Connect with us

TOP SCEINCE

Training algorithm breaks barriers to deep physical neural networks

Published

on

Training algorithm breaks barriers to deep physical neural networks


EPFL researchers have developed an algorithm to train an analog neural network just as accurately as a digital one, enabling the development of more efficient alternatives to power-hungry deep learning hardware.

With their ability to process vast amounts of data through algorithmic ‘learning’ rather than traditional programming, it often seems like the potential of deep neural networks like Chat-GPT is limitless. But as the scope and impact of these systems have grown, so have their size, complexity, and energy consumption — the latter of which is significant enough to raise concerns about contributions to global carbon emissions.

And while we often think of technological advancement in terms of shifting from analog to digital, researchers are now looking for answers to this problem in physical alternatives to digital deep neural networks. One such researcher is Romain Fleury of EPFL’s Laboratory of Wave Engineering in the School of Engineering. In a paper published in Science, he and his colleagues describe an algorithm for training physical systems that shows improved speed, enhanced robustness, and reduced power consumption compared to other methods.

“We successfully tested our training algorithm on three wave-based physical systems that use sound waves, light waves, and microwaves to carry information, rather than electrons. But our versatile approach can be used to train any physical system,” says first author and LWE researcher Ali Momeni.

A “more biologically plausible” approach

Neural network training refers to helping systems learn to generate optimal values of parameters for a task like image or speech recognition. It traditionally involves two steps: a forward pass, where data is sent through the network and an error function is calculated based on the output; and a backward pass (also known as backpropagation, or BP), where a gradient of the error function with respect to all network parameters is calculated.

Over repeated iterations, the system updates itself based on these two calculations to return increasingly accurate values. The problem? In addition to being very energy-intensive, BP is poorly suited to physical systems. In fact, training physical systems usually requires a digital twin for the BP step, which is inefficient and carries the risk of a reality-simulation mismatch.

The scientists’ idea was to replace the BP step with a second forward pass through the physical system to update each network layer locally. In addition to decreasing power use and eliminating the need for a digital twin, this method better reflects human learning.

“The structure of neural networks is inspired by the brain, but it is unlikely that the brain learns via BP,” explains Momeni. “The idea here is that if we train each physical layer locally, we can use our actual physical system instead of first building a digital model of it. We have therefore developed an approach that is more biologically plausible.”

The EPFL researchers, with Philipp del Hougne of CNRS IETR and Babak Rahmani of Microsoft Research, used their physical local learning algorithm (PhyLL) to train experimental acoustic and microwave systems and a modeled optical system to classify data like vowel sounds and images. As well as showing comparable accuracy to BP-based training, the method was robust and adaptable — even in systems exposed to unpredictable external perturbations — compared to the state of the art.

An analog future?

While the LWE’s approach is the first BP-free training of deep physical neural networks, some digital updates of the parameters are still required. “It’s a hybrid training approach, but our aim is to decrease digital computation as much as possible,” Momeni says.

The researchers now hope to implement their algorithm on a small-scale optical system, with the ultimate goal of increasing network scalability.

“In our experiments, we used neural networks with up to 10 layers, but would it still work with 100 layers with billions of parameters? This is the next step, and will require overcoming technical limitations of physical systems.”



Source link

Continue Reading
Click to comment

Leave a Reply

TOP SCEINCE

Researchers create artificial cells that act like living cells

Published

on

By

Researchers create artificial cells that act like living cells


In a new study published in Nature Chemistry, UNC-Chapel Hill researcher Ronit Freeman and her colleagues describe the steps they took to manipulate DNA and proteins — essential building blocks of life — to create cells that look and act like cells from the body. This accomplishment, a first in the field, has implications for efforts in regenerative medicine, drug delivery systems, and diagnostic tools.

“With this discovery, we can think of engineering fabrics or tissues that can be sensitive to changes in their environment and behave in dynamic ways,” says Freeman, whose lab is in the Applied Physical Sciences Department of the UNC College of Arts and Sciences.

Cells and tissues are made of proteins that come together to perform tasks and make structures. Proteins are essential for forming the framework of a cell, called the cytoskeleton. Without it, cells wouldn’t be able to function. The cytoskeleton allows cells to be flexible, both in shape and in response to their environment.

Without using natural proteins, the Freeman Lab built cells with functional cytoskeletons that can change shape and react to their surroundings. To do this, they used a new programmable peptide-DNA technology that directs peptides, the building blocks of proteins, and repurposed genetic material to work together to form a cytoskeleton.

