Tag Archives: #artificialintelligence

New Study Investigates Photonics for Artificial Intelligence and Neuromorphic Computing (Computer Science)

Scientists have given a fascinating new insight into the next steps to develop fast, energy-efficient, future computing systems that use light instead of electrons to process and store information – incorporating hardware inspired directly by the functioning of the human brain.

A team of scientists, including Professor C. David Wright from the University of Exeter, has explored the future potential for computer systems by using photonics in place of conventional electronics.

The article is published today (January 29th 2021) in the prestigious journal Nature Photonics.

The study focuses on potential solutions to one of the world’s most pressing computing problems – how to develop computing technologies to process this data in a fast and energy efficient way.

Contemporary computers are based on the von Neumann architecture in which the fast Central Processing Unit (CPU) is physically separated from the much slower program and data memory.

This means computing speed is limited and power is wasted by the need to continuously transfer data to and from the memory and processor over bandwidth-limited and energy-inefficient electrical interconnects – known as the von Neumann bottleneck.

As a result, it has been estimated that more than 50 % of the power of modern computing systems is wasted simply in this moving around of data.

Professor C David Wright, from the University of Exeter’s Department of Engineering, and one of the co-authors of the study explains “Clearly, a new approach is needed – one that can fuse together the core information processing tasks of computing and memory, one that can incorporate directly in hardware the ability to learn, adapt and evolve, and one that does away with energy-sapping and speed-limiting electrical interconnects.”

Photonic neuromorphic computing is one such approach. Here, signals are communicated and processed using light rather than electrons, giving access to much higher bandwidths (processor speeds) and vastly reducing energy losses.

Moreover, the researchers try to make the  computing hardware itself isomorphic with biological processing system (brains), by developing devices to directly mimic the basic functions of brain neurons and synapses, then connecting these together in networks that can offer fast, parallelised, adaptive processing for artificial intelligence and machine learning applications.

The state-of-the-art of such photonic ‘brain-like’ computing, and its likely future development, is the focus of an article entitled “Photonics for artificial intelligence and neuromorphic computing” published in the prestigious journal Nature Photonics by a leading international team of researchers from the USA, Germany and UK.

Featured image: Conceptual layout of a future photonic neuromorphic computer. Picture credit: Thomas Ferreira de Lima


Reference: Shastri, B.J., Tait, A.N., Ferreira de Lima, T. et al. Photonics for artificial intelligence and neuromorphic computing. Nat. Photonics 15, 102–114 (2021). https://doi.org/10.1038/s41566-020-00754-y


Provided by University of Exeter

Robust Artificial intelligence Tools to Predict Future Cancer (Medicine)

Researchers created a risk-assessment algorithm that shows consistent performance across datasets from US, Europe, and Asia.

To catch cancer earlier, we need to predict who is going to get it in the future. The complex nature of forecasting risk has been bolstered by artificial intelligence (AI) tools, but the adoption of AI in medicine has been limited by poor performance on new patient populations and neglect to racial minorities

Two years ago, a team of scientists from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) and Jameel Clinic (J-Clinic) demonstrated a deep learning system to predict cancer risk using just a patient’s mammogram. The model showed significant promise and even improved inclusivity: It was equally accurate for both white and Black women, which is especially important given that Black women are 43 percent more likely to die from breast cancer. 

But to integrate image-based risk models into clinical care and make them widely available, the researchers say the models needed both algorithmic improvements and large-scale validation across several hospitals to prove their robustness. 

To that end, they tailored their new “Mirai” algorithm to capture the unique requirements of risk modeling. Mirai jointly models a patient’s risk across multiple future time points, and can optionally benefit from clinical risk factors such as age or family history, if they are available. The algorithm is also designed to produce predictions that are consistent across minor variances in clinical environments, like the choice of mammography machine.

Video: Robust artificial intelligence tools may be used to predict future breast cancer. © MIT

The team trained Mirai on the same dataset of over 200,000 exams from Massachusetts General Hospital (MGH) from their prior work, and validated it on test sets from MGH, the Karolinska Institute in Sweden, and Chang Gung Memorial Hospital in Taiwan. Mirai is now installed at MGH, and the team’s collaborators are actively working on integrating the model into care. 

