Does This A.I. Think Like a Human?

A new technique compares the reasoning of a machine-learning model to that of a human, so the user can see patterns in the model’s behavior.

From:  Adam Zewe at the M.I.T. News Office

April 6, 2022 -- MIT researchers developed a method that helps a user understand a machine-learning model’s reasoning, and how that reasoning compares to that of a human.

In machine learning, understanding why a model makes certain decisions is often just as important as whether those decisions are correct. For instance, a machine-learning model might correctly predict that a skin lesion is cancerous, but it could have done so using an unrelated blip on a clinical photo.

While tools exist to help experts make sense of a model’s reasoning, often these methods only provide insights on one decision at a time, and each must be manually evaluated. Models are commonly trained using millions of data inputs, making it almost impossible for a human to evaluate enough decisions to identify patterns.

Now, researchers at MIT and IBM Research have created a method that enables a user to aggregate, sort, and rank these individual explanations to rapidly analyze a machine-learning model’s behavior. Their technique, called Shared Interest, incorporates quantifiable metrics that compare how well a model’s reasoning matches that of a human.

Shared Interest could help a user easily uncover concerning trends in a model’s decision-making — for example, perhaps the model often becomes confused by distracting, irrelevant features, like background objects in photos. Aggregating these insights could help the user quickly and quantitatively determine whether a model is trustworthy and ready to be deployed in a real-world situation.

“In developing Shared Interest, our goal is to be able to scale up this analysis process so that you could understand on a more global level what your model’s behavior is,” says lead author Angie Boggust, a graduate student in the Visualization Group of the Computer Science and Artificial Intelligence Laboratory (CSAIL).

Boggust wrote the paper with her advisor, Arvind Satyanarayan, an assistant professor of computer science who leads the Visualization Group, as well as Benjamin Hoover and senior author Hendrik Strobelt, both of IBM Research. The paper will be presented at the Conference on Human Factors in Computing Systems.

Boggust began working on this project during a summer internship at IBM, under the mentorship of Strobelt. After returning to MIT, Boggust and Satyanarayan expanded on the project and continued the collaboration with Strobelt and Hoover, who helped deploy the case studies that show how the technique could be used in practice.

Human-AI alignment

Shared Interest leverages popular techniques that show how a machine-learning model made a specific decision, known as saliency methods. If the model is classifying images, saliency methods highlight areas of an image that are important to the model when it made its decision. These areas are visualized as a type of heatmap, called a saliency map, that is often overlaid on the original image. If the model classified the image as a dog, and the dog’s head is highlighted, that means those pixels were important to the model when it decided the image contains a dog.

Shared Interest works by comparing saliency methods to ground-truth data. In an image dataset, ground-truth data are typically human-generated annotations that surround the relevant parts of each image. In the previous example, the box would surround the entire dog in the photo. When evaluating an image classification model, Shared Interest compares the model-generated saliency data and the human-generated ground-truth data for the same image to see how well they align.

The technique uses several metrics to quantify that alignment (or misalignment) and then sorts a particular decision into one of eight categories. The categories run the gamut from perfectly human-aligned (the model makes a correct prediction and the highlighted area in the saliency map is identical to the human-generated box) to completely distracted (the model makes an incorrect prediction and does not use any image features found in the human-generated box).

“On one end of the spectrum, your model made the decision for the exact same reason a human did, and on the other end of the spectrum, your model and the human are making this decision for totally different reasons. By quantifying that for all the images in your dataset, you can use that quantification to sort through them,” Boggust explains.

The technique works similarly with text-based data, where key words are highlighted instead of image regions.

Rapid analysis

The researchers used three case studies to show how Shared Interest could be useful to both nonexperts and machine-learning researchers.

In the first case study, they used Shared Interest to help a dermatologist determine if he should trust a machine-learning model designed to help diagnose cancer from photos of skin lesions. Shared Interest enabled the dermatologist to quickly see examples of the model’s correct and incorrect predictions. Ultimately, the dermatologist decided he could not trust the model because it made too many predictions based on image artifacts, rather than actual lesions.

