Categories
AI

Wildlife photos are a treasure trove for AI-driven conservation research

If you look at a photograph of leopards, would you be able to tell which two were related based on their spots?

Unless you’re a leopard expert, the answer is most likely not, says Tanya Berger-Wolf, director of the Translational Data Analytics Institute (TDAI) at Ohio State University. But, she says, computers can.

Berger-Wolf and her team are pioneering a new field of study called imageomics. As the name suggests, imageomics uses machine learning to extract biological data from photos and videos of living organisms. Berger-Wolf and her team have recently begun collaborating with researchers studying leopards in India to compare spot patterns of moms and children using algorithms.

“Images have become the most abundant source of information now, and we have the technology, too. We have computer vision machine learning,” says Berger-Wolf. She compares this technology to the invention of the microscope, offering scientists a completely different way to look at wildlife.

Building on TDAI’s open-source platform called Wildbook, which helps wildlife researchers gather and analyze photos, the team is now focusing on generative AI approaches. These programs use existing content to generate meaningful data. In this case, they are attempting to analyze crowdsourced images to make biological traits that humans may naturally miss computable, like the curvature of a fish’s fin — or a leopard’s spots. The algorithms scan images of leopards publicly available online, from social media to digitized museum collections.

In simple terms, the algorithms “quantify the similarity,” she says. The aim is to help wildlife researchers overcome a data deficiency problem and, ultimately, better protect animals at risk of extinction.

Machine learning can be used to identify all the relevant parts of a photo, including wildlife of interest.
Photo by Tanya Berger-Wolf

Ecologists and other wildlife researchers are currently facing a data crunch — it’s tedious, expensive, and time-consuming for people to spend time in the field monitoring animals. Due to these challenges, 20,054 species on the International Union for Conservation of Nature’s (IUCN) Red List of Threatened Species are labeled as “data deficient,” meaning there’s not enough information to make a proper assessment of its risk of extinction. As Berger-Wolf sums it up, “biologists are making decisions without having good data on what we’re losing and how fast.”

The platform started with supervised learning — Berger-Wolf says the computer uses algorithms “simpler than Siri” to count how many animals are in the image, as well as where it was taken and when, which could contribute to metrics like population counts. Not only can AI do this at a much lower cost than hiring people but also at a faster rate. In August 2021, the platform analyzed 17 million images automatically.

There are also barriers that only a computer can seem to overcome. “Humans are not the best ones at figuring out what’s the informative aspect,” she says, noting how humans are biased in how we see nature, focusing mostly on facial features. Instead, AI can scan for features humans would likely miss, like the color range of the wings on a tiger moth. A March 2022 study found that the human eye couldn’t tell male polymorphic wood tiger moth genotypes apart — but moth vision models with ultraviolet light sensitivity could.

“That’s where all the true innovation in all of this is,” Berger-Wolf says. The team is implementing algorithms that create pixel values of patterned animals, like leopards, zebras, and whale sharks, and analyze those hot spots where the pixel values change most — it’s like comparing fingerprints. Having these fingerprints means researchers can track animals non-invasively and without GPS collars, count them to estimate population sizes, understand migration patterns, and more.

As Berger-Wolf points out, population size is the most basic metric of a species’ well-being. The platform scanned 11,000 images of whale sharks to create hot spots and help researchers identify individual whale sharks and track their movement, which led to updated information about their population size. This new data pushed the IUCN to change the conservation status of the whale shark from “vulnerable” to “endangered” in 2016.

There are also algorithms using facial recognition for primates and cats, shown to be about 90 percent accurate, compared to humans being about 42 percent accurate.

Generative AI is still a burgeoning field when it comes to wildlife conservation, but Berger-Wolf is hopeful. For now, the team is cleaning the preliminary data of the leopard hot spots to ensure the results are not data artifacts — or flawed — and are true biologically meaningful information. If meaningful, the data could teach researchers how species are responding to changing habitats and climates and show us where humans can step in to help.

Repost: Original Source and Author Link

Categories
AI

Nvidia AI research takes science fiction one step closer to reality

We are excited to bring Transform 2022 back in-person July 19 and virtually July 20 – 28. Join AI and data leaders for insightful talks and exciting networking opportunities. Register today!


Could AI be taught multiple skills at the same time? Are immersive displays using holography closer to reality than ever? No one can say with any certainty what precisely the future of artificial intelligence (AI) will hold. But one way to get a glimpse is by looking at the research that Nvidia will present at Siggraph 2022, to be held August 8-11.

Nvidia is collaborating with researchers to present 16 papers at Siggraph 2022, spanning multiple research topics that impact the intersection of graphics and AI technologies. 

One paper details innovation with reinforcement learning models, done by researchers from the University of Toronto and UC Berkeley, that could help to teach AI multiple skills at the same time.

Another delves into new techniques to help build large-scale virtual worlds with instant neural graphics primitives. Stepping closer to technologies only seen in science fiction, there is also research on holography that could one day pave the way for new display technology that will enable immersive telepresence.

“Our goal is to do work that’s going to impact the company,” David Luebke, vice president of graphics research at Nvidia, told VentureBeat. “It’s about solving problems where people don’t already know the answer and there is no easy engineering solution, so you have to do research.”

The intersection of research and enterprise AI

The 16 papers that Nvidia is helping to present focus on innovations that impact graphics, which is what the Siggraph show is all about. Luebke noted, however, that nearly all the research is also relevant for AI use outside the graphics field.

“I think of graphics as one of the hardest and most interesting applications of computation,” Luebke said. “So it’s no surprise that AI is revolutionizing graphics and graphics is providing a real showcase for AI.”

Luebke said that the researchers who worked on the reinforcement learning model paper actually view themselves as more in the robotics field than graphics. The model has potential applicability to robots as well as any other AI that needs to learn how to perform multiple actions.

“The thing about graphics is that it’s really, really hard and it’s really, really compelling,” he said. “Siggraph is a place where we showcase our graphics accomplishments, but almost everything we do there is applicable in a broader context as well.”

Computational holography and the future of telepresence

Throughout the COVID-19 pandemic, individuals and organizations around the world suddenly become a lot more familiar with video conferencing technologies like Zoom. There has also been a growing use of virtual reality headset usage, connecting to the emerging concept of the metaverse. The metaverse and telepresence could well one day become significantly more immersive.

One of the papers being presented by Nvidia at Siggraph has to do with a concept known as computational holography. Luebke explained that at a basic level, computational holography is a technique that can construct a three-dimensional scene, where the human eye can focus anywhere within that scene and see the correct thing as if it were really there. The research being presented at Siggraph details some new approaches to computational holography that could one day lead to VR headsets that are dramatically thinner than current options, providing a more immersive and lifelike experience.

“That has been kind of a holy grail for computer graphics for years and years,” Luebke said about the work on computational holography. “This research is showing that you can use computation, including neural networks and AI, to improve the quality of holographic displays that work and look good.”

Looking beyond just the papers being presented at Siggraph, Luebke said that Nvidia research is really interested in telepresence innovations. 

VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Learn more about membership.

Repost: Original Source and Author Link

Categories
AI

A new use for AI: summarizing scientific research for seven-year-olds

Academic writing often has a reputation for being hard to follow, but what if you could use machine learning to summarize arguments in scientific papers so that even a seven-year-old could understand them? That’s the idea behind tl;dr papers — a project that leverages recent advances in AI language processing to simplify science.