Mirai was significantly more accurate than prior methods in predicting cancer risk and identifying high-risk groups across all three datasets. When comparing high-risk cohorts on the MGH test set, the team found that their model identified nearly two times more future cancer diagnoses compared the current clinical standard, the Tyrer-Cuzick model. Mirai was similarly accurate across patients of different races, age groups, and breast density categories in the MGH test set, and across different cancer subtypes in the Karolinska test set. 

“Improved breast cancer risk models enable targeted screening strategies that achieve earlier detection, and less screening harm than existing guidelines,” says Adam Yala, CSAIL PhD student and lead author on a paper about Mirai that was published this week in Science Translational Medicine. “Our goal is to make these advances part of the standard of care. We are partnering with clinicians from Novant Health in North Carolina, Emory in Georgia, Maccabi in Israel, TecSalud in Mexico, Apollo in India, and Barretos in Brazil to further validate the model on diverse populations and study how to best clinically implement it.” 

How it works 

Despite the wide adoption of breast cancer screening, the researchers say the practice is riddled with controversy: More-aggressive screening strategies aim to maximize the benefits of early detection, whereas less-frequent screenings aim to reduce false positives, anxiety, and costs for those who will never even develop breast cancer.  

Current clinical guidelines use risk models to determine which patients should be recommended for supplemental imaging and MRI. Some guidelines use risk models with just age to determine if, and how often, a woman should get screened; others combine multiple factors related to age, hormones, genetics, and breast density to determine further testing. Despite decades of effort, the accuracy of risk models used in clinical practice remains modest.  

Recently, deep learning mammography-based risk models have shown promising performance. To bring this technology to the clinic, the team identified three innovations they believe are critical for risk modeling: jointly modeling time, the optional use of non-image risk factors, and methods to ensure consistent performance across clinical settings. 

1. Time

Inherent to risk modeling is learning from patients with different amounts of follow-up, and assessing risk at different time points: this can determine how often they get screened, whether they should have supplemental imaging, or even consider preventive treatments. 

Although it’s possible to train separate models to assess risk for each time point, this approach can result in risk assessments that don’t make sense — like predicting that a patient has a higher risk of developing cancer within two years than they do within five years. To address this, the team designed their model to predict risk at all time points simultaneously, by using a tool called an “additive-hazard layer.” 

The additive-hazard layer works as follows: Their network predicts a patient’s risk at a time point, such as five years, as an extension of their risk at the previous time point, such as four years. In doing so, their model can learn from data with variable amounts of follow-up, and then produce self-consistent risk assessments. 

2. Non-image risk factors

While this method primarily focuses on mammograms, the team wanted to also use non-image risk factors such as age and hormonal factors if they were available — but not require them at the time of the test. One approach would be to add these factors as an input to the model with the image, but this design would prevent the majority of hospitals (such as Karolinska and CGMH), which don’t have this infrastructure, from using the model. 

For Mirai to benefit from risk factors without requiring them, the network predicts that information at training time, and if it’s not there, it can use its own predictive version. Mammograms are rich sources of health information, and so many traditional risk factors such as age and menopausal status can be easily predicted from their imaging. As a result of this design, the same model could be used by any clinic globally, and if they have that additional information, they can use it. 

3. Consistent performance across clinical environments

To incorporate deep-learning risk models into clinical guidelines, the models must perform consistently across diverse clinical environments, and its predictions cannot be affected by minor variations like which machine the mammogram was taken on. Even across a single hospital, the scientists found that standard training did not produce consistent predictions before and after a change in mammography machines, as the algorithm could learn to rely on different cues specific to the environment. To de-bias the model, the team used an adversarial scheme where the model specifically learns mammogram representations that are invariant to the source clinical environment, to produce consistent predictions. 

To further test these updates across diverse clinical settings, the scientists evaluated Mirai on new test sets from Karolinska in Sweden and Chang Gung Memorial Hospital in Taiwan, and found it obtained consistent performance. The team also analyzed the model’s performance across races, ages, and breast density categories in the MGH test set, and across cancer subtypes on the Karolinska dataset, and found it performed similarly across all subgroups. 