“The value here is that using Shared Interest, we are able to see these patterns emerge in our model’s behavior. In about half an hour, the dermatologist was able to make a confident decision of whether or not to trust the model and whether or not to deploy it,” Boggust says.

In the second case study, they worked with a machine-learning researcher to show how Shared Interest can evaluate a particular saliency method by revealing previously unknown pitfalls in the model. Their technique enabled the researcher to analyze thousands of correct and incorrect decisions in a fraction of the time required by typical manual methods.

In the third case study, they used Shared Interest to dive deeper into a specific image classification example. By manipulating the ground-truth area of the image, they were able to conduct a what-if analysis to see which image features were most important for particular predictions.   

The researchers were impressed by how well Shared Interest performed in these case studies, but Boggust cautions that the technique is only as good as the saliency methods it is based upon. If those techniques contain bias or are inaccurate, then Shared Interest will inherit those limitations.

In the future, the researchers want to apply Shared Interest to different types of data, particularly tabular data which is used in medical records. They also want to use Shared Interest to help improve current saliency techniques. Boggust hopes this research inspires more work that seeks to quantify machine-learning model behavior in ways that make sense to humans.

This work is funded, in part, by the MIT-IBM Watson AI Lab, the United States Air Force Research Laboratory, and the United States Air Force Artificial Intelligence Accelerator.

    https://news.mit.edu/2022/does-this-artificial-intelligence-think-human-0406

  

The Road to Happiness

“The only way to deep happiness is to do something you love to the best of your ability.”  So said Richard Feynman, May 11, 1918 to February 15, 1988, the legendary physics professor at Cal Tech who became known as The Great Explainer.  He also said, “Study hard what interests you the most.”

Reducing Cardiac Risk Factor Lp(a)

Cleveland Clinic-Led Trial Finds That Experimental ‘Gene Silencing’ Therapy Reduces Lipoprotein(a), an Important Risk Factor of Heart Disease, By Up To 98%

From:  Cleveland Clinic

April 3, 2022 -- Findings from a new Cleveland Clinic-led phase 1 trial show that an experimental “gene silencing” therapy reduced blood levels of lipoprotein(a), a key driver of heart disease risk, by up to 98%.

Findings from the “APOLLO Trial: Magnitude and Duration of Effects of a Short-interfering RNA Targeting Lipoprotein(a): A Placebo-controlled Double-blind Dose-ranging Trial” were presented today during a late-breaking science session at the American College of Cardiology’s 71^st Annual Scientific Session and simultaneously published online in the Journal of the American Medical Association.

In the trial, participants who received higher doses of SLN360 – a small interfering RNA (siRNA) therapeutic that “silences” the gene responsible for lipoprotein(a) production – saw their lipoprotein(a) levels  drop by as much as 96%-98%. Five months later, these participants’ lipoprotein(a) – also known as Lp(a) – levels remained 71%-81% lower than baseline.

The findings suggest this siRNA therapy could be a promising treatment to help prevent premature heart disease in people with high levels of Lp(a), which is estimated to affect 64 million people in the United States and 1.4 billion people worldwide. It is estimated that nearly 20 to 25% of the world’s population has elevated Lp(a).

“These results showed the safety and strong efficacy of this experimental treatment at reducing levels of Lp(a), a common, but previously untreatable, genetically-determined risk factor that leads to premature heart attack, stroke and aortic stenosis,” said the study’s lead author Steven E. Nissen, M.D., Chief Academic Officer of the Heart, Vascular & Thoracic Institute at Cleveland Clinic. “We hope that further development of this therapy also will be shown to reduce the consequences of Lp(a) in the clinical setting through future studies.”

Lp(a) has similarities to LDL, also known as bad cholesterol. Lp(a) is made in the liver, where an extra protein called apolipoprotein(a) is attached to an LDL-like particle. Unlike other types of cholesterol particles, Lp(a) levels are 80 to 90% genetically determined. The structure of the Lp(a) particle causes the accumulation of plaques in arteries, which play a significant role in heart disease. Elevated Lp(a) greatly increases the risk of heart attacks and strokes.