Work on the site began two years ago by university friends Yash Dani and Cindy Wu as a way to “learn more about software development,” Dani tells The Verge, but the service went viral on Twitter over the weekend when academics started sharing AI summaries of their research. The AI-generated results are sometimes inaccurate or simplified to the point of idiocy. But just as often, they are satisfyingly and surprisingly concise, cutting through academic jargon to deliver what could be mistaken for child-like wisdom.

Take this summary of a paper by Professor Michelle Ryan, director of the Global Institute for Women’s Leadership at the Australian National University. Ryan has written on the concept of the “glass cliff,” a form of gender discrimination in which women are placed in leadership roles at times when institutions are at their greatest risk of failure. The AI summary of her work? “The glass cliff is a place where a lot of women get put. It’s a bad place to be.”

“It is just excellent,” as Ryan put it.

Ryan tells The Verge the summary was “accurate and pithy,” though it did elide a lot of nuances around the concept. In part, this is because of a crucial caveat: tl;dr papers only analyzes the abstract of a scientific paper, which is itself a condensed version of a researcher’s argument. (Being able to condense an entire paper would be a much greater challenge, though it’s something machine learning researchers are already working on.)

Ryan says that although tl;dr papers is undoubtedly a very fun tool, it also offers “a good illustration of what good science communication should look like.” “I think many of us could write in a way that is more reader-friendly,” she says. “And the target audience of a second-grader is a good place to start.”

Zane Griffin Talley Cooper, a PhD candidate at the Annenberg School for Communication at the University of Pennsylvania, described the AI summaries as “refreshingly transparent.” He used the site to condense a paper he’d written on “data peripheries,” which traces the physical history of materials essential to big data infrastructure. Or, as tl;dr papers put it:

“Big data is stored on hard disk drives. These hard disk drives are made of very small magnets. The magnets are mined out of the ground.“

Cooper says although the tool is a “joke on the surface,” systems like this could have serious applications in teaching and study. AI summarizers could be used by students as a way into complex papers, or they could be incorporated into online journals, automatically producing simplified abstracts for public consumption. “Of course,” says Cooper, this should be only done “if framed properly and with discussion of limitations and what it means (both practically and ethically) to use machine learning as a writing tool.”

These limitations are still being explored by the companies that make these AI systems, even as the software is incorporated into ever-more mainstream tools. tl;dr papers itself was run on GPT-3, which is one of the best-known AI writing tools and is made by OpenAI, a combined research lab and commercial startup that works closely with Microsoft.

Microsoft has used GPT-3 and its ilk to build tools like autocomplete software for coders and recently began offering businesses access to the system as part of its cloud suite. The company says GPT-3 can be used to analyze the sentiment of text, generate ideas for businesses, and — yes — condense documents like the transcripts of meetings or email exchanges. And already, tools similar to GPT-3 are being used in popular services like Google’s Gmail and Docs, which offer AI-powered autocomplete features to users.

But the deployment of these AI-language systems is controversial. Time and time again, it’s been shown that these tools encode and amplify harmful language based on their training data (which is usually just vast volumes of text scraped off the internet). They repeat racist and sexist stereotypes and slurs and may be biased in more subtle ways, too.

A different set of worries stems from the inaccuracy of these systems. These tools only manipulate language on a statistical level: they have no human-equivalent understanding of what they’re “reading,” and this can lead to some very basic mistakes. In one notorious example that surfaced last year, Google search — which uses AI to summarize search topics — provided misleading medical advice to a query asking what to do if someone suffers a seizure. While last December, Amazon’s Alexa responded to a child asking for a fun challenge to do by telling them to touch a penny to the exposed prongs of a plug socket.

The specific danger to life posed by these scenarios is unusual, but they offer vivid illustrations of the structural weaknesses of these models. Jathan Sadowski, a senior research fellow in the Emerging Technologies Research Lab at Monash University, was another academic entertained by tl;dr papers’ summary of his research. He says AI systems like this should be handled with care, but they can serve a purpose in the right context.

“Maybe one day [this technology will] be so sophisticated that it can be this automated research assistant who is going and providing you a perfect, accurate, high quality annotated bibliography of academic literature while you sleep. But we are extremely far from that point right now,” Sadowski told The Verge. “The real, immediate usefulness from the tool is — first and foremost — as a novelty and joke. But more practically, I could see it as a creativity catalyst. Something that provides you this alien perspective on your work.”

Sadowski says the summaries provided by tl;dr papers often have a sort of “accidental wisdom” to them — a byproduct, perhaps, of machine learning’s inability to fully understand language. In other scenarios, artists have used these AI tools to write books and music, and Sadowski says a machine’s perspective could be useful for academics who’ve burrowed too deep in their subject. “It can give you artificial distance from a thing you’ve spent a lot of time really close to, that way you can maybe see it in a different light,” he says.

In this way, AI systems like tl;dr papers might even find a place similar to tools designed to promote creativity. Take, for example, “Oblique Strategies,” a deck of cards created by Brian Eno and Peter Schmidt. It offers pithy advice to struggling artists like “ask your body” or “try faking it!” Are these words of wisdom imbued with deep intelligence? Maybe, maybe not. But their primary role is to provoke the reader into new patterns of thinking. AI could offer similar services, and indeed, some companies already sell AI creative writing assistants.

Unfortunately, although tl;dr papers has had a rapturous reception among the academic world, its time in the spotlight looks limited. After going viral this weekend, the website has been labeled “under maintenance,” and the site’s creators say they have no plans to maintain it in the future. (They also mention that other tools have been built that perform the same task.)

Dani told The Verge that tl;dr papers “was designed to be an experiment to see if we can make learning about science a little easier, more fun, and engaging.” He says: “I appreciate all of the attention the app has received and thank all of the people who have tried it out [but] given this was always intended to be an educational project, I plan to sunset tl;dr papers in the coming days to focus on exploring new things.”



Repost: Original Source and Author Link

Categories
AI

Google Research changes the game for medical imaging with self-supervised learning

Deep learning shows a lot of promise in health care, especially in medical imaging, where it can be utilized to improve the speed and accuracy of diagnosing patient conditions. But it also faces a serious barrier: the shortage of labeled training data.

In medical contexts, training data comes at great costs, which makes it very difficult to use deep learning for many applications.

To overcome this hurdle, scientists have explored several solutions to various degrees of success. In a new paper, artificial intelligence researchers at Google suggest a new technique that uses self-supervised learning to train deep learning models for medical imaging. Early results show that the technique can reduce the need for annotated data and improve the performance of deep learning models in medical applications.

Supervised pretraining

Convolutional neural networks have proven to be very efficient at computer vision tasks. Google is one of several organizations that has been exploring its use in medical imaging. In recent years, the company’s research arm has built several medical imaging models in domains like ophthalmology, dermatology, mammography, and pathology.

“There is a lot of excitement around applying deep learning to health, but it remains challenging because highly accurate and robust DL models are needed in an area like health care,” said Shekoofeh Azizi, AI resident at Google Research and lead author of the self-supervised paper.

