Peter van Hengstum, associate professor in the Department of Marine and Coastal Environment Science at Texas A&M-Galveston, and colleagues have had their findings published in PNAS (Proceedings of the National Academy of Sciences).

Researchers generated a new environmental record from the Blackwood Sinkhole, which is flooded with 120 feet of groundwater without dissolved oxygen. This is important because it has pristinely preserved organic material for the last 3,000 years. Using core samples and radiocarbon dating, the team examined charcoal deposits from human fires thousands of years ago, indicating that the first settlers arrived in the Bahamas sooner than previously thought.

“The Bahamas were the last place colonized by people in the Caribbean region, and previous physical evidence indicated that it may have taken hundreds of years for indigenous people of the Bahamas — called the Lucayans — to move through the Bahamian archipelago that spans about 500 miles,” van Hengstum said.

While people were present in Florida more than 14,000 years ago at the end of the last ice age, he said, these people never crossed the Florida Straits to nearby Bahamian islands, only 50 to 65 miles away. Meanwhile, the Caribbean islands were populated by people migrating from South American northward. Van Hengstum said the oldest archaeological sites in the southernmost Bahamian archipelago from the Turks and Caicos Islands indicate human arrival likely by 700 A.D.

“But in the northern Bahamian Great Abaco Island, the earliest physical evidence of human occupation are skeletons preserved in sinkholes and blueholes,” he said. “These two skeletons from Abaco date from 1200 to 1300 A.D. Our new record of landscape disturbance from people indicates that slash-and-burn agriculture likely began around 830 A.D., meaning the Lucayans rapidly migrated through the Bahamian archipelago in likely a century, or spanning just a few human generations.”

The team’s other findings show how the Lucayans changed the new land.

When the Lucayans arrived, Great Abaco Island was mostly covered with pine and palm forests, and had a unique reptile-dominated ecosystem of giant tortoises and crocodiles. Increased deforestation and burning allowed pine trees to colonize and out-compete native palms and hardwoods.

Large land reptiles began to disappear after 1000 A.D. A significant increase in intense regional hurricane activity around 1500 AD is thought to have caused considerable damage to the new pine tree forests, as indicated by a decrease in pine pollen in the sediment core.

“The pollen record indicates that the pre-contact forest was not significantly impacted earlier in the record during known times when intense hurricane strike events were more frequent,” van Hengstum said. “In our current world where the intensity of the largest hurricanes is expected to increase over the coming decades, the current pine trees in the northern Bahamas may not be as resilient to environmental impacts of these changes in hurricane activity.”

The study was funded by the National Science Foundation.

Story Source:

Materials provided by Texas A&M University. Original written by Keith Randall. Note: Content may be edited for style and length.

Researchers at the NYU Tandon School of Engineering, who explored the machine-learning frameworks behind these tools, found that the answer is “not very.” In the paper “Subverting Privacy-Preserving GANs: Hiding Secrets in Sanitized Images,” presented last month at the 35th AAAI Conference on Artificial Intelligence, a team led by Siddharth Garg, Institute Associate Professor of electrical and computer engineering at NYU Tandon, explored whether private data could still be recovered from images that had been “sanitized” by such deep-learning discriminators as privacy protecting GANs (PP-GANs) and that had even passed empirical tests. The team, including lead author Kang Liu, a Ph.D. candidate, and Benjamin Tan, research assistant professor of electrical and computer engineering, found that PP-GAN designs can, in fact, be subverted to pass privacy checks, while still allowing secret information to be extracted from sanitized images.

Machine-learning-based privacy tools have broad applicability, potentially in any privacy sensitive domain, including removing location-relevant information from vehicular camera data, obfuscating the identity of a person who produced a handwriting sample, or removing barcodes from images. The design and training of GAN-based tools are outsourced to vendors because of the complexity involved.

“Many third-party tools for protecting the privacy of people who may show up on a surveillance or data-gathering camera use these PP-GANs to manipulate images,” said Garg. “Versions of these systems are designed to sanitize images of faces and other sensitive data so that only application-critical information is retained. While our adversarial PP-GAN passed all existing privacy checks, we found that it actually hid secret data pertaining to the sensitive attributes, even allowing for reconstruction of the original private image.”

The study provides background on PP-GANs and associated empirical privacy checks, formulates an attack scenario to ask if empirical privacy checks can be subverted, and outlines an approach for circumventing empirical privacy checks.

The team provides the first comprehensive security analysis of privacy-preserving GANs and demonstrate that existing privacy checks are inadequate to detect leakage of sensitive information.

Using a novel steganographic approach, they adversarially modify a state-of-the-art PP-GAN to hide a secret (the user ID), from purportedly sanitized face images.

They show that their proposed adversarial PP-GAN can successfully hide sensitive attributes in “sanitized” output images that pass privacy checks, with 100% secret recovery rate.

Noting that empirical metrics are dependent on discriminators’ learning capacities and training budgets, Garg and his collaborators argue that such privacy checks lack the necessary rigor for guaranteeing privacy.