“DNA does not normally appear in a cytoskeleton,” Freeman says. “We reprogrammed sequences of DNA so that it acts as an architectural material, binding the peptides together. Once this programmed material was placed in a droplet of water, the structures took shape.”

The ability to program DNA in this way means scientists can create cells to serve specific functions and even fine-tune a cell’s response to external stressors. While living cells are more complex than the synthetic ones created by the Freeman Lab, they are also more unpredictable and more susceptible to hostile environments, like severe temperatures.

“The synthetic cells were stable even at 122 degrees Fahrenheit, opening up the possibility of manufacturing cells with extraordinary capabilities in environments normally unsuitable to human life,” Freeman says.

Instead of creating materials that are made to last, Freeman says their materials are made to task — perform a specific function and then modify themselves to serve a new function. Their application can be customized by adding different peptide or DNA designs to program cells in materials like fabrics or tissues. These new materials can integrate with other synthetic cell technologies, all with potential applications that could revolutionize fields like biotechnology and medicine.

“This research helps us understand what makes life,” Freeman says. “This synthetic cell technology will not just enable us to reproduce what nature does, but also make materials that surpass biology.”



Source link

Continue Reading

TOP SCEINCE

This alloy is kinky

Published

on

By

This alloy is kinky


Researchers have uncovered a remarkable metal alloy that won’t crack at extreme temperatures due to kinking, or bending, of crystals in the alloy at the atomic level.  A metal alloy composed of niobium, tantalum, titanium, and hafnium has shocked materials scientists with its impressive strength and toughness at both extremely hot and cold temperatures, a combination of properties that seemed so far to be nearly impossible to achieve. In this context, strength is defined as how much force a material can withstand before it is permanently deformed from its original shape, and toughness is its resistance to fracturing (cracking). The alloy’s resilience to bending and fracture across an enormous range of conditions could open the door for a novel class of materials for next-generation engines that can operate at higher efficiencies.

The team, led by Robert Ritchie at Lawrence Berkeley National Laboratory (Berkeley Lab) and UC Berkeley, in collaboration with the groups led by professors Diran Apelian at UC Irvine and Enrique Lavernia at Texas A&M University, discovered the alloy’s surprising properties and then figured out how they arise from interactions in the atomic structure. Their work is described in a study that was published April 11, 2024 in Science.

“The efficiency of converting heat to electricity or thrust is determined by the temperature at which fuel is burned — the hotter, the better. However, the operating temperature is limited by the structural materials which must withstand it,” said first author David Cook, a Ph.D. student in Ritchie’s lab. “We have exhausted the ability to further optimize the materials we currently use at high temperatures, and there’s a big need for novel metallic materials. That’s what this alloy shows promise in.”

The alloy in this study is from a new class of metals known as refractory high or medium entropy alloys (RHEAs/RMEAs). Most of the metals we see in commercial or industrial applications are alloys made of one main metal mixed with small quantities of other elements, but RHEAs and RMEAs are made by mixing near-equal quantities of metallic elements with very high melting temperatures, which gives them unique properties that scientists are still unraveling. Ritchie’s group has been investigating these alloys for several years because of their potential for high-temperature applications.

“Our team has done previous work on RHEAs and RMEAs and we have found that these materials are very strong, but generally possess extremely low fracture toughness, which is why we were shocked when this alloy displayed exceptionally high toughness,” said co-corresponding author Punit Kumar, a postdoctoral researcher in the group.

According to Cook, most RMEAs have a fracture toughness less than 10 MPa√m, which makes them some of the most brittle metals on record. The best cryogenic steels, specially engineered to resist fracture, are about 20 times tougher than these materials. Yet the niobium, tantalum, titanium, and hafnium (Nb45Ta25Ti15Hf15) RMEA alloy was able to beat even the cryogenic steel, clocking in at over 25 times tougher than typical RMEAs at room temperature.

But engines don’t operate at room temperature. The scientists evaluated strength and toughness at five temperatures total: -196°C (the temperature of liquid nitrogen), 25°C (room temperature), 800°C, 950°C, and 1200°C. The last temperature is about 1/5 the surface temperature of the sun.

The team found that the alloy had the highest strength in the cold and became slightly weaker as the temperature rose, but still boasted impressive figures throughout the wide range. The fracture toughness, which is calculated from how much force it takes to propagate an existing crack in a material, was high at all temperatures.