One of the key challenges of deep learning is the need for huge amounts of annotated data. Large neural networks require millions of labeled examples to reach optimal accuracy. In medical settings, data labeling is a complicated and costly endeavor.

“Acquiring these ‘labels’ in medical settings is challenging for a variety of reasons: it can be time-consuming and expensive for clinical experts, and data must meet relevant privacy requirements before being shared,” Azizi said.

For some conditions, examples are scarce, to begin with, and in others, such as breast cancer screening, it may take many years for the clinical outcomes to manifest after a medical image is taken.

Further complicating the data requirements of medical imaging applications are distribution shifts between training data and deployment environments, such as changes in the patient population, disease prevalence or presentation, and the medical technology used for imaging acquisition, Azizi added.

One popular way to address the shortage of medical data is to use supervised pretraining. In this approach, a convolutional neural network is initially trained on a dataset of labeled images, such as ImageNet. This phase tunes the parameters of the model’s layers to the general patterns found in all kinds of images. The trained deep learning model can then be fine-tuned on a limited set of labeled examples for the target task.

Several studies have shown supervised pretraining to be helpful in applications such as medical imaging, where labeled data is scarce. However, supervised pretraining also has its limits.

“The common paradigm for training medical imaging models is transfer learning, where models are first pretrained using supervised learning on ImageNet. However, there is a large domain shift between natural images in ImageNet and medical images, and previous research has shown such supervised pretraining on ImageNet may not be optimal for developing medical imaging models,” Azizi said.

Self-supervised pretraining

Self-supervised learning has emerged as a promising area of research in recent years. In self-supervised learning, the deep learning models learn the representations of the training data without the need for labels. If done right, self-supervised learning can be of great advantage in domains where labeled data is scarce and unlabeled data is abundant.

Outside of medical settings, Google has developed several self-supervised learning techniques to train neural networks for computer vision tasks. Among them is the Simple Framework for Contrastive Learning (SimCLR), which was presented at the ICML 2020 conference. Contrastive learning uses different crops and variations of the same image to train a neural network until it learns representations that are robust to changes.

In their new work, the Google Research team used a variation of the SimCLR framework called Multi-Instance Contrastive Learning (MICLe), which learns stronger representations by using multiple images of the same condition. This is often the case in medical datasets, where there are multiple images of the same patient, though the images might not be annotated for supervised learning.

“Unlabeled data is often available in large quantities in various medical domains. One important difference is that we utilize multiple views of the underlying pathology commonly present in medical imaging datasets to construct image pairs for contrastive self-supervised learning,” Azizi said.

When a self-supervised deep learning model is trained on different viewing angles of the same target, it learns more representations that are more robust to changes in viewpoint, imaging conditions, and other factors that might negatively affect its performance.

Putting it all together

The self-supervised learning framework the Google researchers used involved three steps. First, the target neural network was trained on examples from the ImageNet dataset using SimCLR. Next, the model was further trained using MICLe on a medical dataset that has multiple images for each patient. Finally, the model is fine-tuned on a limited dataset of labeled images for the target application.

The researchers tested the framework on two dermatology and chest x-ray interpretation tasks. When compared to supervised pretraining, the self-supervised method provides a significant improvement in the accuracy, label efficiency, and out-of-distribution generalization of medical imaging models, which is especially important for clinical applications. Plus, it requires much less labeled data.

“Using self-supervised learning, we show that we can significantly reduce the need for expensive annotated data to build medical image classification models,” Azizi said. In particular, on the dermatology task, they were able to train the neural networks to match the baseline model performance while using only a fifth of the annotated data.

“This hopefully translates to significant cost and time savings for developing medical AI models. We hope this method will inspire explorations in new health care applications where acquiring annotated data has been challenging,” Azizi said.

Ben Dickson is a software engineer and the founder of TechTalks. He writes about technology, business, and politics.

This story originally appeared on Bdtechtalks.com. Copyright 2021

VentureBeat

VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative technology and transact.

Our site delivers essential information on data technologies and strategies to guide you as you lead your organizations. We invite you to become a member of our community, to access:

  • up-to-date information on the subjects of interest to you
  • our newsletters
  • gated thought-leader content and discounted access to our prized events, such as Transform 2021: Learn More
  • networking features, and more

Become a member

Repost: Original Source and Author Link

Categories
AI

DeepMind takes next step in robotics research

DeepMind is mostly known for its work in deep reinforcement learning, especially in mastering complicated games and predicting protein structures. Now, it is taking its next step in robotics research.

According to a blog post on DeepMind’s website, the company has acquired the rigid-body physics simulator MuJoCo and has made it freely available to the research community. MuJoCo is now one of several open-source platforms for training artificial intelligence agents used in robotics applications. Its free availability will have a positive impact on the work of scientists who are struggling with the costs of robotics research. It can also be an important factor for DeepMind’s future, both as a science lab seeking artificial general intelligence and as a business unit of one of the largest tech companies in the world.

Simulating the real world

Simulation platforms are a big deal in robotics. Training and testing robots in the real world is expensive and slow. Simulated environments, on the other hand, allow researchers to train multiple AI agents in parallel and at speeds that are much faster than real life. Today, most robotics research teams carry out the bulk of training their AI models in simulated environments. The trained models are then tested and further fine-tuned on real physical robots.

The past few years have seen the launch of several simulation environments for reinforcement learning and robotics.

MuJoCo, which stands for Multi-Joint Dynamics with Contact, is not the only game in town. There are other physics simulators such as PyBullet, Roboschool, and Isaac Gym. But what makes MuJoCo stand out from others is the fine-grained detail that has gone into simulating contact surfaces. MuJoCo performs a more accurate modeling of the laws of physics, which is shown in the emergence of physical phenomena such as Newton’s Cradle.

MuJoCo also has built-in features that support the simulation of musculoskeletal models of humans and animals, which is especially important in bipedal and quadruped robots.

The increased accuracy of the physics environment can help reduce the differences between the simulated environment and the real world. Called the “sim2real gap,” these differences cause a degradation in the performance of the AI models when they are transferred from simulation to the real world. A smaller sim2real gap reduces the need for adjustments in the physical world.

Making MuJoCo available for free

Before DeepMind open-sourced MuJuCo, many researchers were frustrated with its license costs and opted to use the free PyBullet platform. In 2017, OpenAI released Roboschool, a license-free alternative to MuJoCo, for Gym, its toolkit for training deep reinforcement learning models for robotics and other applications.

“After we launched Gym, one issue we heard from many users was that the MuJoCo component required a paid license … Roboschool removes this constraint, letting everyone conduct research regardless of their budget,” OpenAI wrote in a blog post.

A more recent paper by researchers in Cardiff University states that “The cost of a Mujoco institutional license is at least $3000 per year, which is often unaffordable for many small research teams, especially when a long-term project depends on it.”

DeepMind’s blog refers to a recent article in PNAS that discusses the use of simulation in robotics. The authors recommend better support for the development of open-source simulation platforms and write, “A robust and feature-rich set of four or five simulation tools available in the open-source domain is critical to advancing the state of the art in robotics.”

“In line with these aims, we’re committed to developing and maintaining MuJoCo as a free, open-source, community-driven project with best-in-class capabilities,” DeepMind’s blog post states.