“From a practical standpoint, our results sound a note of caution against the use of data sanitization tools, and specifically PP-GANs, designed by third parties,” explained Garg. “Our experimental results highlighted the insufficiency of existing DL-based privacy checks and the potential risks of using untrusted third-party PP-GAN tools.”

Story Source:

Materials provided by NYU Tandon School of Engineering. Note: Content may be edited for style and length.

But how does this intricate system ensure that more active parts of the brain receive enough nourishment versus less demanding areas? That’s a century-old problem in neuroscience that scientists at the University of California San Diego have helped answer in a newly published study.

Studying the brains of mice, a team of researchers led by Xiang Ji, David Kleinfeld and their colleagues has deciphered the question of brain energy consumption and blood vessel density through newly developed maps that detail brain wiring to a resolution finer than a millionth of a meter, or one-hundredth of the thickness of a human hair.

A result of work at the crossroads of biology and physics, the new maps provide novel insights into these “microvessels” and their various functions in blood supply chains. The techniques and technologies underlying the results are described March 2 in the journal Neuron.

“We developed an experimental and computational pipeline to label, image and reconstruct the microvascular system in whole mouse brains with unprecedented completeness and precision,” said Kleinfeld, a professor in the UC San Diego Department of Physics (Division of Physical Sciences) and Section of Neurobiology (Division of Biological Sciences). Kleinfeld says the effort was akin to reverse engineering nature. “This allowed Xiang to carry out sophisticated calculations that not just related brain energy use to vessel density, but also predicted a tipping point between the loss of brain capillaries and a sudden drop in brain health.”

Questions surrounding how blood vessels carry nourishment to active and less active regions were posed as a general issue in physiology as far back as 1920. By the 1980s, a technology known as autoradiography, the predecessor of modern-day positron emission tomography (PET), allowed scientists to measure the distribution of sugar metabolism across the mouse brain.

To fully grasp and solve the problem, Ji, Kleinfeld and their colleagues at the Howard Hughes Medical Institute’s Janelia Research Campus and UC San Diego Jacobs School of Engineering filled 99.9 percent of the vessels in the mouse brain — a count of nearly 6.5 million — with a dye-labeled gel. They then imaged the full extent of the brain with sub-micrometer precision. This resulted in fifteen trillion voxels, or individual volumetric elements, per brain, that were transformed into a digital vascular network that could be analyzed with the tools of data science.

With their new maps in hand, the researchers determined that the concentration of oxygen is roughly the same in every region of the brain. But they found that small blood vessels are the key components that compensate for varying energy requirements. For example, white matter tracts, which transfer nerve impulses across the two brain hemispheres and to the spinal cord, are regions of low energy needs. The researchers identified lower levels of blood vessels there. By contrast, brain regions that coordinate the perception of sound use three times more energy and, they discovered, were found with a much greater level of blood vessel density.

“In the era of increasing complexities being unraveled in biological systems, it is fascinating to observe the emergence of shared simple and quantitative design rules underlaying the seemingly complicated networks across mammalian brains,” said Ji, a graduate student in physics.

Up next, the researchers hope to drill down into the finer aspects of their new maps to determine the detailed patterns of blood flow into and out from the entire brain. They will also pursue the largely uncharted relationship between the brain and the immune system.

Authors on the paper include Xiang Ji, Tiago Ferreira, Beth Friedman, Rui Liu, Hannah Liechty, Erhan Bas, Jayaram Chandrashekar and David Kleinfeld.

Story Source:

Materials provided by University of California – San Diego. Original written by Mario Aguilera. Note: Content may be edited for style and length.

Every aspect of modern computing, from the smallest chip to the largest data center comes with a carbon price tag. For the better part of a century, the tech industry and the field of computation as a whole have focused on building smaller, faster, more powerful devices — but few have considered their overall environmental impact.

Researchers at the Harvard John A. Paulson School of Engineering and Applied Sciences (SEAS) are trying to change that.

“Over the next decade, the demand, number and types of devices is only going to grow,” said Udit Gupta, a PhD candidate in Computer Science at SEAS. “We want to know what impact that will have on the environment and how we, as a field, should be thinking about how we adopt more sustainable practices.”

Gupta, along with Gu-Yeon Wei, the Robert and Suzanne Case Professor of Electrical Engineering and Computer Science, and David Brooks, the Haley Family Professor of Computer Science, will present a paper on the environmental footprint of computing at the IEEE International Symposium on High-Performance Computer Architecture on March 3rd, 2021.

The SEAS research is part of a collaboration with Facebook, where Gupta is an intern, and Arizona State University.

The team not only explored every aspect of computing, from chip architecture to data center design, but also mapped the entire lifetime of a device, from manufacturing to recycling, to identify the stages where the most emissions occur.

The team found that most emissions related to modern mobile and data-center equipment come from hardware manufacturing and infrastructure.

