The new platform technology, modeled after the brain, consists of a mesh of silver wires on a bed of electrodes.
The new platform technology, modeled after the brain, consists of a mesh of silver wires on a bed of electrodes.
Tyler O'Neal, Staff Editor NETWORKS

Experimental brain-like supercomputing system: Unlocking the potential of neuromorphic nanowire networks

Scientists and researchers have been inspired by the intricacies of the human brain in creating advanced computing systems. The California NanoSystems Institute at UCLA has been at the forefront of developing a brain-inspired platform technology for computation. Their experimental computing system has shown remarkable potential in learning and identifying handwritten numbers with an impressive accuracy of 93.4%. This breakthrough is attributed to a novel training algorithm that provides continuous real-time feedback to the system, surpassing conventional machine-learning approaches.

The Brain-Like System: A Tangled Network of Nanowires

The brain-like computing system developed by researchers at UCLA and the University of Sydney is composed of a network of nanowires made of silver and selenium. These nanoscale wires self-organize into a complex, tangled structure on top of an array of electrodes. Unlike traditional computers, where memory and processing modules are separate entities, this nanowire network physically reconfigures itself in response to stimuli, with memory distributed throughout its atomic structure. The connections between wires can form or break, similar to the behavior of synapses in the biological brain.

Training the Brain-Like System with Handwritten Numbers

To train and test the brain-like system, researchers utilized a dataset of handwritten numbers provided by the National Institute of Standards and Technology. The images of these numbers were communicated to the system pixel-by-pixel using pulses of electricity, with varying voltages representing light or dark pixels. The streamlined algorithm developed by the University of Sydney allowed the system to process multiple streams of data simultaneously and adapt dynamically, leveraging its brain-like capabilities.

Real-Time Feedback: A Key to Enhanced Learning

The groundbreaking aspect of this experimental system lies in the real-time feedback provided during the training process. Unlike conventional approaches where training is performed after processing a batch of data, the brain-like system received continuous information about its success at the task as it learned. This constant feedback loop proved to be highly effective, resulting in an accuracy of 93.4% in identifying handwritten numbers. In comparison, a conventional machine-learning approach achieved an accuracy of 91.4%.

The brain-like computing system has some unique features that distinguish it from other computing approaches. One of these features is the system's distributed memory, which stores past inputs within the nanowire network. This embedded memory enhances the system's learning capabilities, making it highly accurate in identifying handwritten numbers.

Compared to silicon-based artificial intelligence systems, the brain-like system has the potential to operate with significantly lower power consumption. It can perform tasks that current AI systems find difficult, such as analyzing patterns in weather, traffic, and other dynamic systems.

The research team employed a co-design approach, developing both the hardware and software simultaneously. This approach ensures optimal integration between the brain-like system and its custom algorithm, resulting in enhanced performance. The combination of brain-like memory and processing capabilities embedded in physical systems with continuous adaptation and learning opens up new possibilities for edge computing.

Edge computing processes complex data on the spot without relying on communication with remote servers, making it suitable for various applications in robotics, autonomous navigation, smart devices, health monitoring, and more.

The brain-like computing system is still in the development phase, but its potential impact on various industries is immense. Its ability to perform complex tasks with lower energy consumption makes it an attractive alternative to traditional AI systems. The neuromorphic nanowire networks can unlock new opportunities in fields such as robotics, autonomous vehicles, the Internet of Things, and multi-location sensor coordination.

The experimental brain-like computing system represents a significant advancement in the field of neuromorphic computing. By harnessing the unique properties of nanowire networks and integrating them with custom algorithms, researchers have shown the potential for creating highly efficient and adaptable computing systems. As this technology continues to evolve, we can expect to see further breakthroughs in AI, edge computing, and various other domains, transforming the way we process and analyze complex data.

The study's corresponding authors include James Gimzewski, a distinguished professor of chemistry and member of the California NanoSystems Institute at UCLA, Adam Stieg, a research scientist and associate director of CNSI, Zdenka Kuncic, a professor of physics at the University of Sydney, and Ruomin Zhu, a doctoral student at the University of Sydney and first author of the study. Other co-authors include Sam Lilak, Alon Loeffler, and Joseph Lizier, all contributing to the research at UCLA and the University of Sydney.

