The gravity fed water distribution network design (WDND) optimization problem consists in determining the pipe diameters of a water network such that hydraulic constraints are satisfied and the total cost is minimized. Traditionally, such design decisions are made on the basis of expert experience. When networks increase in size, however, rules of thumb will rarely lead to near optimal decisions. Over the past thirty years, a large number of techniques have been developed to tackle the problem of optimally designing a water distribution network. In this paper, we tackle the NP-hard water distribution network design (WDND) optimization problem in a multi-period setting where time varying demand patterns occur. We propose a new simulation-based iterated local search metaheuristic which further explores the structure of the problem in an attempt to obtain high quality solutions. Computational experiments show that our approach is very competitive as it is able to improve over a state-of-the-art metaheuristic for most of the performed tests. Furthermore, it converges much faster to low cost solutions and demonstrates a more robust performance in that it obtains smaller deviations from the best known solutions.

Efforts to improve the learning abilities of neural networks have focused mostly on the role of optimization methods rather than on weight initializations. Recent findings, however, suggest that neural networks rely on lucky random initial weights of subnetworks called "lottery tickets" that converge quickly to a solution. To investigate how weight initializations affect performance, we examine small convolutional networks that are trained to predict n steps of the two-dimensional cellular automaton Conway's Game of Life, the update rules of which can be implemented efficiently in a 2n+1 layer convolutional network. We find that networks of this architecture trained on this task rarely converge. Rather, networks require substantially more parameters to consistently converge. In addition, near-minimal architectures are sensitive to tiny changes in parameters: changing the sign of a single weight can cause the network to fail to learn. Finally, we observe a critical value d_0 such that training minimal networks with examples in which cells are alive with probability d_0 dramatically increases the chance of convergence to a solution. We conclude that training convolutional neural networks to learn the input/output function represented by n steps of Game of Life exhibits many characteristics predicted by the lottery ticket hypothesis, namely, that the size of the networks required to learn this function are often significantly larger than the minimal network required to implement the function.

Multivariate time series forecasting is of great importance to many scientific disciplines and industrial sectors. The evolution of a multivariate time series depends on the dynamics of its variables and the connectivity network of causal interrelationships among them. Most of the existing time series models do not account for the causal effects among the system's variables and even if they do they rely just on determining the between-variables causality network. Knowing the structure of such a complex network and even more specifically knowing the exact lagged variables that contribute to the underlying process is crucial for the task of multivariate time series forecasting. The latter is a rather unexplored source of information to leverage. In this direction, here a novel neural network-based architecture is proposed, termed LAgged VAriable Representation NETwork (LAVARNET), which intrinsically estimates the importance of lagged variables and combines high dimensional latent representations of them to predict future values of time series. Our model is compared with other baseline and state of the art neural network architectures on one simulated data set and four real data sets from meteorology, music, solar activity, and finance areas. The proposed architecture outperforms the competitive architectures in most of the experiments.

Graph Neural Networks (GNNs) are powerful and flexible neural networks that use the naturally sparse connectivity information of the data. GNNs represent this connectivity as sparse matrices, which have lower arithmetic intensity and thus higher communication costs compared to dense matrices, making GNNs harder to scale to high concurrencies than convolutional or fully-connected neural networks. We introduce a family of parallel algorithms for training GNNs and show that they can asymptotically reduce communication compared to previous parallel GNN training methods. We implement these algorithms, which are based on 1D, 1.5D, 2D, and 3D sparse-dense matrix multiplication, using torch.distributed on GPU-equipped clusters. Our algorithms optimize communication across the full GNN training pipeline. We train GNNs on over a hundred GPUs on multiple datasets, including a protein network with over a billion edges.

The next time you need to fill up your car, you might pay for your gas using your voice. Amazon, ExxonMobil and payment provider Fiserv have announced that you can pay with Alexa at some pumps. The Alexa skill was announced at CES in January, and it's now available at more than 11,500 Exxon and Mobil stations across the US. When you pull up to a pump, you can simply ask the voice assistant to "pay for gas" using your phone or through your car itself if you have an Alexa-enabled vehicle or Echo Auto. Alexa will confirm the station and pump number with you, then it'll activate the pump.

But the kind of power generation that Edison pioneered on that September day in 1882 put us on a trajectory that has had unfortunate outcomes. He kicked into overdrive our reliance on fossil fuels for energy, allowing it to permeate all aspects of our lives--from the electricity we need to power our homes, offices and factories, to the petroleum we need to run our cars, ships and planes. This forced us down a path of high-energy consumption that has resulted in the rapid depletion of naturally occurring carbon-based fuel sources and inflicted near-irreversible damage on our planet. Edison's choice of coal as the fuel source for his power plant should not be taken as indicative of his support for fossil fuels as a source of energy. At least in the context of transportation, he believed that automobiles should run on electricity--not petrol--and even built a vehicle powered by alkaline batteries of his own invention.

Artificial intelligence is about to trigger explosive changes in our lives, work, and leisure, but few understand what the technology can do beyond Amazon AMZN 2%'s Alexa or Apple AAPL 3.4%'s Siri. These are examples of virtual assistant or'weak AI' technology -- the most common example of AI application. But in the data-driven energy sector, sophisticated machine learning is paving the way for'strong AI' to improve efficiency, forecasting, trading, and user accessibility. Electricity is a commodity that can be bought, sold, and traded in open markets. For these markets to function efficiently, massive amounts of data -- from weather forecasting to grid demand/supply balance -- must be constantly analyzed by power sellers, buyers, and brokers.

In 2018, a wildfire caused by a faulty electric transmission line owned by Pacific Gas & Electric (PG&E) tore through Northern California's Butte County, killing 85 people and destroying nearly 19,000 buildings. In the wake of the fire, PG&E, which serves 5.2 million households in Northern California, set out to develop an AI technology suite that leverages computer vision to help it identify high fire-risk areas. Dubbed Sherlock Suite, the solution has helped PG&E automate inspections of its field equipment. Get the insights by signing up for our newsletters. "The Sherlock Suite allows desktop inspectors to mark up potential equipment problems on high-resolution images, further training computer-vision models to automatically detect potential issues and adding metadata to enable searchability of these images across the enterprise," says Kunal Datta, product manager for the Sherlock Suite at PG&E.

The robot possesses an infrared thermal imager and a visual light camera, thereby giving them the ability to replace 24-hour manual inspection. Artificial intelligence is about to trigger explosive changes in our lives, work, and leisure, but few understand what the technology can do beyond Amazon AMZN's Alexa or Apple AAPL's Siri. These are examples of virtual assistant or'weak AI' technology -- the most common example of AI application. But in the data-driven energy sector, sophisticated machine learning is paving the way for'strong AI' to improve efficiency, forecasting, trading, and user accessibility. Electricity is a commodity that can be bought, sold, and traded in open markets.

LSWI Optimization Methodology. A software tool that combines advanced statistical analysis, machine learning, artificial intelligence, and data- …