It is worth noting, however, that license fees account for a very small part of the costs of training AI models for robots. The computational costs of robotics research tend to rise along with the complexity of the application.

MuJoCo only runs on CPUs, according to its documentation. It hasn’t been designed to leverage the power of GPUs, which have many more computation cores than traditional processors.

A recent paper by researchers at the University of Toronto, Nvidia, and other organizations highlights the limits of simulation platforms that work on CPUs only. For example, Dactyl, a robotic hand developed by OpenAI, was trained on a compute cluster comprising around 30,000 CPU cores. These kinds of costs remain a challenge with CPU-based platforms such as MuJoCo.

DeepMind’s view on intelligence

DeepMind’s mission is to develop artificial general intelligence (AGI), the flexible kind of innate and learned problem-solving capabilities found in humans and animals. While the path to AGI (and whether we will ever reach it or not) is hotly debated among scientists, DeepMind has a clearly expressed view on it.

In a paper published earlier this year, some of DeepMind’s top scientists suggested that “reward is enough” to reach AGI. According to DeepMind’s scientists, if you have a complex environment, a well-defined reward, and a good reinforcement learning algorithm, you can develop AI agents that will acquire the traits of general intelligence. Richard Sutton, who is among the co-authors of the paper, is one of the pioneers of reinforcement learning and describes it as “the first computational theory of intelligence.”

The acquisition of MuJoCo can provide DeepMind with a powerful tool to test this hypothesis and gradually build on top of its results. By making it available to small research teams, DeepMind can also help nurture talent it will hire in the future.

MuJoCo can also boost DeepMind’s efforts to turn in profits for its parent company, Alphabet. In 2020, the AI lab recorded its first profit after six years of sizable costs for Alphabet. DeepMind is already home to some of the brightest scientists in AI. And with autonomous mobile robots such as Boston Dynamics’ Spot slowly finding their market, DeepMind might be able to develop a business model that serves both its scientific goal and its owner’s interests.

Ben Dickson is a software engineer and the founder of TechTalks. He writes about technology, business, and politics.

This story originally appeared on Bdtechtalks.com. Copyright 2021

VentureBeat

VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative technology and transact.

Our site delivers essential information on data technologies and strategies to guide you as you lead your organizations. We invite you to become a member of our community, to access:

  • up-to-date information on the subjects of interest to you
  • our newsletters
  • gated thought-leader content and discounted access to our prized events, such as Transform 2021: Learn More
  • networking features, and more

Become a member

Repost: Original Source and Author Link

Categories
AI

Meet the AI research pioneer who wants to redefine ‘progress’

All the sessions from Transform 2021 are available on-demand now. Watch now.


Women in the AI field are making research breakthroughs, spearheading vital ethical discussions, and inspiring the next generation of AI professionals. We created the VentureBeat Women in AI Awards to emphasize the importance of their voices, work, and experience and to shine a light on some of these leaders. In this series, publishing Fridays, we’re diving deeper into conversations with this year’s winners, whom we honored recently at Transform 2021. Check out last week’s interview with a winner of our AI responsibility and ethics award.

Think of an AI technology, and Dr. Nuria Oliver was likely working on it decades ago when it still felt like science fiction. Her research and inventions have ignited advancements across the industry, and now drive many of the products and services we use every day.

But while Oliver, the winner of our AI Research Award, has published more than 150 scientific papers and earned 41 patents, she doesn’t believe in technology advancement for the sake of it. Above all, she is today focused on responsible AI and “developing technology that’s on our side, that really has our interests and our well being as the main objective function.”

“To me, progress is an improvement to the quality of life for all people, all the beings on the planet, and the planet itself — not just some people,” she told VentureBeat. “So I think it’s very important before we invest in any technology, to think whether that development is continuing progress. Or if it’s not, maybe we shouldn’t do it.”

Oliver is activating this belief beyond just her own research, speaking regularly on the topic and also creating the Institute for Humanity Centric AI, a non-profit focused on the impact of AI. She’s also leading efforts to bring more women into the industry, and asks any young girls who may be reading this to consider the opportunities in the field. Oliver herself was the first woman computer scientist in Spain to be named an ACM Distinguished Scientist and an ACM Fellow. She was also the first woman scientific director of R&D at Telefonica, and continues making waves today as the chief scientific advisor of Vodafone Institute.

We’re thrilled to offer Oliver this much-deserved award. We recently caught up with her to learn more about her research and discuss responsible AI, the challenges in the industry, and how business leaders can make sense of the quickly evolving field.

This interview has been edited for brevity and clarity.

VentureBeat: How did you become an AI researcher? And what interests you most about the work?

Dr. Nuria Oliver: I discovered AI when I was studying telecommunications engineering in Spain. It’s a six year degree, and when I was in the third or fourth year, a professor from the math department asked me to write a paper for an international conference. I chose to write about neural networks and human intelligence versus artificial intelligence, and I became fascinated with the topic. And so I decided to do my master’s thesis project on computer vision. My PhD in the U.S. was also on AI. So I guess it all started in my third year of university, but I think before that what really fascinated me and still fascinates me about AI is also human intelligence.

VentureBeat: Of all your inventions and research, is there one that sticks out to you as the most impactful for the field of AI? Or the most impactful in another way?

Oliver: That’s like asking someone if they have a preferred child. But I guess my main area of expertise is building computational models of human behavior and building intelligent interactive systems that understand humans. And in terms of a landmark project, I would say the work I did on modeling human interactions using machine learning techniques, because that was one of the early works on detecting and modeling human interactions. I also did a system that was able to predict the most likely maneuver in a car before anyone was talking about autonomous driving — like back in 19999. So that was also a really complex but very exciting project.

I’m also proud of the first project I did at MIT, which was a real-time facial expression recognition system. That commercially exists today, but it was like science fiction back then in 1995. All the work I’ve done on the intersection between mobile phones, health, and wellness has also been really exciting, because it was sort of trying to really change the way we perceived phones. A lot of that work has also become mainstream today with wearables. And then finally, I would say all the work I’ve done on using data and AI for social good. That’s an area that I’m very passionate about, and I feel it’s had a lot of impact. I created the area for using data and AI for social good at Telefonica, and I created the area at Vodafone.

VentureBeat: Well that’s an amazing body of work, and it sounds like you’re always ahead of the time. So what are you working on now that we might see more of in the future? Is there any emerging area of research that you really have your eye on right now?

Oliver: I’m very interested in developing technology that’s on our side, that really has our interests and our well being as the main objective function. And this is not the case today. Why don’t we design technology that suggests we turn it off if it’s having a negative impact on us? Why is the expectation that the technology we use is designed to maximize the amount of time that we spend using it? I’m also working a lot on some of the key challenges of AI systems that are used for decision making: algorithmic bias, discrimination, opacity, violations of privacy, the subliminal manipulation of human behavior. Right now, I don’t think the impact is necessarily positive. So that’s a big area of focus right now of my work, and I recently created a nonprofit foundation called the Institute for Humanity Centric AI. A lot of the work I just described is part of the research agenda of this new foundation we just created.

VentureBeat: You mentioned some of the big ones like bias and privacy, but I’m wondering what you think are some of the lesser known hurdles with AI research today.