“African-American women continue to present with breast cancer at younger ages, and often at later stages,” says Salewai Oseni, a breast surgeon at Massachusetts General Hospital who was not involved with the work. “This, coupled with the higher instance of triple-negative breast cancer in this group, has resulted in increased breast cancer mortality. This study demonstrates the development of a risk model whose prediction has notable accuracy across race. The opportunity for its use clinically is high.” 

Here’s how Mirai works: 

1. The mammogram image is put through something called an “image encoder.”

2. Each image representation, as well as which view it came from, is aggregated with other images from other views to obtain a representation of the entire mammogram.

3. With the mammogram, a patient’s traditional risk factors are predicted using a Tyrer-Cuzick model (age, weight, hormonal factors). If unavailable, predicted values are used. 

4. With this information, the additive-hazard layer predicts a patient’s risk for each year over the next five years. 

Improving Mirai 

Although the current model doesn’t look at any of the patient’s previous imaging results, changes in imaging over time contain a wealth of information. In the future the team aims to create methods that can effectively utilize a patient’s full imaging history.

In a similar fashion, the team notes that the model could be further improved by utilizing “tomosynthesis,” an X-ray technique for screening asymptomatic cancer patients. Beyond improving accuracy, additional research is required to determine how to adapt image-based risk models to different mammography devices with limited data. 

“We know MRI can catch cancers earlier than mammography, and that earlier detection improves patient outcomes,” says Yala. “But for patients at low risk of cancer, the risk of false-positives can outweigh the benefits. With improved risk models, we can design more nuanced risk-screening guidelines that offer more sensitive screening, like MRI, to patients who will develop cancer, to get better outcomes while reducing unnecessary screening and over-treatment for the rest.” 

“We’re both excited and humbled to ask the question if this AI system will work for African-American populations,” says Judy Gichoya, MD, MS and assistant professor of interventional radiology and informatics at Emory University, who was not involved with the work. “We’re extensively studying this question, and how to detect failure.” 

Yala wrote the paper on Mirai alongside MIT research specialist Peter G. Mikhael, radiologist Fredrik Strand of Karolinska University Hospital, Gigin Lin of Chang Gung Memorial Hospital, Associate Professor Kevin Smith of KTH Royal Institute of Technology, Professor Yung-Liang Wan of Chang Gung University, Leslie Lamb of MGH, Kevin Hughes of MGH, senior author and Harvard Medical School Professor Constance Lehman of MGH, and senior author and MIT Professor Regina Barzilay. 

The work was supported by grants from Susan G Komen, Breast Cancer Research Foundation, Quanta Computing, and the MIT Jameel Clinic. It was also supported by Chang Gung Medical Foundation Grant, and by Stockholm Läns Landsting HMT Grant.

Featured image: MIT researchers have improved their machine learning system developed to predict cancer risk from mammogram images, and validated their effectiveness with studies across several hospitals. Credits: Images courtesy of the researchers.


Reference: Adam Yala, Peter G. Mikhael, Fredrik Strand, Gigin Lin, Kevin Smith, Yung-Liang Wan, Leslie Lamb, Kevin Hughes, Constance Lehman, Regina Barzilay, “Toward robust mammography-based models for breast cancer risk”, Science Translational Medicine 27 Jan 2021: Vol. 13, Issue 578, eaba4373 DOI: 10.1126/scitranslmed.aba4373 https://stm.sciencemag.org/content/13/578/eaba4373


Provided by MIT

Artificial Intelligence Beats Us in Chess, But Not in Memory (Engineering)

The brain strategy for storing memories is more efficient than AI’s one, a new study reveals.

In the last decades, Artificial Intelligence has shown to be very good at achieving exceptional goals in several fields. Chess is one of them: in 1996, for the first time, the computer Deep Blue beat a human player, chess champion Garry Kasparov. A new piece of research shows now that the brain strategy for storing memories may lead to imperfect memories, but in turn, allows it to store more memories, and with less hassle than AI. The new study, carried out by SISSA scientists in collaboration with Kavli Institute for Systems Neuroscience & Centre for Neural Computation, Trondheim, Norway, has just been published in Physical Review Letters.

