Ensure that you have the right strategy to deliver the most value to your learners and avoid giving them something cool that doesn't really help them learn. My advice is to just get on with it. Make a point of learning something about AI and machine learning every day, always with an eye to how you might be able to use it in your own organization.

It may be the most exclusive school in the world. Housed somewhere inside the sprawling Hawthorne, Calif., headquarters of rocket manufacturer SpaceX, Ad Astra reportedly has less than 50 students between seven and 14-years old, perpetually-evolving curriculum, and no formal grading. Instead, Ars Technica reports, the mysterious not-for-profit school functions like a "venture capital incubator" in which students work in teams to drill into some of the most daunting topics of our time: robotics, nuclear politics and the dangers posed by artificial intelligence. The latter is no surprise considering that Ad Astra was founded by Elon Musk, the billionaire inventor who has been stridently warning about the risks posed by intelligent machines for several years now. "I just didn't see that the regular schools were doing the things that I thought should be done," Musk told a Chinese TV station in 2015.