Although effective therapies to reduce the risk of heart disease by lowering LDL cholesterol and other lipids exist, currently there are no approved treatments to lower Lp(a). Since Lp(a) levels are determined by a person’s genes, lifestyle changes such as diet or exercise have no effect. In the current study, the siRNA therapy reduces Lp(a) levels by “silencing” the gene responsible for Lp(a) production and blocking creation of apolipoprotein(a) in the liver.

In the APOLLO trial, researchers enrolled 32 people at five medical centers in three countries. All participants had Lp(a) levels above 150 nmol/L, with a median level of 224 nmol/L (75 nmol/L or less is considered normal). Eight participants received a placebo and the remaining received one of four doses of SLN360 via a single subcutaneous injection. The doses were 30 mg, 100 mg, 300 mg and 600 mg. Participants were closely observed for the first 24 hours after their injection and then assessed periodically for five months.

Participants receiving 300 mg and 600 mg of SLN360 had a maximum of 96% and 98% reduction in Lp(a) levels, and a reduction of 71% and 81% at five months compared to baseline. Those receiving a placebo saw no change in Lp(a) levels. The highest doses also reduced LDL cholesterol by about 20%-25%. There were no major safety consequences reported and the most common side effect was temporary soreness at the injection site. The study was extended and researchers will continue to follow participants for a total of one year.

The APOLLO trial was funded by Silence Therapeutics plc (Nasdaq: SLN), London, UK. Dr. Nissen has served as a consultant for many pharmaceutical companies and has overseen clinical trials for Amgen, AstraZeneca, Bristol Myers Squibb, Eli Lilly, Esperion, Novartis, Novo Nordisk, Orexigen, Takeda and Pfizer. However, he does not accept honoraria, consulting fees or other compensation from commercial entities.

The trial was coordinated by the Cleveland Clinic Coordinating Center for Clinical Research (C5Research) and sponsored by Silence Therapeutics plc (Nasdaq: SLN), London, UK.

https://newsroom.clevelandclinic.org/2022/04/03/cleveland-clinic-led-trial-finds-that-experimental-gene-silencing-therapy-reduces-lipoproteina-an-important-risk-factor-of-heart-disease-by-up-to-98/

 

Birds and Bees Make Coffee Better

Study calculates winged helpers’ effects on coffee—while pioneering a better way to measure nature’s ‘unpaid labor’

From:  University of Vermont

April 4, 2022 -- A groundbreaking study reveals that without birds and bees working together, some traveling thousands of miles, coffee farmers would see a whopping 25% drop in crop yields. Coffee is bigger and more plentiful when birds and bees team up to protect and pollinate coffee plants. The study is also the first to show, with real-world experiments, that the contributions of nature -- ie. from bees and birds -- are larger combined than their individual contributions. This suggests researchers may be underestimating how much the environment benefits society.

Without these winged helpers, some traveling thousands of miles, coffee farmers would see a 25% drop in crop yields, a loss of roughly $1,066 per hectare of coffee.

That's important for the $26 billion coffee industry -- including consumers, farmers, and corporations who depend on nature's unpaid labor for their morning buzz -- but the research has even broader implications.

The forthcoming study in the Proceedings of the National Academy of Sciences is the first to show, using real-world experiments at 30 coffee farms, that the contributions of nature -- in this case, bee pollination and pest control by birds -- are larger combined than their individual contributions.

"Until now, researchers have typically calculated the benefits of nature separately, and then simply added them up," says lead author Alejandra Martínez-Salinas of the Tropical Agricultural Research and Higher Education Center (CATIE). "But nature is an interacting system, full of important synergies and trade-offs. We show the ecological and economic importance of these interactions, in one of the first experiments at realistic scales in actual farms."

"These results suggest that past assessments of individual ecological services -- including major global efforts like IPBES -- may actually underestimate the benefits biodiversity provides to agriculture and human wellbeing," says Taylor Ricketts of the University of Vermont's Gund Institute for Environment. "These positive interactions mean ecosystem services are more valuable together than separately."