The brain strategy for storing memories may lead to imperfect memories, but in turn, allows it to store more memories, and with less hassle than AI. © Shahab Mohsenin

Neural networks, real or artificial, learn by tweaking the connections between neurons. Making them stronger or weaker, some neurons become more active, some less, until a pattern of activity emerges. This pattern is what we call “a memory”. The AI strategy is to use complex long algorithms, which iteratively tune and optimize the connections. The brain does it much simpler: each connection between neurons changes just based on how active the two neurons are at the same time. When compared to the AI algorithm, this had long been thought to permit the storage of fewer memories. But, in terms of memory capacity and retrieval, this wisdom is largely based on analysing networks assuming a fundamental simplification: that neurons can be considered as binary units.

The new research, however, shows otherwise: the fewer number of memories stored using the brain strategy depends on such unrealistic assumption. When the simple strategy used by the brain to change the connections is combined with biologically plausible models for single neurons response, that strategy performs as well as, or even better, than AI algorithms. How could this be the case? Paradoxically, the answer is in introducing errors: when a memory is effectively retrieved this can be identical to the original input-to-be-memorized or correlated to it. The brain strategy leads to the retrieval of memories which are not identical to the original input, silencing the activity of those neurons that are only barely active in each pattern. Those silenced neurons, indeed, do not play a crucial role in distinguishing among the different memories stored within a same network. By ignoring them, neural resources can be focused on those neurons that do matter in an input-to-be-memorized and enable a higher capacity.

Overall, this research highlights how biologically plausible self-organized learning procedures can be just as efficient as slow and neurally implausible training algorithms.

Reference: Schönsberg, Francesca, Yasser Roudi, and Alessandro Treves, “Efficiency of Local Learning Rules in Threshold-Linear Associative Networks”, Physical Review Letters 126.1 (2021): 018301 https://journals.aps.org/prl/abstract/10.1103/PhysRevLett.126.018301

Provided by SISSA

Artificial Intelligence Discovers 100,000 New Craters on the Moon (Planetary Science)

More than 100,000 new craters have been identified on the Moon thanks to an artificial intelligence system based on machine learning, developed by a group of researchers led by Yang Chen of the Chinese University of Jilin (a former member for several years of the Department of Information Engineering and Computer Science -DISI, first as a doctoral student and then as a post-doctoral researcher) with the contribution of Lorenzo Bruzzone, professor at DISI.

The research work was published in the prestigious Nature Communication journal in an article entitled “Lunar impact crater identification and age estimation with Chang’E data by deep and transfer learning”This is the largest database of lunar craters in the world, a dozen times larger than what has been reported in previous databases so far.

The sensational achievement is the result of a machine learning algorithm based on a particular convolutional neural network architecture (CNN or ConvNet). The algorithm used the entire database of images acquired by the two Chinese lunar missions Chang’e-1 and Chang’e-2 and, based on approximately 10,000 known craters defined by the International Astronomical Union (IAU) in previous decades, was able to automatically learn which are the most effective features to model craters and to search for them on the entire lunar surface. In this way, scientists were able to identify 117,240 new craters ranging in diameter from about 1 km up to 532 km, mainly distributed in the mid- and low-latitude regions of the Moon. 

Thanks to the machine learning system, the researchers were able not only to detect irregular or degraded craters, but also small craters difficult to identify with conventional systems (88.14% of the detected craters have a diameter of less than 10 km ). 

Furthermore, based on morphological and stratigraphic information, the scientists developed a second neural network architecture capable of automatically establishing thegeological age of nearly 19,000 new craters located in the mid- and low-latitude regions of the Moon, thus creating a database 13 times larger than any existing one.The scientific achievement is extremely important as Lunar craters, generated by impacts with asteroids and comets, can be considered as fossils that describe both the evolution of the Moon and of the Earth that were impacted by the same population over time. However, on our planet  tectonic plate activity and erosion have deleted many of these traces.  

“The automatic learning methodology we have adopted” – explains prof. Bruzzone – is based on Transfer Learning (TL). It is aimed at exploiting what has been learned on the low-resolution images of the Chang’e-1 mission for the analysis of the high-resolution images of the Chang’e-2 mission. Basically, this machine learning system is similar to a supervisor passing on its knowledge and experience from one generation to the next, applying what it has learned previously to solve new problems. This approach allows an automatic, accurate and consistent analysis of craters distribution over an entire celestial body”. 