Oliver: There are different types of challenges. This is a very active research area, so there are a lot of technical challenges. In addition to what we already said, there’s inferring causality versus correlations. For a lot of big, important problems, we want to understand the causal relationships between different factors, but that is very difficult to do with many of today’s methods, which are very good at finding correlations but not necessarily causation. There are challenges related to data access and combining data from different sources. And for many impactful use cases like helping with a natural disaster or even the pandemic, you want to be able to make decisions in real time.

And then there are more human-related issues in terms of education and capacity building. I’ve been saying for like 10 years now that we should really transform the compulsory education system so it’s more aligned with the 21st century. I think the education system in many countries is from the second industrial revolution, but we’re in the fourth industrial revolution. I also think we need to invest more in developing human skills that have been very important for our own survival: our social intelligence, emotional intelligence, creativity, our ability to work together, to adapt. And beyond the formal education, I think it’s very important to invest in upskilling and reskilling programs for professionals whose jobs are being impacted by AI. And I think there’s a connection there with some of the other VentureBeat awards like the AI Mentorship Award Katia Walsh won. And then also investing in education for the general population and policymakers, so we can actually make informed decisions about this very important discipline of AI.

And I mentioned it briefly, but there are many challenges related to the data: accessing, sharing, analyzing, ensuring quality, and privacy implications. Because even if the data is non-personal data, you can infer personal attributes like political views, sexual orientation, gender, or age. And of course, there are many barriers related to the governance of these systems and the ethical frameworks necessary to make sure the huge power AI has is going to actually be used for social good. I always say we shouldn’t confuse technological development with progress.

VentureBeat: There are new AI papers and findings coming out every day, and like you said, advancements aren’t always progress. So what advice do you have for technical professionals and decision makers for how they can keep up, understand changes in the field, and parse what research is truly impactful?

That’s a very good question because the field has grown exponentially to the point where papers are being published constantly. And in fact, many influential papers aren’t even published in scientific conferences anymore; they’re published in open repository systems like arXiv without any peer review. So I think it’s important to understand that this work is incremental. If you’re a practitioner or a business leader, understand the main concepts and both the capabilities and limitations of existing AI systems. Try to think of how they can benefit your business without necessarily maybe going into all the details of the latest papers.

VentureBeat: Throughout the conversation, we’ve been touching on this idea of responsible and ethical AI. What do you feel is the role of AI researchers in regards to this and preventing the potential harms of these technologies? How is the responsibility the same or different from that of entrepreneurs and enterprises?

Oliver: Increasingly, leading machine learning conferences are asking for a clear ethical discussion on the implications of the work. So that’s really a step in the right direction. Many universities are now including ethics in the computer science degree as well. My main message here would be that if you’re using AI, develop a human-centric approach from the beginning. Take the direction the field and legislation are going into account. I think Europe is recognizing that if there is no regulation of AI systems, the negative unintended consequences of these systems can be pretty bad. And as I said, you know, we might not have progress at all.

VentureBeat

VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative technology and transact.

Our site delivers essential information on data technologies and strategies to guide you as you lead your organizations. We invite you to become a member of our community, to access:

  • up-to-date information on the subjects of interest to you
  • our newsletters
  • gated thought-leader content and discounted access to our prized events, such as Transform 2021: Learn More
  • networking features, and more

Become a member

Repost: Original Source and Author Link

Categories
AI

OpenAI disbands its robotics research team

All the sessions from Transform 2021 are available on-demand now. Watch now.


OpenAI has disbanded its robotics team after years of research into machines that can learn to perform tasks like solving a Rubik’s Cube. Company cofounder Wojciech Zaremba quietly revealed on a podcast hosted by startup Weights & Biases that OpenAI has shifted its focus to other domains, where data is more readily available.

“So it turns out that we can make a gigantic progress whenever we have access to data. And I kept all of our machinery unsupervised, [using] reinforcement learning — [it] work[s] extremely well. There [are] actually plenty of domains that are very, very rich with data. And ultimately that was holding us back in terms of robotics,” Zaremba said. “The decision [to disband the robotics team] was quite hard for me. But I got the realization some time ago that actually, that’s for the best from the perspective of the company.”

In a statement, an OpenAI spokesperson told VentureBeat: “After advancing the state of the art in reinforcement learning through our Rubik’s Cube project and other initiatives, last October we decided not to pursue further robotics research and instead refocus the team on other projects. Because of the rapid progress in AI and its capabilities, we’ve found that other approaches, such as reinforcement learning with human feedback, lead to faster progress in our reinforcement learning research.”

OpenAI first widely demonstrated its robotics work in October 2019, when it published research detailing a five-fingered robotic hand guided by an AI model with 13,000 years of cumulative experience. The best-performing system could successfully unscramble Rubik’s Cubes about 20% to 60% of the time, which might not seem especially impressive. But the model notably discovered techniques to recover from challenges, like when the robot’s fingers were tied together and when the hand was wearing a leather glove.

This was the culmination of over two years of work. In May 2017, OpenAI released Roboschool, open source software for controlling robotics in simulation. That same year, the company said it had created a robotics system, trained entirely in simulation and deployed on a physical robot, that could learn a new task after seeing it done once. And in 2018, OpenAI made available simulated robotics environments and a baseline implementation of Hindsight Experience Replay, a reinforcement learning algorithm that can learn from failure.

“The sad thing is, if we were a robotics company, the mission of the company would be different, and I think we would continue. I quite strongly in the approach that [the] robotics [team] took and the direction,” Zaremba added. “But from the perspective of what we want to achieve, which is to build [artificial general intelligence], there were some components missing.”

Artificial general intelligence

OpenAI has long asserted that immense computational horsepower is a necessary step on the road to artificial general intelligence (AGI), or AI that can learn any task a human can. While luminaries like Mila founder Yoshua Bengio and Facebook VP and chief AI scientist Yann LeCun argue that AGI can’t exist, OpenAI’s cofounders and backers — among them Greg Brockman, chief scientist Ilya Sutskever, Elon Musk, Reid Hoffman, and former Y Combinator president Sam Altman — believe powerful computers in conjunction with reinforcement learning, pretraining, and other techniques can achieve paradigm-shifting AI advances.

As MIT Technology Review reported in 2020, a team within OpenAI called Foresight runs experiments to test how far they can push AI capabilities by training algorithms with increasingly large amounts of data and compute. According to that same report, OpenAI is developing a system trained on images, text, and other data using massive computational resources that the company’s leadership believes is the most promising path toward AGI.

One of the fruits of this effort is DALL-E, a text-to-image engine that’s essentially a visual idea generator. Given a text prompt, the OpenAI system generates images to match the prompt, filling in the blanks when the prompt implies the image must contain a detail that isn’t explicitly stated. DALL-E can combine disparate ideas to synthesize objects, some of which are unlikely to exist in the real world — like a hybrid of a snail and a harp.

Brockman and Altman in particular believe AGI will be able to master more fields than any one person, chiefly by identifying complex cross-disciplinary connections that elude human experts. Furthermore, they predict that responsibly deployed AGI — in other words, AGI deployed in “close collaboration” with researchers in relevant fields, like social science — might help solve longstanding challenges in climate change, health care, and education.

