Learning how to act when there are many available actions in each state is a challenging task for Reinforcement Learning (RL) agents, especially when many of the actions are redundant or irrelevant. In such cases, it is sometimes easier to learn which actions not to take. In this work, we propose the Action-Elimination Deep Q-Network (AE-DQN) architecture that combines a Deep RL algorithm with an Action Elimination Network (AEN) that eliminates sub-optimal actions. The AEN is trained to predict invalid actions, supervised by an external elimination signal provided by the environment. Simulations demonstrate a considerable speedup and added robustness over vanilla DQN in text-based games with over a thousand discrete actions.

Abstract: Wide reaching and continuously evolving value propositions are the gears of network orchestrator's business models (NOBMs). Digital transformation enters when reliable data and meaningful information became digital enabler (DE) fuel of NOBMs. Moreover, Machine Learning (ML) capabilities can work as a catalyzer to increase knowledge rate acquisition for business processes or economical activities. This paper sets up DE and ML example binds for 4 different industries, proposing that high-quality data obtained from a rich context augments the profitability of the model. Finally, we conclude that the highly variable context from México provides an ideal environment in which ML augments harmonization between DE and NOBMs.

I intend to develop an expert system in energy sector. I have enough expertise in this domanin and I'm lookin for a shell to build such project and,

Approximate computing is being considered as a promising design paradigm to overcome the energy and performance challenges in computationally demanding applications. If the case where the accuracy can be configured, the quality level versus energy efficiency or delay also may be traded-off. For this technique to be used, one needs to make sure a satisfactory user experience. This requires employing error predictors to detect unacceptable approximation errors. In this work, we propose a scheduling-aware feature selection method which leverages the intermediate results of the hardware accelerator to improve the prediction accuracy. Additionally, it configures the error predictors according to the energy consumption and latency of the system. The approach enjoys the flexibility of the prediction time for a higher accuracy. The results on various benchmarks demonstrate significant improvements in the prediction accuracy compared to the prior works which used only the accelerator inputs for the prediction.

Many applications in different domains produce large amount of time series data. Making accurate forecasting is critical for many decision makers. Various time series forecasting methods exist which use linear and nonlinear models separately or combination of both. Studies show that combining of linear and nonlinear models can be effective to improve forecasting performance. However, some assumptions that those existing methods make, might restrict their performance in certain situations. We provide a new Autoregressive Integrated Moving Average (ARIMA)-Artificial Neural Network(ANN) hybrid method that work in a more general framework. Experimental results show that strategies for decomposing the original data and for combining linear and nonlinear models throughout the hybridization process are key factors in the forecasting performance of the methods. By using appropriate strategies, our hybrid method can be an effective way to improve forecasting accuracy obtained by traditional hybrid methods and also either of the individual methods used separately.

AI could have a key role to play in climate change after the technology was used by scientists to identify greenhouse gas leaks in oil and gas wells. Research conducted at the University of Vermont used machine learning algorithms to predict whether the wells would emit significant amounts of methane – one of the most harmful gases contributing to global warming. It tested 38,391 wells in Alberta, Canada, and was able to determine which wells leaked – and those that didn't – with up to 87% accuracy. Professor George Pinder, who conducted the research alongside former doctoral student James Montague, said: "The big picture is that we can now have tool that could help us much more efficiently identify leaking wells. "Given that methane is such a significant contributor to global warming, this is powerful information that should be put to use." The analysis yielded a cluster of 16 traits that predicted whether a well would fail and leak. Researchers were given access to more complete information, including the fluid properties of the oil or natural gas being mined, for 4,000 wells. For these wells, the machine learning algorithm identified leaks with 87% accuracy. For a larger sample of about 28,500 wells, where the fluid property was not known and taken into account, the accuracy level was 62%. Companies in Alberta are required to test wells at the time they begin operating to determine if they have failed and are leaking methane. They must also keep careful records of each well's construction characteristics. Professor Anthony R Ingraffea – based at Cornell University's School of Civil and Environmental Engineering, in Ithaca, New York – is an expert in oil and natural gas well design and construction, but was not involved in the study. He said: "Provincial and state regulatory agencies never have enough inspectors or financial resources to locate, let alone repair, leaking wells.

Deep learning has been used to identify 1.47 million solar installations across the United States, exceeding the latest estimate of 1.02 million. What's new: Solar panels are becoming increasingly popular across the US, but it's proved difficult to pinpoint their exact number. Researchers from Stanford University have got us much closer, thanks to a new system called DeepSolar, which uses deep learning to scan satellite images for solar panels. How it worked: The team trained DeepSolar on 370,000 satellite images by teaching it which ones included solar panels. The program then worked out how to spot solar panels, finding them correctly 93% of the time. It took about a month for the system to scan the billion images needed to reach its final figure.

The reduced-price replacements last until the end of the year, at which point the cost will dramatically increase. For the moment, a new battery costs only £25 – but once the new year arrives, that will rocket up to as much as £65. Old batteries can cause significant problems for their owners as iPhones age. With use, the power begins to drop – something that can lead to phones lasting for much less time, and to Apple having to slow down phones to ensure that they don't crash because they're not getting enough power. It was the revelation that Apple was doing that – throttling performance on older phones, in line with more spectacular rumours that swirled before it was admitted – that led to the cheap repairs in the first place.

What were the main developments in AI, Machine Learning, Analytics & Data Science in 2018 and what key trends do you expect in 2019? For the industry's take on what happened this year and what will happen next, we have collected insights from Domino Data Lab, dotData, Figure Eight, GoodData, KNIME, MapR, MathWorks, OpenText, ParallelM, Salesforce, Splice Machine, Splunk, and Zoomdata. Key themes singled out by these experts include the changing analytics landscape, how data science will continue to influence business, and the emerging technologies that will be leveraged to do so. Be sure to check out collected opinions we shared last week when we asked a group of experts the related question, "What were the main developments in Data Science and Analytics in 2018 and what key trends do you expect in 2019?" Josh Poduska is Chief Data Scientist at Domino Data Lab. The honeymoon is officially over for Artificial Intelligence.