In the future, the methodology developed by the researchers could therefore be adapted to other Solar System bodies, for example: Mars, Mercury, Venus, Vesta and Ceres; this will make it possible to automatically and reliably extract information on a global scale that are difficult to derive by conventional manual analyses or automatic techniques of previous generation.

Reference: Yang, C., Zhao, H., Bruzzone, L. et al. Lunar impact crater identification and age estimation with Chang’E data by deep and transfer learning. Nat Commun 11, 6358 (2020). https://doi.org/10.1038/s41467-020-20215-y

Provided by University of Trento

Electronic Skin Could Help Robots Get In Touch With Their Feelings (Electronics /AI / Material Science)

A new type of energy-generating synthetic skin could create more affordable prosthetic limbs and robots capable of mimicking the sense of touch, scientists say.

In an early-view paper published in the journal IEEE Transactions on Robotics, researchers from the University of Glasgow describe how a robotic hand wrapped in their flexible solar skin is capable of interacting with objects without using dedicated and expensive touch sensors.

Ravindra Dhaiya with robotic hand ©University of Glasgow

Instead, the skin puts the array of miniaturised solar cells integrated on its soft polymer surface to a clever dual use. The cells generate enough energy to power the micro-actuators which control the hand’s movements, but they also provide the hand with its unique sense of ‘touch’ by measuring the variations in the solar cells’ output.

As objects get closer to the surface of a cell, they reduce the amount of light which reaches it. The amount of power the cell generates drops as the light gets dimmer, eventually reaching zero when an object touches and covers it. By making clever interpretations of the levels of power produced in each cell, the skin is capable of detecting the shape of an incoming object.

A second set of simple LEDs, integrated between the solar cells in the skin, transmit infra-red light towards objects. By measuring the time the light takes to reflect from the object, the skin can sense the distance between the object and the hand.

Combining the information collected from the solar cells and LEDs allows the skin’s processor to deduce an object’s proximity, location, and edges, replicating many of the parameters measured by more traditional touch sensors. Together, the data allows the hand to grasp objects like rubber balls placed in front of it.

It’s the latest development in electronic skin from the University of Glasgow’s Bendable Electronics and Sensing Technologies (BEST) Group, led by Professor Ravinder Dahiya.

Professor Dahiya, of the University’s James Watt School of Engineering, said: “Touch-sensitive electronic skin has found numerous experimental applications in prosthetics and robotics in recent years, but our project is the first energy-generating e-skin capable of offering touch feedback without using dedicated touch sensors.

“That lack of sensors means the skin requires no conventional power source to work, unlike other equivalent devices which include touch sensors. In fact, the skin itself is the source of energy, capable of powering the hand and devices attached to it. The generated power can be stored in devices such as flexible supercapacitors we’ve developed to work alongside the skin, so it doesn’t have to be constantly exposed to the sun in order to work.

“It’s one step closer to a completely self-powered prosthetic wrapped in flexible skin made from relatively inexpensive components. The sensing capabilities built into the skin could even lead to skin that can ‘see’ – further refinements could help the skin identify approaching objects even before they make contact.”

“We’ve also experimented with adding the hand to the end of a robot arm, similar to the ones found in places like car manufacturing facilities. The skin’s sensors are capable of stopping the arm’s motion when it senses an unexpected object, which we believe could help prevent future industrial accidents.”

The team’s paper, titled ‘Energy Generating Electronic Skin With Intrinsic Tactile Sensing Without Touch Sensors’, is published in early access in EEE Transactions on Robotics

The research was supported by funding from the Engineering and Physical Sciences Research Council (EPSRC).

Provided by University of Glasgow

Machine Learning: A Breakthrough In The Study of Stellar Nurseries (Astronomy)

Artificial intelligence can make it possible to see astrophysical phenomena that were previously beyond reach. This has now been demonstrated by scientists from the CNRS, IRAM, Observatoire de Paris-PSL, Ecole Centrale Marseille and Ecole Centrale Lille, working together in the ORION-B programme. In a series of three papers published in Astronomy & Astrophysics on 19 November 2020, they present the most comprehensive observations yet carried out of one of the star-forming regions closest to the Earth.