For the experiment, researchers from Latin America and the U.S. manipulated coffee plants across 30 farms, excluding birds and bees with a combination of large nets and small lace bags. They tested for four key scenarios: bird activity alone (pest control), bee activity alone (pollination), no bird and bee activity at all, and finally, a natural environment, where bees and birds were free to pollinate and eat insects like the coffee berry borer, one of the most damaging pests affecting coffee production worldwide.

The combined positive effects of birds and bees on fruit set, fruit weight, and fruit uniformity -- key factors in quality and price -- were greater than their individual effects, the study shows. Without birds and bees, the average yield declined nearly 25%, valued at roughly $1,066 per hectare.

"One important reason we measure these contributions is to help protect and conserve the many species that we depend on, and sometimes take for granted," says Natalia Aristizábal, a PhD candidate at UVM's Gund Institute for Environment and Rubenstein School of Environment and Natural Resources. "Birds, bees, and millions of other species support our lives and livelihoods, but face threats like habitat destruction and climate change."

One of the most surprising aspects of the study was that many birds providing pest control to coffee plants in Costa Rica had migrated thousands of miles from Canada and the U.S., including Vermont, where the UVM team is based. The team is also studying how changing farm landscapes impact birds' and bees' ability to deliver benefits to coffee production. They are supported by the U.S. Fish and Wildlife Service through the Neotropical Migratory Bird Conservation Act.

In addition to Martínez-Salinas (Nicaragua), Ricketts (USA), Aristizábal (Colombia), the international research team from CATIE included Adina Chain-Guadarrama (México), Sergio Vilchez Mendoza (Nicaragua), and Rolando Cerda (Bolivia).

           https://www.sciencedaily.com/releases/2022/04/220404152702.htm

 

The Planet Mercury Has Magnetic Storms

An international team of scientists has proved that Mercury, our solar system’s smallest planet, has geomagnetic storms similar to those on Earth.

From:  University of Alaska Fairbanks

By Rod Boyce

March 29, 2022 -- The research by scientists in the United States, Canada and China includes work by Hui Zhang, a space physics professor at the University of Alaska Fairbanks Geophysical Institute.

Their finding, a first, answers the question of whether other planets, including those outside our solar system, can have geomagnetic storms regardless of the size of their magnetosphere or whether they have an Earth-like ionosphere. 

The research was published in two papers in February. Zhang is among the co-authors of each paper.

The first of those papers proves the planet has a ring current, a doughnut-shaped field of charged particles flowing laterally around the planet and excluding the poles. The second proves the existence of geomagnetic storms triggered by the ring current.

A geomagnetic storm is a major disturbance in a planet’s magnetosphere caused by the transfer of energy from the solar wind. Such storms in Earth’s magnetosphere produce the aurora and can disrupt radio communications.

The geomagnetic storms finding was published Feb. 18 in the journal Science China Technological Sciences. QiuGang Zong of the Institute of Space Physics and Applied Technology at Peking University and the Polar Research Institute of China is the author.

That paper built on a finding published one day earlier that verified through data observation earlier suggestions that Mercury has a ring current. Earth also has a ring current.

The ring current paper, published in Nature Communications, is authored by Jiutong Zhao, also of the Institute of Space Physics and Applied Technology at Peking University.

Seven of the 14 scientists involved worked on both papers.

“The processes are quite similar to here on Earth,’ Zhang said of Mercury’s magnetic storms. “The main differences are the size of the planet and Mercury has a weak magnetic field and virtually no atmosphere.”

Confirmation about geomagnetic storms on Mercury results from research made possible by a fortuitous coincidence: a series of coronal mass ejections from the sun on April 8-18, 2015, and the end of NASA’s Messenger space probe, which launched in 2004 and crashed into the planet’s surface on April 30, 2015, at the expected end of its mission.

A coronal mass ejection, or CME, is an ejected cloud of the sun’s plasma — a gas made of charged particles. That cloud includes the plasma’s embedded magnetic field.

The coronal mass ejection of April 14 proved to be the key for scientists. It compressed Mercury’s ring current on the sun-facing side and increased the current’s energy.

New analysis of data from Messenger, which had dropped closer to the planet, shows “the presence of a ring current intensification that is essential for triggering magnetic storms,” the second of the two papers reads.