The research was supported by the University of Sydney and the Australian-American Fulbright Commission. The brain-like computing system has some unique features that distinguish it from other computing approaches. One of these features is the system's distributed memory, which stores past inputs within the nanowire network. This embedded memory enhances the system's learning capabilities, making it highly accurate in identifying handwritten numbers.

Compared to silicon-based artificial intelligence systems, the brain-like system has the potential to operate with significantly lower power consumption. It can perform tasks that current AI systems find difficult, such as analyzing patterns in weather, traffic, and other dynamic systems.

The research team employed a co-design approach, developing both the hardware and software simultaneously. This approach ensures optimal integration between the brain-like system and its custom algorithm, resulting in enhanced performance. The combination of brain-like memory and processing capabilities embedded in physical systems with continuous adaptation and learning opens up new possibilities for edge computing.

Edge computing processes complex data on the spot without relying on communication with remote servers, making it suitable for various applications in robotics, autonomous navigation, smart devices, health monitoring, and more.

The brain-like computing system is still in the development phase, but its potential impact on various industries is immense. Its ability to perform complex tasks with lower energy consumption makes it an attractive alternative to traditional AI systems. The neuromorphic nanowire networks can unlock new opportunities in fields such as robotics, autonomous vehicles, the Internet of Things, and multi-location sensor coordination.

The experimental brain-like computing system represents a significant advancement in the field of neuromorphic computing. By harnessing the unique properties of nanowire networks and integrating them with custom algorithms, researchers have shown the potential for creating highly efficient and adaptable computing systems. As this technology continues to evolve, we can expect to see further breakthroughs in AI, edge computing, and various other domains, transforming the way we process and analyze complex data.

The study's corresponding authors include James Gimzewski, a distinguished professor of chemistry and member of the California NanoSystems Institute at UCLA, Adam Stieg, a research scientist and associate director of CNSI, Zdenka Kuncic, a professor of physics at the University of Sydney, and Ruomin Zhu, a doctoral student at the University of Sydney and first author of the study. Other co-authors include Sam Lilak, Alon Loeffler, and Joseph Lizier, all contributing to the research at UCLA and the University of Sydney.

The research was supported by the University of Sydney and the Australian-American Fulbright Commission.

Assistant Professor Cancer Center Member Ph.D., Yale University, 2015
Assistant Professor Cancer Center Member Ph.D., Yale University, 2015

CSHL prof Koo builds EvoAug for improving the interpretability of genomic deep neural networks with evolution-inspired data augmentations

Tyler O'Neal, Staff Editor NETWORKS

Genes make up only a small fraction of the human genome. Between them are wide sequences of DNA that direct cells when, where, and how much each gene should be used. These biological instruction manuals are known as regulatory motifs. If that sounds complex, well, it is.

The instructions for gene regulation are written in a complicated code, and scientists have turned to artificial intelligence to crack it. To learn the rules of DNA regulation, they’re using deep neural networks (DNNs), which excel at finding patterns in large datasets. DNNs are at the core of popular AI tools like ChatGPT. Thanks to a new tool developed by Cold Spring Harbor Laboratory Assistant Professor Peter Koo, genome-analyzing DNNs can now be trained with far more data than can be obtained through experiments alone. The name EvoAug stands for evolution augmentations. The Koo lab built its new AI-training model by feeding it augmented data based on the genetic mutations that have driven evolution. Image: © VectorMine – stock.adobe.com

“With DNNs, the mantra is the more data, the better," Koo says. “We really need these models to see a diversity of genomes so they can learn robust motif signals. But in some situations, the biology itself is the limiting factor, because we can’t generate more data than exists inside the cell.”

If an AI learns from too few examples, it may misinterpret how a regulatory motif impacts gene function. The problem is that some motifs are uncommon. Very few examples are found in nature.

To overcome this limitation, Koo and his colleagues developed EvoAug—a new method of augmenting the data used to train DNNs. EvoAug was inspired by a dataset hiding in plain sight—evolution. The process begins by generating artificial DNA sequences that nearly match real sequences found in cells. The sequences are tweaked in the same way genetic mutations have naturally altered the genome during evolution.