Emission from carbon monoxide in the Orion B molecular cloud © J. Pety/ORION-B Collaboration/IRAM

The gas clouds in which stars are born and evolve are vast regions of the Universe that are extremely rich in matter, and hence in physical processes. All these processes are intertwined on different size and time scales, making it almost impossible to fully understand such stellar nurseries. However, the scientists in the ORION-B programme have now shown that statistics and artificial intelligence can help to break down the barriers still standing in the way of astrophysicists.

With the aim of providing the most detailed analysis yet of the Orion molecular cloud, one of the star-forming regions nearest the Earth, the ORION-B team included in its ranks scientists specialising in massive data processing. This enabled them to develop novel methods based on statistical learning and machine learning to study observations of the cloud made at 240 000 frequencies of light.

Based on artificial intelligence algorithms, these tools make it possible to retrieve new information from a large mass of data such as that used in the ORION-B project. This enabled the scientists to uncover a certain number of ‘laws’ governing the Orion molecular cloud.

For instance, they were able to discover the relationships between the light emitted by certain molecules and information that was previously inaccessible, namely, the quantity of hydrogen and of free electrons in the cloud, which they were able to estimate from their calculations without observing them directly. By analysing all the data available to them, the research team was also able to determine ways of further improving their observations by eliminating a certain amount of unwanted information.

The ORION-B teams now wish to put this theoretical work to the test, by applying the estimates and recommendations obtained and verifying them under real conditions. Another major theoretical challenge will be to extract information about the speed of molecules, and hence visualise the motion of matter in order to see how it moves within the cloud.

References: (1) P. Gratier et al. Quantitative inference of the H2 column densities from 3mm molecular emission: Case study towards Orion B, Astronomy & Astrophysics (2020). DOI: 10.1051/0004-6361/202037871 (2) E. Bron et al. Tracers of the ionization fraction in dense and translucent gas. I. Automated exploitation of massive astrochemical model grids, Astronomy & Astrophysics (2020). DOI: 10.1051/0004-6361/202038040 (3) Roueff et al., C18O, 13CO, and 12CO abundances and excitation temperatures in the Orion B molecular cloud: An analysis of the precision achievable when modeling spectral line within the Local Thermodynamic Equilibrium approximation. arxiv.org/abs/2005.08317

Provided by CNRS

Physics Can Assist With Key Challenges in Artificial Intelligence (Engineering)

A physical mechanism a priori reveals how many examples in deep learning are required to achieve a desired test accuracy. It surprisingly indicates that learning each example once is almost equivalent to learning examples repeatedly.

Current research and applications in the field of artificial intelligence (AI) include several key challenges. These include: (a) A priori estimation of the required dataset size to achieve a desired test accuracy. For example, how many handwritten digits does a machine have to learn before being able to predict a new one with a success rate of 99%? Similarly, how many specific types of circumstances does an autonomous vehicle have to learn before its reaction will not lead to an accident? (b) The achievement of reliable decision-making under a limited number of examples, where each example can be trained only once, i.e., observed only for a short period. This type of realization of fast on-line decision making is representative of many aspects of human activity, robotic control and network optimization.

In an article published today in the journal Scientific Reports, researchers from Bar-Ilan University show how two challenges in current research and applications in the field of artificial intelligence are solved by adopting a physical concept that was introduced a century ago to describe the formation of a magnet during a process of iron bulk cooling. Using a careful optimization procedure and exhaustive simulations, the scientists have demonstrated the usefulness of the physical concept of power-law scaling to deep learning. This central concept in physics, which arises from diverse phenomena, including the timing and magnitude of earthquakes, Internet topology and social networks, stock price fluctuations, word frequencies in linguistics, and signal amplitudes in brain activity, has also been found to be applicable in the ever-growing field of AI, and especially deep learning. Image Rapid decision making: A deep learning neural network where each handwritten digit is presented only once to the trained network. ©Prof. Ido Kanter, Bar-Ilan University

In an article published today in the journal Scientific Reports, researchers show how these two challenges are solved by adopting a physical concept that was introduced a century ago to describe the formation of a magnet during a process of iron bulk cooling.

Using a careful optimization procedure and exhaustive simulations, a group of scientists from Bar-Ilan University has demonstrated the usefulness of the physical concept of power-law scaling to deep learning. This central concept in physics, which arises from diverse phenomena, including the timing and magnitude of earthquakes, Internet topology and social networks, stock price fluctuations, word frequencies in linguistics, and signal amplitudes in brain activity, has also been found to be applicable in the ever-growing field of AI, and especially deep learning.

