In this paper we propose strategies for estimating performance of a classifier when labels cannot be obtained for the whole test set. The number of test instances which can be labeled is very small compared to the whole test data size. The goal then is to obtain a precise estimate of classifier performance using as little labeling resource as possible. Specifically, we try to answer, how to select a subset of the large test set for labeling such that the performance of a classifier estimated on this subset is as close as possible to the one on the whole test set. We propose strategies based on stratified sampling for selecting this subset. We show that these strategies can reduce the variance in estimation of classifier accuracy by a significant amount compared to simple random sampling (over 65% in several cases). Hence, our proposed methods are much more precise compared to random sampling for accuracy estimation under restricted labeling resources. The reduction in number of samples required (compared to random sampling) to estimate the classifier accuracy with only 1% error is high as 60% in some cases.

Scientific and engineering processes produce massive high-dimensional data sets that are generated as highly non-linear transformations of an initial state and few process parameters. Mapping such data to a low-dimensional manifold can facilitate better understanding of the underlying process, and ultimately their optimization. We show that off-the-shelf non-linear spectral dimensionality methods, such as Isomap, fail for such data, primarily due to the presence of strong temporal correlation among observations belonging to the same process pathways. We propose a novel method, Entropy-Isomap, to address this issue. The proposed method is successfully applied to morphology evolution data of the organic thin film fabrication process. The resulting output is ordered by the process variables. It allows for low-dimensional visualization of the morphological pathways, and provides key insights to guide subsequent design and exploration.

We propose a conditional non-autoregressive neural sequence model based on iterative refinement. The proposed model is designed based on the principles of latent variable models and denoising autoencoders, and is generally applicable to any sequence generation task. We extensively evaluate the proposed model on machine translation (En-De and En-Ro) and image caption generation, and observe that it significantly speeds up decoding while maintaining the generation quality comparable to the autoregressive counterpart.

The rise of artificial intelligence will help improve productivity and lead to equitable development, said Prime Minister Narendra Modi at the University of Mumbai on Sunday. Modi, who inaugurated the Wadhwani Institute for Artificial Intelligence on the Kalina campus of the university, downplayed fears of humans losing jobs to robots. "With each wave of new technology, new opportunities arise. It opens an entirely new paradigm of opportunities. New opportunities have always outnumbered old ones," said Modi. "This optimism spells from my firm faith in the ancient Indian thinking that blended science and spirituality and found harmony between the two for the greater good of mankind," he said.

Artificial intelligence (AI), by definition, is somehow made, rather than innate. But how do you make an intelligence? It is a billion, trillion dollar question in the real world, and we don't appear all that close to finding out the answer. But science fiction authors have lots of ideas about how to create intelligence and give it personality. Why does an AI need a personality?

The Fairbanks Daily News-Miner reports the children in attendance Saturday at the Fairbanks Children's Museum participated in dusting for fingerprints, handcuffing, evidence searches and using a facial recognition and sketching program for the Crime Busters program.

This article originally appeared in the Motley Fool. NVIDIA (NASDAQ:NVDA) stock has been a huge winner in recent years. Since 2016, shares have returned 640%, versus the S&P 500's nearly 40% return. This exceptional performance has left investors hungry for articles about the fast-growing graphics processing unit (GPU) specialist and artificial intelligence (AI) player, which in turn means that financial writers are coming out of the woodwork to meet this demand. If said woodwork was filled only with folks who were both capable of grasping meaty tech topics and willing to do adequate homework before gracing the digital world with their words and analysis, this en masse emergence wouldn't be an issue.

As healthcare organizations take on more accountability through value-based contracting, they are making business and clinical intelligence solutions an integral part of their population health strategy. These solutions have significantly advanced in recent years and healthcare organizations are using them to better inform decisions and drive actions for success in managing patient populations. Four new developments in analytics solutions strengthen the tools available to support providers in effectively managing the health of their patients. The evolution of algorithms and models that analyze healthcare indicators and system-specific patterns enables organizations to more precisely identify at-risk and rising-risk patients within their population. This helps organizations optimize care interventions with patients that have been identified with the potential to encounter elevated healthcare needs.

We have all heard the dire predictions about robots coming to steal our jobs. Some would even have us believe these silicon bogeymen are coming to kill us. It plays straight into people's darkest fears about technology. When futurists talk about things that haven't happened yet, they are free to parade educated guesses as fact. In a recent article, the MIT Technology Review tabulated the results of "every study we could find on what automation will do to jobs".

Time series are one of the most common data types encountered in daily life. Financial prices, weather, home energy usage, and even weight are all examples of data that can be collected at regular intervals. Almost every data scientist will encounter time series in their daily work and learning how to model them is an important skill in the data science toolbox. One powerful yet simple method for analyzing and predicting periodic data is the additive model. The idea is straightforward: represent a time-series as a combination of patterns at different scales such as daily, weekly, seasonally, and yearly, along with an overall trend.