Connect with us

TOP SCEINCE

How to use AI for discovery — without leading science astray

Published

on

How to use AI for discovery — without leading science astray


Over the past decade, AI has permeated nearly every corner of science: Machine learning models have been used to predict protein structures, estimate the fraction of the Amazon rainforest that has been lost to deforestation and even classify faraway galaxies that might be home to exoplanets.

But while AI can be used to speed scientific discovery — helping researchers make predictions about phenomena that may be difficult or costly to study in the real world — it can also lead scientists astray. In the same way that chatbots sometimes “hallucinate,” or make things up, machine learning models can sometimes present misleading or downright false results.

In a paper published online today (Thursday, Nov. 9) in Science, researchers at the University of California, Berkeley, present a new statistical technique for safely using the predictions obtained from machine learning models to test scientific hypotheses.

The technique, called prediction-powered inference (PPI), uses a small amount of real-world data to correct the output of large, general models — such as AlphaFold, which predicts protein structures — in the context of specific scientific questions.

“These models are meant to be general: They can answer many questions, but we don’t know which questions they answer well and which questions they answer badly — and if you use them naively, without knowing which case you’re in, you can get bad answers,” said study author Michael Jordan, the Pehong Chen Distinguished Professor of electrical engineering and computer science and of statistics at UC Berkeley. “With PPI, you’re able to use the model, but correct for possible errors, even when you don’t know the nature of those errors at the outset.”

The risk of hidden biases

When scientists conduct experiments, they’re not just looking for a single answer — they want to obtain a range of plausible answers. This is done by calculating a “confidence interval,” which, in the simplest case, can be found by repeating an experiment many times and seeing how the results vary.


In most science studies, a confidence interval usually refers to a summary or combined statistic, not individual data points. Unfortunately, machine learning systems focus on individual data points, and thus do not provide scientists with the kinds of uncertainty assessments that they care about. For instance, AlphaFold predicts the structure of a single protein, but it doesn’t provide a notion of confidence for that structure, nor a way to obtain confidence intervals that refer to general properties of proteins.

Scientists may be tempted to use the predictions from AlphaFold as if they were data to compute classical confidence intervals, ignoring the fact that these predictions are not data. The problem with this approach is that machine learning systems have many hidden biases that can skew the results. These biases arise, in part, from the data on which they are trained, which are generally existing scientific research that may not have had the same focus as the current study.

“Indeed, in scientific problems, we’re often interested in phenomena which are at the edge between the known and the unknown,” Jordan said. “Very often, there aren’t much data from the past that are at that edge, and that makes generative AI models even more likely to ‘hallucinate,’ producing output that is unrealistic.”

Calculating valid confidence intervals

PPI allows scientists to incorporate the predictions from models like AlphaFold without making any assumptions about how the model was built or the data it was trained on. To do this, PPI requires a small amount of data that is unbiased, with respect to the specific hypothesis being investigated, paired with machine learning predictions corresponding to that data. By bringing these two sources of evidence together, PPI is able to form valid confidence intervals.

For example, the research team applied the PPI technique to algorithms that can pinpoint areas of deforestation in the Amazon using satellite imagery. These models were accurate, overall, when tested individually on regions in the forest; however, when these assessments were combined to estimate deforestation across the entire Amazon, the confidence intervals became highly skewed. This is likely because the model struggled to recognize certain newer patterns of deforestation.


With PPI, the team was able to correct for the bias in the confidence interval using a small number of human-labeled regions of deforestation.

The team also showed how the technique can be applied to a variety of other research, including questions about protein folding, galaxy classification, gene expression levels, counting plankton, and the relationship between income and private health insurance.

“There’s really no limit on the type of questions that this approach could be applied to,” Jordan said. “We think that PPI is a much-needed component of modern data-intensive, model-intensive and collaborative science.”

Additional co-authors include Anastasios N. Angelopoulos, Stephen Bates, Clara Fannjiang and Tijana Zrnic of UC Berkeley. This research was supported by the Office of Naval Research (N00014-21-1-2840) and the National Science Foundation.



Source link

Continue Reading
Click to comment

Leave a Reply

TOP SCEINCE

The unintended consequences of success against malaria

Published

on

By

How to use AI for discovery — without leading science astray


For decades, insecticide-treated bed nets and indoor insecticide spraying regimens have been important — and widely successful — treatments against mosquitoes that transmit malaria, a dangerous global disease. Yet these treatments also — for a time — suppressed undesirable household insects like bed bugs, cockroaches and flies.

Now, a new North Carolina State University study reviewing the academic literature on indoor pest control shows that as the household insects developed resistance to the insecticides targeting mosquitoes, the return of these bed bugs, cockroaches and flies into homes has led to community distrust and often abandonment of these treatments — and to rising rates of malaria.