Zaremba asserts that pretraining is a particularly powerful technique in the creation of large, sophisticated AI systems. At a high level, pretraining helps the model learn general features that can be reused on the target task to boost its accuracy. Pretraining was used to develop OpenAI’s Codex, a model that’s trained on billions of lines of public code to power Copilot, GitHub’s service that provides suggestions for whole lines of code inside development environments like Microsoft Visual Studio. Codex is a fine-tuned version of OpenAI’s GPT-3, a language model pretrained on over a trillion words from websites, books, Wikipedia, and other web sources.

“When we created robotics [systems], we thought that we could go very far with self-generated data and reinforcement learning. At the moment, I believe that pretraining [gives] model[s] 100 times cheaper ‘IQ points,’” Zaremba said. “That might be followed with other techniques.”

Commercial realities

OpenAI’s move away from robotics might be a reflection of the economic realities the company faces. DeepMind, the Alphabet-owned AI research lab, has undergone a similar shift in recent years as R&D costs mount, moving away from prestige projects in favor of work with commercial applications, like protein shape prediction.

It’s an open secret that robotics is a capital-intensive field. Industrial robotics company Rethink Robotics closed its doors months after attempting unsuccessfully to find an acquirer. Boston Dynamics, considered among the most advanced robotics firms, was acquired by Google and then sold to SoftBank before Hyundai agreed to buy a controlling stake for $1.1 billion. And Honda retired its Asimo robotics project after over a decade in development.

Roughly a year ago, Microsoft announced it would invest $1 billion in San Francisco-based OpenAI to jointly develop new technologies for Microsoft’s Azure cloud platform. In exchange, OpenAI agreed to license some of its intellectual property to Microsoft, which the company would then package and sell to partners, and to train and run AI models on Azure as OpenAI worked to develop next-generation computing hardware.

In the months that followed, OpenAI released a Microsoft Azure-powered API that allows developers to explore GPT-3’s capabilities.(OpenAI said recently that GPT-3 is now being used in more than 300 different apps by “tens of thousands” of developers and producing 4.5 billion words per day.) Toward the end of 2020, Microsoft announced that it would exclusively license GPT-3 to develop and deliver AI solutions for customers, as well as creating new products that harness the power of natural language generation.

Microsoft recently announced that GPT-3 will be integrated “deeply” with Power Apps, its low-code app development platform — specifically for formula generation. The AI-powered features will allow a user building an ecommerce app, for example, to describe a programming goal using conversational language like “find products where the name starts with ‘kids.’”

As for projects like DALL-E and Jukebox — an AI system that can generate music in any style from scratch, complete with vocals — they also have obvious and immediate business applications. OpenAI predicts that DALL-E could someday augment or even replace 3D rendering engines. For example, architects could use the tool to visualize buildings, while graphic artists could apply it to software and video game design.

VentureBeat

VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative technology and transact.

Our site delivers essential information on data technologies and strategies to guide you as you lead your organizations. We invite you to become a member of our community, to access:

  • up-to-date information on the subjects of interest to you
  • our newsletters
  • gated thought-leader content and discounted access to our prized events, such as Transform 2021: Learn More
  • networking features, and more

Become a member

Repost: Original Source and Author Link

Categories
AI

Data labeling for AI research is highly inconsistent, study finds

Join AI & data leaders at Transform 2021 on July 12th for the AI/ML Automation Technology Summit. Register today.


Supervised machine learning, in which machine learning models learn from labeled training data, is only as good as the quality of that data. In a study published in the journal Quantitative Science Studies, researchers at consultancy Webster Pacific and the University of California, San Diego and Berkeley investigate to what extent best practices around data labeling are followed in AI research papers, focusing on human-labeled data. They found that the types of labeled data range widely from paper to paper and that a “plurality” of the studies they surveyed gave no information about who performed labeling — or where the data came from.

While labeled data is usually equated with ground truth, datasets can — and do — contain errors. The processes used to build them are inherently error-prone, which becomes problematic when these errors reach test sets, the subsets of datasets researchers use to compare progress. A recent MIT paper identified thousands to millions of mislabeled samples in datasets used to train commercial systems. These errors could lead scientists to draw incorrect conclusions about which models perform best in the real world, undermining benchmarks.

The coauthors of the Quantitative Science Studies paper examined 141 AI studies across a range of different disciplines, including social sciences and humanities, biomedical and life sciences, and physical and environmental sciences. Out of all of the papers, 41% tapped an existing human-labeled dataset, 27% produced a novel human-labeled dataset, and 5% didn’t disclose either way. (The remaining 27% used machine-labeled datasets.) Only half of the projects using human-labeled data revealed whether the annotators were given documents or videos containing guidelines, definitions, and examples they could reference as aids. Moreover, there was a “wide variation” in the metrics used to rate whether annotators agreed or disagreed with particular labels, with some papers failing to note this altogether.

Compensation and reproducibility

As a previous study by Cornell and Princeton scientists pointed out, a major venue for crowdsourcing labeling work is Amazon Mechanical Turk, where annotators mostly originate from the U.S. and India. This can lead to an imbalance of cultural and social perspectives. For example, research has found that models trained on ImageNet and Open Images, two large, publicly available image datasets, perform worse on images from Global South countries. Images of grooms are classified with lower accuracy when they come from Ethiopia and Pakistan compared to images of grooms from the U.S.

For annotators, labeling tasks tend to be monotonous and low-paying — ImageNet workers made a median of $2 per hour in wages. Unfortunately, the Quantitative Science Studies survey shows that the AI field leaves the issue of fair compensation largely unaddressed. Most publications didn’t indicate what type of reward they offered to labelers or even include a link to the training dataset.

Beyond doing a disservice to labelers, the lack of links threatens to exacerbate the reproducibility problem in AI. At ICML 2019, 30% of authors failed to submit code with their papers by the start of the conference. And one report found that 60% to 70% of answers given by natural language processing models were embedded somewhere in the benchmark training sets, indicating that the models were often simply memorizing answers.

“Some of the papers we analyzed described in great detail how the people who labeled their dataset were chosen for their expertise, from seasoned medical practitioners diagnosing diseases to youth familiar with social media slang in multiple languages. That said, not all labeling tasks require years of specialized expertise, such as more straightforward tasks we saw, like distinguishing positive versus negative business reviews or identifying different hand gestures,” the coauthors of the Quantitative Science Studies paper wrote. “Even the more seemingly straightforward classification tasks can still have substantial room for ambiguity and error for the inevitable edge cases, which require training and verification processes to ensure a standardized dataset.”

Moving forward

The researchers avoid advocating for a single, one-size-fits-all solution to human data labeling. However, they call for data scientists who choose to reuse datasets to exercise as much caution around the decision as they would if they were labeling the data themselves — lest bias creep in. An earlier version of ImageNet was found to contain photos of naked children, porn actresses, and college parties, all scraped from the web without those individuals’ consent. Another popular dataset, 80 Million Tiny Images, was taken offline after an audit surfaced racist, sexist, and otherwise offensive annotations, such as nearly 2,000 images labeled with the N-word and labels like “rape suspect” and “child molester.”