“Test errors with online learning, where each example is trained only once, are in close agreement with state-of-the-art algorithms consisting of a very large number of epochs, where each example is trained many times. This result has an important implication on rapid decision making such as robotic control,” said Prof. Ido Kanter, of Bar-Ilan’s Department of Physics and Gonda (Goldshmied) Multidisciplinary Brain Research Center, who led the research. “The power-law scaling, governing different dynamical rules and network architectures, enables the classification and hierarchy creation among the different examined classification or decision problems,” he added.

“One of the important ingredients of the advanced deep learning algorithm is the recent new bridge between experimental neuroscience and advanced artificial intelligence learning algorithms,” said PhD student Shira Sardi, a co-author of the study. Our new type of experiments on neuronal cultures indicate that an increase in the training frequency enables us to significantly accelerate the neuronal adaptation process. “This accelerated brain-inspired mechanism enables building advanced deep learning algorithms which outperform existing ones,” said PhD student Yuval Meir, another co-author.

The reconstructed bridge from physics and experimental neuroscience to machine learning is expected to advance artificial intelligence and especially ultrafast decision making under limited training examples as to contribute to the formation of a theoretical framework of the field of deep learning.

References: Meir, Y., Sardi, S., Hodassman, S. et al. Power-law scaling to assist with key challenges in artificial intelligence. Sci Rep 10, 19628 (2020). https://doi.org/10.1038/s41598-020-76764-1

Provided by Bar-Ilan University

Artificial Intelligence Reveals Hundreds Of Millions Of Trees In The Sahara (Earth Science)

If you think that the Sahara is covered only by golden dunes and scorched rocks, you aren’t alone. Perhaps it’s time to shelve that notion. In an area of West Africa 30 times larger than Denmark, an international team, led by University of Copenhagen and NASA researchers, has counted over 1.8 billion trees and shrubs. The 1.3 million km2 area covers the western-most portion of the Sahara Desert, the Sahel and what are known as sub-humid zones of West Africa.

Dryland landscape in West Africa. © Martin Brandt

“We were very surprised to see that quite a few trees actually grow in the Sahara Desert, because up until now, most people thought that virtually none existed. We counted hundreds of millions of trees in the desert alone. Doing so wouldn’t have been possible without this technology. Indeed, I think it marks the beginning of a new scientific era,” asserts Assistant Professor Martin Brandt of the University of Copenhagen’s Department of Geosciences and Natural Resource Management, lead author of the study’s scientific article, now published in Nature.

The work was achieved through a combination of detailed satellite imagery provided by NASA, and deep learning — an advanced artificial intelligence method. Normal satellite imagery is unable to identify individual trees, they remain literally invisible. Moreover, a limited interest in counting trees outside of forested areas led to the prevailing view that there were almost no trees in this particular region. This is the first time that trees across a large dryland region have been counted.

The role of trees in the global carbon budget

New knowledge about trees in dryland areas like this is important for several reasons, according to Martin Brandt. For example, they represent an unknown factor when it comes to the global carbon budget:

“Trees outside of forested areas are usually not included in climate models, and we know very little about their carbon stocks. They are basically a white spot on maps and an unknown component in the global carbon cycle,” explains Martin Brandt.

The red rectangle marks the area where the trees were mapped. ©Martin Brandt

Furthermore, the new study can contribute to better understanding the importance of trees for biodiversity and ecosystems and for the people living in these areas. In particular, enhanced knowledge about trees is also important for developing programmes that promote agroforestry, which plays a major environmental and socio-economic role in arid regions.

“Thus, we are also interested in using satellites to determine tree species, as tree types are significant in relation to their value to local populations who use wood resources as part of their livelihoods. Trees and their fruit are consumed by both livestock and humans, and when preserved in the fields, trees have a positive effect on crop yields because they improve the balance of water and nutrients,” explains Professor Rasmus Fensholt of the Department of Geosciences and Natural Resource Management.

Technology with a high potential

The research was conducted in collaboration with the University of Copenhagen’s Department of Computer Science, where researchers developed the deep learning algorithm that made the counting of trees over such a large area possible.