“The sudden intensification of a ring current causes the main phase of a magnetic storm,” Zhang said.

But this doesn’t mean Mercury has auroral displays like those on Earth. 

On Earth, the storms produce aurora displays when solar wind particles interact with the particles of the atmosphere. On Mercury, however, solar wind particles don’t encounter an atmosphere. Instead, they reach the surface unimpeded and may therefore be visible only through X-ray and gamma ray examination.

The results of the two papers show that magnetic storms are “potentially a common feature of magnetized planets,” the second of the papers reads.

“The results obtained from Messenger provide a further fascinating insight into Mercury’s place in the evolution of the solar system following the discovery of its intrinsic planetary magnetic field,” it concludes.

Other institutions involved in the research include the University of Alberta, Edmonton; University of Michigan and the Heliophysics Science Division at NASA’s Goddard Space Flight Center.

   https://uaf.edu/news/uaf-researcher-in-papers-that-prove-mercury-has-magnetic-storms.php

 

Draft Human Genome Now Out of Date

Researchers generate the first complete, gapless sequence of a human genome

From:  National Institutes of Health

March 31, 2022 -- Scientists have published the first complete, gapless sequence of a human genome, two decades after the Human Genome Project produced the first draft human genome sequence. According to researchers, having a complete, gap-free sequence of the roughly 3 billion bases (or “letters”) in our DNA is critical for understanding the full spectrum of human genomic variation and for understanding the genetic contributions to certain diseases. The work was done by the Telomere to Telomere (T2T) consortium, which included leadership from researchers at the National Human Genome Research Institute (NHGRI), part of the National Institutes of Health; University of California, Santa Cruz; and University of Washington, Seattle. NHGRI was the primary funder of the study.

Analyses of the complete genome sequence will significantly add to our knowledge of chromosomes, including more accurate maps for five chromosome arms, which opens new lines of research. This helps answer basic biology questions about how chromosomes properly segregate and divide. The T2T consortium used the now-complete genome sequence as a reference to discover more than 2 million additional variants in the human genome. These studies provide more accurate information about the genomic variants within 622 medically relevant genes. 

“Generating a truly complete human genome sequence represents an incredible scientific achievement, providing the first comprehensive view of our DNA blueprint,” said Eric Green, M.D., Ph.D., director of NHGRI. “This foundational information will strengthen the many ongoing efforts to understand all the functional nuances of the human genome, which in turn will empower genetic studies of human disease.” 

The now-complete human genome sequence will be particularly valuable for studies that aim to establish comprehensive views of human genomic variation, or how people’s DNA differs. Such insights are vital for understanding the genetic contributions to certain diseases and for using genome sequence as a routine part of clinical care in the future. Many research groups have already started using a pre-release version of the complete human genome sequence for their research.  

The full sequencing builds upon the work of the Human Genome Project, which mapped about 92% of the genome, and research undertaken since then. Thousands of researchers have developed better laboratory tools, computational methods and strategic approaches to decipher the complex sequence. Six papers encompassing the completed sequence appear in Science(link is external), along with companion papers in several other journals. 

That last 8% includes numerous genes and repetitive DNA and is comparable in size to an entire chromosome. Researchers generated the complete genome sequence using a special cell line that has two identical copies of each chromosome, unlike most human cells, which carry two slightly different copies. The researchers noted that most of the newly added DNA sequences were near the repetitive telomeres (long, trailing ends of each chromosome) and centromeres (dense middle sections of each chromosome).

“Ever since we had the first draft human genome sequence, determining the exact sequence of complex genomic regions has been challenging,” said Evan Eichler, Ph.D., researcher at the University of Washington School of Medicine and T2T consortium co-chair. “I am thrilled that we got the job done. The complete blueprint is going to revolutionize the way we think about human genomic variation, disease and evolution.” 

The cost of sequencing a human genome using “short-read” technologies, which provide several hundred bases of DNA sequence at a time, is only a few hundred dollars, having fallen significantly since the end of the Human Genome Project. However, using these short-read methods alone still leaves some gaps in assembled genome sequences. The massive drop in DNA sequencing costs comes hand-in-hand with increased investments in new DNA sequencing technologies to generate longer DNA sequence reads without compromising the accuracy. 