“We see a role for the classic principle of reproducibility, but for data labeling: does the paper provide enough detail so that another researcher could hypothetically recruit a similar team of labelers, give them the same instructions and training, reconcile disagreements similarly, and have them produce a similarly labeled dataset?” the researchers wrote. “[Our work gives] evidence to the claim that there is substantial and wide variation in the practices around human labeling, training data curation, and research documentation … We call on the institutions of science — publications, funders, disciplinary societies, and educators — to play a major role in working out solutions to these issues of data quality and research documentation.”

VentureBeat

VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative technology and transact.

Our site delivers essential information on data technologies and strategies to guide you as you lead your organizations. We invite you to become a member of our community, to access:

  • up-to-date information on the subjects of interest to you
  • our newsletters
  • gated thought-leader content and discounted access to our prized events, such as Transform 2021: Learn More
  • networking features, and more

Become a member

Repost: Original Source and Author Link

Categories
AI

Nvidia launches $100M supercomputer for U.K. health research

Where does your enterprise stand on the AI adoption curve? Take our AI survey to find out.


Nvidia is launching the $100 million Cambridge-1, the most powerful supercomputer in the United Kingdom, and it is making it available to external researchers in the U.K. health care industry.

The machine will be used for AI research in health care, and it’s one of the world’s fastest supercomputers. Nvidia will make it available to accelerate research in digital biology, genomics, and quantum computing.

Nvidia is collaborating with AstraZeneca, maker of one of the COVID-19 vaccines, to fuel faster drug discoveries and creating a transformer-based generative AI model for chemical structures. Transformer-based neural network architectures, which have become available only in the last several years, allow researchers to leverage massive datasets using self-supervised training methods, avoiding the need for manually labeled examples during pre-training.

Kimberly Powell, vice president of healthcare at Nvidia, said that AstraZeneca, GSK, Guy’s and St Thomas’ NHS Foundation Trust, King’s College London, and Oxford Nanopore Technologies (ONT) are using the supercomputer to develop a deeper understanding of brain diseases like dementia, using AI to design new drugs, and improve the accuracy of finding disease-causing variations in human genomes.

“This is an Nvidia industrial supercomputer owned and operated by Nvidia, and it’s the first one that we’re opening up to public use,” Powell said. “We believe that there is a massive opportunity in the area of health as all the stars have aligned. We’ve been working on simulations for 15 years and AI is having a rapid amount of progress. We know how to build these computers and use them to their maximum capacity better than anyone in the world. And some of the world’s best researchers are in health care.”

I asked Powell if Nvidia was doing this in the hopes of convincing European Union regulators that they should approve Nvidia’s $40 billion acquisition of Cambridge, England-based Arm. But she said the supercomputer project is unrelated to that and it has been in the works for a long time.

At No. 41 on the top 500 supercomputers list, the Cambridge-1 uses an Nvidia DGX SuperPod supercomputing cluster. Nvidia hopes it could have a global impact on health care around the world and contribute to the continuing efforts to fight COVID-19. On top of that, a report by Frontier Economics, an economic consulting firm, estimated that Cambridge-1 will have an economic value of £600 million ($831 million) over the next 10 years.

If it is successful, the Cambridge-1 could be a model for other industries or supercomputers in other regions as well. It’s like a reference design or showroom where Nvidia can show off the best of its technology and get more people to adopt it, Powell said.

Powell said that Nvidia’s CUDA designs and graphics processing unit technology have enabled Moore’s law to progress a million times over the past decade, rather than just a thousand times if Moore’s law was left to itself with the normal evolution of chips. AI models have also grown at an exponential rate with the success of network architectures and the ability to train large language models.

“Over the last 15 years, we’ve literally increased the progress of modeling computational biology by 10 million times,” Powell said. “So that rate of progress is what we’re calling the super exponential. And that gives a sense of why this is applicable in the area of biology and health. And so being able to have this level of computing to work with the leaders in the health care industry is what the supercomputer is all about.”

Data for AI to understand dementia

Above: GPUs in the Nvidia Cambridge-1.

Image Credit: Nvidia

King’s College London and Guy’s and St Thomas’ NHS Foundation Trust are using Cambridge-1 to generate synthetic image databases based on tens of thousands of MRI brain scans, from various ages and diseases. The goal is to use AI to gain a better understanding of diseases like dementia, cancer, and multiple sclerosis, enabling earlier diagnosis and treatment.

Early detection is imperative because existing medicines often are not able to treat these diseases given the severity of the neurological impact. This research will leverage the U.K.’s world-leading health care resources through close collaboration with the National Health Service and the UK Biobank, one of the richest biomedical databases in the world. King’s College London intends to share this dataset with the greater research and startup community.

Ian Abbs, CEO of Guy’s and St Thomas’ NHS Foundation Trust, said in a statement that AI in health care will speed up diagnosis for patients, improve services such as breast cancer screening, and support the way doctors assess risk for patients.

“It’s our investment to collaborate with the world’s leading health care institutions on large-scale computing problems,” Powell said. “Nvidia Cambridge-1 is an industrial supercomputer, and it’s going to be dedicated to AI and health care. What’s really cool about it is it’s built off of the Nvidia DGX SuperPod architecture.”

It has 80 DGX 80GB processors with eight ampere A100 tensor core GPUs with a total of 640 GPUs.

“What’s awesome about this is the DGX SuperPod architecture allows you to build datacenters in a matter of weeks,” Powell said. “The SuperPod architecture is a turnkey AI data center. We’ve already figured out the storage, networking, and compute cooling management tools because we have a digital twin of it called Seline, which is Nvidia’s industrial supercomputer.”

Most supercomputers take months or years to build. Nvidia wants to democratize AI computing for industry research and development, Powell said.

“We’re doing just that in health care. Not only is it a turnkey AI datacenter, it’s really a datacenter as a product,” Powell said. “The other really awesome feature to know about this is that it’s cloud native. And what that means is all the application development that Nvidia or our ecosystem does means that this system is going to get better over time. We can redeploy the full stack on these systems. So it’s going to get better over time.”

Scalable, real-time genomics

The Nvidia Cambridge-1 costs $100 million.

Above: The Nvidia Cambridge-1 costs $100 million.

Image Credit: Nvidia

ONT’s long-read sequencing technology is being used in more than 100 countries to gain genomic insights across a breadth of research areas, from human and plant health to environmental monitoring and antimicrobial resistance.

ONT deploys Nvidia technology in a variety of genomic sequencing platforms in the effort to create AI tools to improve not only the speed but the accuracy of genomic analysis. With access to Cambridge-1, researchers will be able to analyze the DNA samples in hours rather than days. That will help scientists gain more insights than ever before, said Rosemary Sinclair Dokos, vice president of product at ONT, in a statement.

The MegaMolBART drug discovery model being developed by Nvidia and AstraZeneca is slated to be used in reaction prediction, molecular optimization, and de novo molecular generation and will optimize the drug development process. It is based on AstraZeneca’s MolBART transformer model and is being trained on the ZINC chemical compound database — using Nvidia’s Megatron framework to enable massively scaled-out training on supercomputing infrastructure. This model will be open-sourced, available to researchers and developers in the Nvidia NGC software catalog.

Additionally, GSK is working with Nvidia to put their vast data sources to work toward the discovery of medicines and vaccines. GSK will use Cambridge-1 to help discover new therapeutics faster by combining genetic and clinical data for the next generation of drug discovery.