Unraveling the atomic arrangements

Almost all metallic alloys are crystalline, meaning that the atoms inside the material are arranged in repeating units. However, no crystal is perfect, they all contain defects. The most prominent defect that moves is called the dislocation, which is an unfinished plane of atoms in the crystal. When force is applied to a metal it causes many dislocations to move to accommodate the shape change. For example, when you bend a paper clip which is made of aluminum, the movement of dislocations inside the paper clip accommodates the shape change. However, the movement of dislocations becomes more difficult at lower temperatures and as a result many materials become brittle at low temperatures because dislocations cannot move. This is why the steel hull of the Titanic fractured when it hit an iceberg. Elements with high melting temperatures and their alloys take this to the extreme, with many remaining brittle up to even 800°C. However, this RMEA bucks the trend, withstanding snapping even at temperatures as low as liquid nitrogen (-196°C).

To understand what was happening inside the remarkable metal, co-investigator Andrew Minor and his team analyzed the stressed samples, alongside unbent and uncracked control samples, using four-dimensional scanning transmission electron microscopy (4D-STEM) and scanning transmission electron microscopy (STEM) at the National Center for Electron Microscopy, part of Berkeley Lab’s Molecular Foundry.

The electron microscopy data revealed that the alloy’s unusual toughness comes from an unexpected side effect of a rare defect called a kink band. Kink bands form in a crystal when an applied force causes strips of the crystal to collapse on themselves and abruptly bend. The direction in which the crystal bends in these strips increases the force that dislocations feel, causing them to move more easily. On the bulk level, this phenomenon causes the material to soften (meaning that less force has to be applied to the material as it is deformed). The team knew from past research that kink bands formed easily in RMEAs, but assumed that the softening effect would make the material less tough by making it easier for a crack to spread through the lattice. But in reality, this is not the case.

“We show, for the first time, that in the presence of a sharp crack between atoms, kink bands actually resist the propagation of a crack by distributing damage away from it, preventing fracture and leading to extraordinarily high fracture toughness,” said Cook.

The Nb45Ta25Ti15Hf15 alloy will need to undergo a lot more fundamental research and engineering testing before anything like a jet plane turbine or SpaceX rocket nozzle is made from it, said Ritchie, because mechanical engineers rightfully require a deep understanding of how their materials perform before they use them in the real world. However, this study indicates that the metal has potential to build the engines of the future.

This research was conducted by David H. Cook, Punit Kumar, Madelyn I. Payne, Calvin H. Belcher, Pedro Borges, Wenqing Wang, Flynn Walsh, Zehao Li, Arun Devaraj, Mingwei Zhang, Mark Asta, Andrew M. Minor, Enrique J. Lavernia, Diran Apelian, and Robert O. Ritchie, scientists at Berkeley Lab, UC Berkeley, Pacific Northwest National Laboratory, and UC Irvine, with funding from the Department of Energy (DOE) Office of Science. Experimental and computational analysis was conducted at the Molecular Foundry and the National Energy Research Scientific Computing Center — both are DOE Office of Science user facilities.



Source link

Continue Reading

TOP SCEINCE

3 in 5 parents play short order cook for young children who don’t like family meal

Published

on

By

3 in 5 parents play short order cook for young children who don’t like family meal


While most parents of preschool and elementary aged children strive to give their children a balanced, nutritional diet, some of their strategies to promote healthy eating may backfire, a national poll suggests.

One in eight parents require children to eat everything on their plate, according to the University of Michigan Health C.S. Mott Children’s Hospital National Poll on Children’s Health. And while just one in three believe the standard American diet is healthy for kids, few have tried alternative, potentially more nutritional menus at home.

“Feeding young children can be difficult due to general pickiness, hesitancy to try unfamiliar foods and constantly evolving food preferences,” said Mott Poll co-director and Mott pediatrician Susan Woolford, M.D.

“The preschool and elementary age is an important time to establish healthy eating patterns. Yet parents’ concern about whether their child is eating enough or if they’re getting the nutrients they need may lead them to adopt practices that actually sabotage their efforts to get kids to have healthy eating habits in the short and long term.”

The nationally representative report is based on 1,083 responses of parents of children ages 3-10 surveyed in February.

More on poll findings:

Parents’ beliefs on nutritional diets vary

Just a third of parents think the standard American diet is healthy compared to half who seem to rank the Mediterranean higher in nutritional value. Still, few have tried alternative diets for their child.