Over the past decade, two new DNA sequencing technologies emerged that produced much longer sequence reads. The Oxford Nanopore DNA sequencing method can read up to 1 million DNA letters in a single read with modest accuracy, while the PacBio HiFi DNA sequencing method can read about 20,000 letters with nearly perfect accuracy. Researchers in the T2T consortium used both DNA sequencing methods to generate the complete human genome sequence.  

"Using long-read methods, we have made breakthroughs in our understanding of the most difficult, repeat-rich parts of the human genome," says Karen Miga, Ph.D., a co-chair of the T2T consortium whose research group at the University of California, Santa Cruz is funded by NHGRI. “This complete human genome sequence has already provided new insight into genome biology, and I look forward to the next decade of discoveries about these newly revealed regions.” 

According to consortium co-chair Adam Phillippy, Ph.D., whose research group at NHGRI led the finishing effort, sequencing a person’s entire genome should get less expensive and more straightforward in the coming years.  

"In the future, when someone has their genome sequenced, we will be able to identify all of the variants in their DNA and use that information to better guide their healthcare,” Phillippy said. “Truly finishing the human genome sequence was like putting on a new pair of glasses. Now that we can clearly see everything, we are one step closer to understanding what it all means.” 

Many early-career researchers and trainees played pivotal roles, including researchers from Johns Hopkins University, Baltimore; University of Connecticut, Storrs; University of California, Davis; Howard Hughes Medical Institute, Chevy Chase, Maryland; and the National Institute of Standards and Technology, Gaithersburg, Maryland. The package of six papers reporting this accomplishment appears in today’s issue of Science, along with companion papers in several other journals.

https://www.nih.gov/news-events/news-releases/researchers-generate-first-complete-gapless-sequence-human-genome

 

Diagnosing a Stroke Accurately

From:  The Internet

Stroke has a new indicator! They say if you forward this to ten people, you stand a chance of saving one life. Will you send this along? Blood Clots/Stroke - They Now Have a Fourth Indicator, the Tongue:

During a BBQ, a woman stumbled and took a little fall - she assured everyone that she was fine (they offered to call paramedics) ...she said she had just tripped over a brick because of her new shoes.

They got her cleaned up and got her a new plate of food. While she appeared a bit shaken up, Jane went about enjoying herself the rest of the evening.

Jane's husband called later telling everyone that his wife had been taken to the hospital - (at 6:00 PM Jane passed away.) She had suffered a stroke at the BBQ. Had they known how to identify the signs of a stroke, perhaps Jane would be with us today. Some don't die. They end up in a helpless, hopeless condition instead.

It only takes a minute to read this.

A neurologist says that if he can get to a stroke victim within 3 hours he can totally reverse the effects of a stroke...totally. He said the trick was getting a stroke recognized, diagnosed, and then getting the patient medically cared for within 3 hours, which is tough.

>>RECOGNIZING A STROKE<<

Thank God for the sense to remember the '3' steps, STR. Read and Learn!

Sometimes symptoms of a stroke are difficult to identify. Unfortunately, the lack of awareness spells disaster. The stroke victim may suffer severe brain damage when people nearby fail to recognize the symptoms of a stroke.

Now doctors say a bystander can recognize a stroke by asking three simple questions:

S *Ask the individual to SMILE.

T *Ask the person to TALK and SPEAK A SIMPLE SENTENCE (Coherently)

(i.e. Chicken Soup)

R *Ask him or her to RAISE BOTH ARMS.

If he or she has trouble with ANY ONE of these tasks, call emergency number immediately and describe the symptoms to the dispatcher.

New Sign of a Stroke -------- Stick out Your Tongue

NOTE: Another 'sign' of a stroke is this: Ask the person to 'stick' out his tongue. If the tongue is 'crooked', if it goes to one side or the other that is also an indication of a stroke.

A cardiologist says if everyone who gets this e-mail sends it to 10 people; you can bet that at least one life will be saved.