“A lot of the focus has been on how we reduce the amount of energy used by computers, but we found that it’s also really important to think about the emissions from just building these processors,” said Brooks. “If manufacturing is really important to emissions, can we design better processors? Can we reduce the complexity of our devices so that manufacturing emissions are lower?”

Take chip design, for example.

Today’s chips are optimized for size, performance and battery life. The typical chip is about 100 square millimeters of silicon and houses billions of transistors. But at any given time, only a portion of that silicon is being used. In fact, if all the transistors were fired up at the same time, the device would exhaust its battery life and overheat. This so-called dark silicon improves a device’s performance and battery life but it’s wildly inefficient if you consider the carbon footprint that goes into manufacturing the chip.

“You have to ask yourself, what is the carbon impact of that added performance,” said Wei. “Dark silicon offers a boost in energy efficiency but what’s the cost in terms of manufacturing? Is there a way to design a smaller and smarter chip that uses all of the silicon available? That is a really intricate, interesting, and exciting problem.”

The same issues face data centers. Today, data centers, some of which span many millions of square feet, account for 1 percent of global energy consumption, a number that is expected to grow.

As cloud computing continues to grow, decisions about where to run applications — on a device or in a data center — are being made based on performance and battery life, not carbon footprint.

“We need to be asking what’s greener, running applications on the device or in a data center,” said Gupta. “These decisions must optimize for global carbon emissions by taking into account application characteristics, efficiency of each hardware device, and varying power grids over the day.”

The researchers are also challenging industry to look at the chemicals used in manufacturing.

Adding environmental impact to the parameters of computational design requires a massive cultural shift in every level of the field, from undergraduate CS students to CEOs.

To that end, Brooks has partnered with Embedded EthiCS, a Harvard program that embeds philosophers directly into computer science courses to teach students how to think through the ethical and social implications of their work. Brooks is including an Embedded EthiCS module on computational sustainability in COMPSCI 146: Computer Architecture this spring.

The researchers also hope to partner with faculty from Environmental Science and Engineering at SEAS and the Harvard University Center for the Environment to explore how to enact change at the policy level.

“The goal of this paper is to raise awareness of the carbon footprint associated with computing and to challenge the field to add carbon footprint to the list of metrics we consider when designing new processes, new computing systems, new hardware, and new ways to use devices. We need this to be a primary objective in the development of computing overall,” said Wei.

The paper was co-authored by Sylvia Lee, Jordan Tse, Hsien-Hsin S. Lee and Carole-Jean Wu from Facebook and Young Geun Kim from Arizona State University.

Connected and automated vehicles, which can interact vehicle to vehicle (V2V) and between vehicles and roadway infrastructure like traffic signals and stop signs (V2I), promise to save energy and improve safety. In a new study published in Transportation Research Part B, engineers from Michigan Technological University propose a modeling framework for V2V and V2I cooperative driving.

Cooperative driving helps cars and their drivers safely and efficiently navigate. The framework uses an eco-driving algorithm that prioritizes saving fuel and reducing emissions. The automated algorithm calculates location-based traffic control devices and roadway constraints using maps and geographic information. The research is led by Kuilin Zhang, associate professor of civil and environmental engineering and affiliated associate professor of computer science at Michigan Tech, along with Shuaidong Zhao ’18, now a senior quantitative analyst at National Grid.

For the past three years, Houghton, Michigan, has been home to roadside units installed on five of the city’s traffic signals that make V2I communication possible. Zhang conducted a simulation analysis using real traffic signal phasing and timing messages from the Ann Arbor connected vehicle test environment and plans to expand testing in the Houghton area.

“The whole idea of cooperative driving automation is that the signals in the intersection tell your car what’s happening ahead,” Zhang said. “The sensor at the intersection can benefit all connected vehicles passing through the intersection. The automated eco-driving algorithm improves the driving decisions of the connected and automated vehicles.”

The simulation results show that the cooperative automated eco-driving algorithm saves energy — 7% under light traffic and 23% under heavy traffic along the corridor.

“The stop and go, stop and go, it may use a lot of energy,” Zhang said. “The concept of eco-driving incorporates how the vehicle makes driving decisions using data not only from vehicles in front of it, but also with information given from a traffic signal.”

Zhang’s model pulls in high-definition (HD) maps, which use a connected vehicle’s hardware and software to provide down-to-the-centimeter accuracy in navigation. HD maps incorporate multiple types of environmental sensing: long-range radar, lidar, camera footage, short/medium-range radar and ultrasound.

Zhang said for autonomous driving, it’s important to know landmarks to control the car’s driving, as well as hill grades; using a hill to slow or accelerate a car can also increase energy savings. It’s easy to conserve energy on a straight highway; on busy arterial streets with traffic and stoplights, energy conservation isn’t so simple. On city streets, Zhang and Zhao’s online predictive connected and automated eco-driving model considers traffic control devices and road geometry constraints under light and heavy traffic conditions.

Story Source:

Materials provided by Michigan Technological University. Original written by Kelley Christensen. Note: Content may be edited for style and length.