Next, the models are trained to recognize regulatory motifs using the new sequences, with one key assumption. It’s assumed the vast majority of tweaks will not disrupt the sequences’ function. Koo compares augmenting the data in this way to training image-recognition software with mirror images of the same cat. The computer learns that a backward cat pic is still a cat pic.

The reality, Koo says, is that some DNA changes do disrupt function. So, EvoAug includes a second training step using only real biological data. This guides the model “back to the biological reality of the dataset,” Koo explains.

Koo’s team found that models trained with EvoAug perform better than those trained on biological data alone. As a result, scientists could soon get a better read of the regulatory DNA that writes the rules of life itself. Ultimately, this could someday provide a whole new understanding of human health.

Sweden shows how cities will need more resilient electricity networks to cope with extreme weather

Tyler O'Neal, Staff Editor NETWORKS

Dense urban areas amplify the effects of higher temperatures, due to the phenomenon of heat islands in cities. This makes cities more vulnerable to extreme climate events. Large investments in the electricity network will be necessary to cool us down during heatwaves and keep us warm during cold snaps, according to a new study led by Lund University in Sweden.

“Unless we account for extreme climate events and continued urbanization, the reliability of electricity supply will fall by up to 30%. An additional outlay of 20-60 percent will be required during the energy transition to guarantee that cities can cope with different kinds of climate,” says Vahid Nik, Professor of Building Physics at Lund University.

The study presents a modeling platform that ties together climate, building, and energy system models to facilitate the simulation and evaluation of cities’ energy transition. The aim is to secure the cities’ resilience against future climate changes at the same time as the densification of urban areas is taking place. In particular, researchers have looked closely at extreme weather events (e.g. heatwaves and cold snaps) by producing simulations of urban microclimates. 

“Our results show that high-density areas give rise to a phenomenon called urban heat islands, which make cities more vulnerable to the effects of extreme climate events, particularly in southern Europe. For example, the outdoor temperature can rise by 17% while the wind speed falls by 61%. Urban densification – a recommended development strategy to reach the UN’s energy and climate goals – could make the electricity network more vulnerable. This must be taken into consideration when designing urban energy systems, says Kavan Javanroodi, Assistant Professor in Building and Urban Physics.

“The framework we have developed connects future climate models to buildings and energy systems at the city level, taking the urban microclimate into account. For the first time, we are getting to grips with several challenges around the issues of future climate uncertainty and extreme weather situations, focussing in particular on what are known as ‘HILP’ or High Impact Low Probability events”, says Vahid Nik.

There is still a large gap between future climate modeling and building and energy analyses and their links to one another. According to Vahid Nik, the model now being developed makes a great contribution to closing that gap. 

“Our results answer questions like ‘How big an effect will extreme weather events have in the future, given the predicted pace of urbanization and several different future climate scenarios?’, ‘How do we take them and the connections between them into account?’ and ‘How does the nature of urban development contribute to exacerbating or mitigating the effects of extreme events at the regional and municipal level?’ “

The results show that the peaks in demand in the energy system increase more than previously thought when extreme microclimates are taken into account, for example with an increase in cooling demand of 68% in Stockholm and 43% in Madrid on the hottest day of the year. Not considering this can lead to incorrect estimates of cities’ energy requirements, which can turn into power shortages and even blackouts. 

“There is a marked deviation between the heat and cooling requirements shown in today’s urban climate models, compared to the outcomes of our calculations when urban morphology, the physical design of the city, is more complex. For example, if we fail to take into account the urban climate in Madrid, we could underestimate the need for cooling by around 28%,” says Kavan Javanroodi.

Vahid Nik explains that an increasing number of countries have become interested in extreme weather events, energy issues, and their impact on public health. At the same time, there are no methods of quantifying the effects of climate change and planning for adapting to them, especially when it comes to extreme weather events and climate variations across space and time. 

“Our efforts can contribute to making societies more prepared for climate change. Future research should aim to examine the relationship between urban density and climate change in energy forecasts. Furthermore, we ought to develop more innovative methods of increasing energy flexibility and climate resilience in cities, which is a major focus of research for our team at the moment,” says Vahid Nik.