The researchers show the deep learning model what a tree looks like: They do so by feeding it thousands of images of various trees. Based upon the recognition of tree shapes, the model can then automatically identify and map trees over large areas and thousands of images. The model needs only hours what would take thousands of humans several years to achieve.

“This technology has enormous potential when it comes to documenting changes on a global scale and ultimately, in contributing towards global climate goals. We are motivated to develop this type of beneficial artificial intelligence,” says professor and co-author Christian Igel of the Department of Computer Science.

The next step is to expand the count to a much larger area in Africa. And in the longer term, the aim is to create a global database of all trees growing outside forest areas.

References: Brandt, M., Tucker, C.J., Kariryaa, A. et al. An unexpectedly large count of trees in the West African Sahara and Sahel. Nature (2020). https://doi.org/10.1038/s41586-020-2824-5 link: http://dx.doi.org/10.1038/s41586-020-2824-5

Provided by University Of Copenhagen

Making New Materials Using AI (Material Science)

There is an old saying, “If rubber is the material that opened the way to the ground, aluminum is the one that opened the way to the sky.” New materials were always discovered at each turning point that changed human history. Materials used in memory devices are also drastically evolving with the emergence of new materials such as doped silicon materials, resistance changing materials, and materials that spontaneously magnetize and polarize. How are these new materials made? A research team from POSTECH has revealed the mechanism behind making materials used in new memory devices by using artificial intelligence.

The research team led by Professor Si-Young Choi of Department of Materials Science and Engineering and the team led by Professor Daesu Lee of the Department of Physics at POSTECH have together succeeded in synthesizing a novel substance that produces electricity by causing polarization (a phenomenon in which the position of negative and positive charges is separated from the negative and positive charges within the crystal) at room temperature and confirmed its variation in the crystal structure by applying deep neural network analysis. This paper was published in the recent issue of Nature Communications.

The atomic structures of perovskite oxides are often distorted and their properties are determined by the oxygen octahedral rotation (OOR) accordingly. In fact, there are only a few stable OOR patterns present at equilibrium and this inevitably limits the properties and functions of perovskite oxides.

The joint research team focused on a perovskite oxide called CaTiO3 [1] which remains nonpolar (or paraelectric) even at the absolute temperature of 0K. Based on the ab-initio calculations, however, the team found that a unique OOR pattern [2] that does not naturally exist would be able to facilitate the ferroelectricity, a powerful polarization at room temperature.

In this light, the research team succeeded in synthesizing a novel material (heteroepitaxial CaTiO3) that possesses the ferroelectricity by applying interface engineering [3] that controls the atomic structures at the interface and accordingly its physical property.

In addition, deep neural network analysis was applied to examine the fine OOR and the variation of a few decades of picometer in the atomic structures, and various atomic structures were simulated and data were utilized for AI analysis to identify artificially controlled OOR patterns.

“We have confirmed that we can create new physical phenomena that do not naturally occur by obtaining the unique OOR pattern through controlling the variation in its atomic structure,” remarked Professor Daesu Lee. “It is especially significant to see that the results of the convergent research of physics and new materials engineering enable calculations for material design, synthesis of novel materials, and analysis to understand new phenomena.”

Professor Si-Young Choi explained, “By applying the deep machine learning to materials research, we have successfully identified atomic-scale variations on tens of picometers that are difficult to identify with the human eye.” He added, “It could be an advanced approach for materials analysis that can help to understand the mechanism for creating new materials with unique physical phenomena.”

The findings are the result of convergence research conducted by the Department of Physics and the Department of Materials Science and Engineering at POSTECH and Seoul National University’s Center for Correlated Electron Systems. It was conducted with the support from the Mid-career Researcher Program and the Global Frontier Hybrid Interface Materials Program of the National Research Foundation of Korea and the POSTECH-Samsung Electronics Industry-Academia Cooperative Research Center.

___________________________________________

  1. CaTiO3
    Oxygen octahedral structure defined as a-b+a-
  2. Oxygen octahedral rotation pattern
    Having an oxygen octahedral structure defined as a-a-a-
  3. Interface engineering
    The boundary between two substances of different materials is called interface. Interface engineering is the study of the conditions and properties of this interface and its surrounding materials.

Provided by POSTECH