In short, the bed nets and insecticide treatments that were so effective in preventing mosquito bites — and therefore malaria — are increasingly viewed as the causes of household pest resurgence.

“These insecticide-treated bed nets were not intended to kill household pests like bed bugs, but they were really good at it,” said Chris Hayes, an NC State Ph.D. student and co-corresponding author of a paper describing the work. “It’s what people really liked, but the insecticides are not working as effectively on household pests anymore.”

“Non-target effects are usually harmful, but in this case they were beneficial,” said Coby Schal, Blanton J. Whitmire Distinguished Professor of Entomology at NC State and co-corresponding author of the paper.

“The value to people wasn’t necessarily in reducing malaria, but was in killing other pests,” Hayes added. “There’s probably a link between use of these nets and widespread insecticide resistance in these house pests, at least in Africa.”

The researchers add that other factors — famine, war, the rural/city divide, and population displacement, for example — also could contribute to rising rates of malaria.

To produce the review, Hayes combed through the academic literature to find research on indoor pests like bed bugs, cockroaches and fleas, as well as papers on malaria, bed nets, pesticides and indoor pest control. The search yielded more than 1,200 papers, which, after an exhaustive review process, was whittled down to a final count of 28 peer-reviewed papers fulfilling the necessary criteria.

One paper — a 2022 survey of 1,000 households in Botswana — found that while 58% were most concerned with mosquitoes in homes, more than 40% were most concerned with cockroaches and flies.

Hayes said a recent paper — published after this NC State review was concluded — showed that people blamed the presence of bed bugs on bed nets.

“There is some evidence that people stop using bed nets when they don’t control pests,” Hayes said.

The researchers say that all hope is not lost, though.

“There are, ideally, two routes,” Schal said. “One would be a two-pronged approach with both mosquito treatment and a separate urban pest management treatment that targets pests. The other would be the discovery of new malaria-control tools that also target these household pests at the same time. For example, the bottom portion of a bed net could be a different chemistry that targets cockroaches and bed bugs.

“If you offer something in bed nets that suppresses pests, you might reduce the vilification of bed nets.”

The study appears in Proceedings of the Royal Society B. The review was supported in part by the Blanton J. Whitmire Endowment at NC State, and grants from the U.S. Department of Housing and Urban Development Healthy Homes program (NCHHU0053-19), the Department of the Army, U.S. Army Contracting Command, Aberdeen Proving Ground, Natick Contracting Division, Ft. Detrick, Maryland (W911QY1910011), and the Triangle Center for Evolutionary Medicine (257367).



Source link

Continue Reading

TOP SCEINCE

Drawing water from dry air

Published

on

By

How to use AI for discovery — without leading science astray


Earth’s atmosphere holds an ocean of water, enough liquid to fill Utah’s Great Salt Lake 800 times.

Extracting some of that moisture is seen as a potential way to provide clean drinking water to billions of people globally who face chronic shortages.

Existing technologies for atmospheric water harvesting (AWH) are saddled with numerous downsides associated with size, cost and efficiency. But new research from University of Utah engineering researchers has yielded insights that could improve efficiencies and bring the world one step closer to tapping the air as a culinary water source in arid places.

The study unveils the first-of-its-kind compact rapid cycling fuel-fired AWH device. This two-step prototype relies on adsorbent materials that draw water molecules out of non-humid air, then applies heat to release those molecules into liquid form, according to Sameer Rao, senior author of the study published Monday and an assistant professor of mechanical engineering.

“Hygroscopic materials intrinsically have affinity to water. They soak up water wherever you go. One of the best examples is the stuff inside diapers,” said Rao, who happens to be the father of an infant son. “We work with a specific type of hygroscopic material called a metal organic framework.”

Rao likened metal organic frameworks to Lego blocks, which can be rearranged to build all sorts of structures. It this case they are arranged to create a molecule ideal for gas separation.

“They can make it specific to adsorb water vapor from the air and nothing else. They’re really selective,” Rao said. Developed with graduate student Nathan Ortiz, the study’s lead author, this prototype uses aluminum fumarate that was fashioned into panels that collect the water as air is drawn through.

“The water molecules themselves get trapped on the surfaces of our material, and that’s a reversible process. And so instead of becoming ingrained into the material itself, it sits on the walls,” Ortiz said. “What’s special about these absorbent materials is they have just an immense amount of internal surface area. There’s so many sites for water molecules to get stuck.”

Just a gram of this material holds as much surface area as two football fields, according to Rao. So just a little material can capture a lot of water.