“Parents may recognize the standard diet in the U.S. includes high amounts of saturated fats, added sugars, sodium, and refined carbohydrates, which can generate an excess intake of calories beyond nutritional needs and contribute to health problems,” Woolford said.

“However, despite this recognition and evidence suggesting that other diet options may help avoid many illnesses, only about 9% have tried the Mediterranean diet for their children and fewer have tried giving their children a vegetarian diet.”

Parents should ensure children are still getting adequate nutrition if they do try diets that eliminate certain food categories, she adds. Diets that limit animal products, for example, will require alternative protein sources such as meat substitutes, tofu, or legumes for children.

And while ketogenic diets have become popular among adults, they are generally not appropriate for children.

Family dining rules may promote or hinder a child’s healthy diet

Fifteen percent of parents say their family rule is that kids finish what’s on their plate, while more than half say children must try some of everything and a little less than a third say no to dessert if meals go unfinished.

But parents who try to force kids to eat may encourage portions that go beyond feeling full, Woolford cautions.

“Requiring children to eat everything on their plate, or withholding dessert unless all other foods are eaten, can lead to overconsumption, especially if portion sizes are too large for the child’s age,” she said.

She agrees with the recommendation that “parents provide, and the child decides.” This makes parents responsible for providing healthy options while allowing children to select which foods they will eat and the amount they want to consume.

Parents often play personal chef

Sixty percent of parents will make something separate if their child doesn’t like the food that’s on the dinner table — and this often leads to a less healthy alternative, Woolford says.

“Rather than allowing the child to choose an alternate menu, parents should provide a balanced meal with at least one option that their child is typically willing to eat,” she said.

“Then if their child chooses not to eat, parents should not worry as this will not cause healthy children any harm and they will be more likely to eat the options presented at the next meal.”

She points out that children learn through watching and imitating, so it’s beneficial for parents to model healthy eating through a well-balanced diet while their child’s eating habits and taste preferences mature.

Avoiding snacks between meals may also help children have a better appetite and increase willingness to eat offered foods.

Picky eating and protesting veggies among biggest battles

Parents describe their biggest challenges with making sure their child gets a healthy diet as the child being a picky eater, the higher cost of healthy food and food waste. Fewer say they don’t have time to prepare healthy food.

Nearly all parents polled report trying at least one strategy to get their child to eat vegetables as part of a healthy diet, such as serving vegetables every day, fixing vegetables how their child prefers, trying vegetables their child hasn’t had before and letting children pick out vegetables at the grocery store.

Others involve children with preparing the vegetables, hide vegetables in other foods or offer a reward for finishing vegetables.

“Unsurprisingly, parents said pickiness and getting kids to eat veggies were among major challenges during mealtimes,” Woolford said.

“Parents should try to include children in meal decisions, avoid pressuring food consumption and provide a variety of healthy options at each meal so kids feel more control.”

Right sizing food may be difficult

Portion size is key to mitigating the risk of childhood obesity, but it can be hard for parents to “right-size” a child portion.

In determining portion size for their child, nearly 70% of parents polled give their child slightly less than adults in the family while fewer let their child choose how much to take, use predetermined portions from the package or give their child the same portions as adults.

Woolford recommends parents seek sources to help. The U.S. Department of Agriculture, for example, provides a visual called “MyPlate” that can help parents estimate the recommended balance of the major food groups and offers guidance on estimating portion size.

Healthy eating starts at the grocery store

When grocery shopping or planning meals, parents polled say they try to limit the amount of certain foods to help their child to maintain a healthy diet, with more than half limiting foods with added sugars and processed foods.

But it may be difficult to identify unhealthy food. Added sugars or processing may be present in foods marketed or packaged as healthy, Woolford says.

Parents should read labels, avoiding the marketing on the front of packages and focusing instead on the details on the back. They should pay particular attention to nutrition information and ingredient lists — especially if they’re long with unrecognizable items — as well as sodium, added sugars, and fat.

Woolford also encourages involving children in grocery trips, spending time in the produce section and asking them what they may like to try.

“Have them help in the process of choosing the healthiest options, not ones that necessarily directly advertise to children, but foods that they are willing to try that are lower in sugar, fat and salt,” she said.

“Spend most of the time in the produce section and try to make it fun by maybe selecting new options from different parts of the world that they haven’t tried before.”



Source link

Continue Reading

Trending

Copyright © 2017 Zox News Theme. Theme by MVP Themes, powered by WordPress.