The Cambridge-1 has 80 Nvidia DGX A100 systems integrating Nvidia A100 GPUs, Nvidia BlueField-2 data processing units (DPUs), and Nvidia HDR InfiniBand networking. The Cambridge-1 is an Nvidia DGX SuperPod that delivers more than 400 petaflops of AI performance and 8 petaflops of Linpack performance. The system is located at a facility operated by Nvidia partner Kao Data, and it will use renewable energy.

Cambridge-1 is the first supercomputer Nvidia has dedicated to advancing industry-specific research in the U.K. The company also intends to build an AI Center for Excellence in Cambridge featuring a new Arm-based supercomputer, which will support more industries across the country.

“By improving the front end of the whole process, we’re going to definitely improve our chances of success and drug discovery going forward,” Powell said.

VentureBeat

VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative technology and transact.

Our site delivers essential information on data technologies and strategies to guide you as you lead your organizations. We invite you to become a member of our community, to access:

  • up-to-date information on the subjects of interest to you
  • our newsletters
  • gated thought-leader content and discounted access to our prized events, such as Transform 2021: Learn More
  • networking features, and more

Become a member

Repost: Original Source and Author Link

Categories
AI

Study finds that few major AI research papers consider negative impacts

Where does your enterprise stand on the AI adoption curve? Take our AI survey to find out.


In recent decades, AI has become a pervasive technology, affecting companies across industries and throughout the world. These innovations arise from research, and the research objectives in the AI field are influenced by many factors. Together, these factors shape patterns in what the research accomplishes, as well as who benefits from it — and who doesn’t.

In an effort to document the factors influencing AI research, researchers at Stanford, the University of California, Berkeley, the University of Washington, and University College Dublin & Lero surveyed 100 highly cited studies submitted to two prominent AI conferences, NeurIPS and ICML. They claim that in the papers they analyzed, which were published in 2008, 2009, 2018, and 2019, the dominant values were operationalized in ways that centralize power, disproportionally benefiting corporations while neglecting society’s least advantaged.

“Our analysis of highly influential papers in the discipline finds that they not only favor the needs of research communities and large firms over broader social needs, but also that they take this favoritism for granted,” the coauthors of the paper wrote. “The favoritism manifests in the choice of projects, the lack of consideration of potential negative impacts, and the prioritization and operationalization of values such as performance, generalization, efficiency, and novelty. These values are operationalized in ways that disfavor societal needs, usually without discussion or acknowledgment.”

In the papers they reviewed, the researchers identified “performance,” “building on past work,” “generalization,” “efficiency,” “quantitative evidence,” and “novelty” as the top values espoused by the coauthors. By contrast, values related to user rights and ethical principles appeared very rarely — if at all. None of the papers mentioned autonomy, justice, or respect for persons, and most only justified how the coauthors achieved certain internal, technical goals. Over two-thirds —  71% — didn’t make any mention of societal need or impact, and just 3% made an attempt to identify links connecting their research to societal needs.

One of the papers included a discussion of negative impacts and a second mentioned the possibility. But tellingly, none of the remaining 98 contained any reference to potential negative impacts, according to the Stanford, Berkeley, Washington, and Dublin researchers. Even after NeurIPS mandated that coauthors who submit papers must state the “potential broader impact of their work” on society, beginning with NeurIPS 2020 last year, the language leaned toward positive consequences, often mentioning negative consequences only briefly or not at all.

“We reject the vague conceptualization of the discipline of [AI] as value-neutral,” the researchers wrote. “The upshot is that the discipline of ML is not value-neutral. We find that it is socially and politically loaded, frequently neglecting societal needs and harms, while prioritizing and promoting the concentration of power in the hands of already powerful actors.”

To this end, the researchers found that ties to corporations — either funding or affiliation — in the papers they examined doubled to 79% from 2008 and 2009 to 2018 and 2019. Meanwhile, ties to universities declined to 81%, putting corporations nearly on par with universities for the most-cited AI research.

The trend is partly attributable to private sector poaching. From 2006 to 2014, the proportion of AI publications with a corporate-affiliated author increased from about 0% to 40%, reflecting the growing movement of researchers from academia to corporations.

But whatever the cause, the researchers assert that the effect is the suppression of values such as beneficence, justice, and inclusion.

“The top stated values of [AI] that we presented in this paper such as performance, generalization, and efficiency … enable and facilitate the realization of Big Tech’s objectives,” they wrote. “A ‘state-of-the-art’ large image dataset, for example, is instrumental for large scale models, further benefiting [AI] researchers and big tech in possession of huge computing power. In the current climate where values such as accuracy, efficiency, and scale, as currently defined, are a priority, user safety, informed consent, or participation may be perceived as costly and time consuming, evading social needs.”

A history of inequality

The study is only the latest to argue that the AI industry is built on inequality. In an analysis of publications at two major machine learning conference venues, NeurIPS 2020 and ICML 2020, none of the top 10 countries in terms of publication index were located in Latin America, Africa, or Southeast Asia. A separate report from Georgetown University’s Center for Security and Emerging Technology found that while 42 of the 62 major AI labs are located outside of the U.S., 68% of the staff are located within the United States.

The imbalances can result in harm, particularly given that the AI field generally lacks clear descriptions of bias and fails to explain how, why, and to whom specific bias is harmful. Previous research has found that ImageNet and OpenImages — two large, publicly available image datasets — are U.S.- and Euro-centric. Models trained on these datasets perform worse on images from Global South countries. For example, images of grooms are classified with lower accuracy when they come from Ethiopia and Pakistan, compared to images of grooms from the United States. Along this vein, because of how images of words like “wedding” or “spices” are presented in distinctly different cultures, publicly available object recognition systems fail to correctly classify many of these objects when they come from the Global South.

Initiatives are underway to turn the tide, like Khipu and Black in AI, which aim to increase the number of Latin American and Black scholars attending and publishing at premiere AI conferences. Other communities based on the African continent, like Data Science AfricaMasakhane, and Deep Learning Indaba, have expanded their efforts with conferences, workshops, dissertation awards, and developed curricula for the wider African AI community.

But substantial gaps remain. AI researcher Timnit Gebru was fired from her position on an AI ethics team at Google reportedly in part over a paper that discusses risks associated with deploying large language models, including the impact of their carbon footprint on marginalized communities and their tendency to perpetuate abusive language, hate speech, microaggressions, stereotypes, and other dehumanizing language aimed at specific groups of people. Google-affiliated coauthors later published a paper pushing back against Gebru’s environmental claims.

“We present this paper in part in order to expose the contingency of the present state of the field; it could be otherwise,” the University College Dublin & Lero researchers and their associates wrote. “For individuals, communities, and institutions wading through difficult-to-pin-down values of the field, as well as those striving toward alternative values, it is a useful tool to have a characterization of the way the field is now, for understanding, shaping, dismantling, or transforming what is, and for articulating and bringing about alternative visions.”

VentureBeat

VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative technology and transact.

Our site delivers essential information on data technologies and strategies to guide you as you lead your organizations. We invite you to become a member of our community, to access:

  • up-to-date information on the subjects of interest to you
  • our newsletters
  • gated thought-leader content and discounted access to our prized events, such as Transform 2021: Learn More
  • networking features, and more

Become a member

Repost: Original Source and Author Link