“All of this surface area is at the molecular scale,” Rao said. “And that’s awesome for us because we want to trap water vapor onto that surface area within the pores of this material.”

Funding for the research came from the DEVCOM Soldier Center, a program run by the Department of Defense to facilitate technology transfer that supports Army modernization. The Army’s interest in the project stems from the need to keep soldiers hydrated while operating in remote areas with few water sources.

“We specifically looked at this for defense applications so that soldiers have a small compact water generation unit and don’t need to lug around a large canteen filled with water,” Rao said. “This would literally produce water on demand.”

Rao and Ortiz have filed for a preliminary patent based on the technology, which addresses non-military needs as well.

“As we were designing the system, I think we also had perspective of the broader water problem. It’s not just a defense issue, it’s very much a civilian issue,” Rao said. “We think in terms water consumption of a household for drinking water per day. That’s about 15 to 20 liters per day.”

In this proof of concept, the prototype achieved its target of producing 5 liters of water per day per kilogram of adsorbent material. In a matter of three days in the field, this devise would outperform packing water, according to Ortiz.

In the device’s second step, the water is precipitated into liquid by applying heat using a standard-issue Army camping stove. This works because of the exothermic nature of its water collecting process.

“As it collects water, it’s releasing little bits of heat. And then to reverse that, we add heat,” Ortiz said. “We just put a flame right under here, anything to get this temperature up. And then as we increase the temperature, we rapidly release the water molecules. Once we have a really humid airstream, that makes condensation at ambient temperature much easier.”

Nascent technologies abound for atmospheric water harvesting, which is more easily accomplished when the air is humid, but none has resulted in equipment that can be put to practical use in arid environments. Ortiz believes his device can be the first, mainly because it is powered with energy-dense fuel like the white gasoline used in camping stoves.

The team decided against using photovoltaics.

“If you’re reliant on solar panels, you’re limited to daytime operation or you need batteries, which is just more weight. You keep stacking challenges. It just takes up so much space,” Ortiz said. “This technology is superior in arid conditions, while refrigeration is best in high humidity.”



Source link

Continue Reading

TOP SCEINCE

3D-printed microstructure forest facilitates solar steam generator desalination

Published

on

By

How to use AI for discovery — without leading science astray


Faced with the world’s impending freshwater scarcity, a team of researchers in Singapore turned to solar steam generators (SSGs), which are emerging as a promising device for seawater desalination. Desalination can be a costly, energy-intensive solution to water scarcity. This renewable-powered approach mimics the natural water cycle by using the sun’s energy to evaporate and isolate water. However, the technology is limited by the need to fabricate complex topologies to increase the surface area necessary to achieve high water evaporation efficiency.

To overcome this barrier, the team sought design inspiration from trees and harnessed the potential of 3D printing. In Applied Physics Reviews, the team presents a state-of-the-art technology for producing efficient SSGs for desalination and introduces a novel method for printing functional nanocomposites for multi-jet fusion (MJF).

“We created SSGs with exceptional photothermal performance and self-cleaning properties,” said Kun Zhou, a professor of mechanical engineering at Nanyang Technological University. “Using a treelike porous structure significantly enhances water evaporation rates and ensures continuous operation by preventing salt accumulation — its performance remains relatively stable even after prolonged testing.”

The physics behind their approach involves light-to-thermal energy conversion, where the SSGs absorb solar energy, convert it to heat, and evaporate the water/seawater. The SSG’s porous structure helps improve self-cleaning by removing accumulated salt to ensure sustained desalination performance.

“By using an effective photothermal fusing agent, MJF printing technology can rapidly create parts with intricate designs,” he said. “To improve the photothermal conversion efficiency of fusing agents and printed parts, we developed a novel type of fusing agent derived from metal-organic frameworks.”

Their SSGs were inspired by plant transpiration and are composed of miniature tree-shaped microstructures, forming an efficient, heat-distributing forest.

“Our bioinspired design increases the surface area of the SSG,” said Zhou. “Using a treelike design increases the surface area of the SSG, which enhances the water transport and boosts evaporation efficiency.”

One big surprise was the high rate of water evaporation observed in both simulated environments and field trials. The desalinated water consistently met standards for drinking water — even after a long-time test.

“This demonstrates the practicality and efficiency of our approach,” Zhou said. “And it can be quickly and easily mass-produced via MJF commercial printers.”

The team’s work shows significant potential for addressing freshwater scarcity.

“Our SSGs can be used in regions with limited access to freshwater to provide a sustainable and efficient desalination solution,” said Zhou. “Beyond desalination, it can be adapted for other applications that require efficient solar energy conversion and water purification.”



Source link

Continue Reading

Trending