Computer scientists have leveraged deep-learning methods to develop Metabolite Translator, a computational tool that predicts what metabolites will result from interactions between enzymes and small molecules such as drugs.

The tool, created at Rice University’s Brown School of Engineering (US) in the lab of Lydia Kavraki, was designed to enable pharmaceutical companies to better understand how their experimental drugs will perform in vivo in the human body.

Using deep-learning methods and the availability of massive reaction datasets, the team developed a system which they say is not constrained by rules that companies use to determine metabolic reactions, opening a path to novel discoveries.

“When you are trying to determine if a compound is a potential drug, you have to check for toxicity. You want to confirm that it does what it should, but you also want to know what else might happen,” said Kavraki, the Noah Harding Professor of Computer Science, a professor of bioengineering, mechanical engineering and electrical and computer engineering and director of Rice’s Ken Kennedy Institute.

The researchers trained Metabolite Translator to predict metabolites by pre-training it on 900,000 known chemical reactions and then fine-tuning its capabilities with data on human metabolic transformations. As a result, they say it can predict metabolites through any enzyme.

Metabolite Translator is based on SMILES (simplified molecular-input line-entry system), a notation method that uses plain text to represent chemical molecules, rather than diagrams.

To validate Metabolite Translator, they used it to analyse the SMILES sequences of 65 drugs and 179 metabolizing enzymes, then compared the results with those produced by several other predictive techniques. Though Metabolite Translator was trained on a general dataset not specific to drugs, the researchers said it performed as well as commonly used rule-based methods specifically developed for drugs. It also identified enzymes that are not commonly involved in drug metabolism and were not detected by existing methods.

Kavraki concluded: “Using a machine learning-based method, we are training a system to understand human metabolism without the need for explicitly encoding this knowledge in the form of rules. This work would not have been possible two years ago.”

The post Using deep learning to predict drug metabolites and toxicity appeared first on Artificial Intelligence.

Artificial intelligence (AI) is rapidly becoming integral to how organizations are run. This should not be a surprise; when analyzing sales calls and market trends, for example, the judgments of computational algorithms can be considered superior to those of humans. As a result, AI techniques are increasingly used to make decisions. Organizations are employing algorithms to allocate valuable resources, design work schedules, analyze employee performance, and even decide whether employees can stay on the job.

This creates a new set of problems even as it solves old ones. As algorithmic decision-making’s role in calculating the distribution of limited resources increases, and as humans become more dependent on and vulnerable to the decisions of AI, anxieties about fairness are rising. How unbiased can an automated decision-making process with humans as the recipients really be?

To address this issue, computer scientists and engineers are focusing primarily on how to govern the use of data provided to help the algorithm learn (that is, data mining) and how to use guiding principles and techniques that can promote interpretable AI: systems that allow us to understand how the results emerged. Both approaches rely, for the most part, on the development of computational methods that factor in certain features believed to be related to fairness.

At the heart of the problem is the fact that algorithms calculate optimal models from the data they’re given — meaning they can end up replicating the problems they’re meant to correct. A 2014 effort to remove human bias in recruitment at Amazon, for example, rated candidates in gender-biased ways; the historical job performance data it was given showed that the tech industry was dominated by men, so it assessed hiring men to be a good bet. The Correctional Offender Management Profiling for Alternative Sanctions, an AI-run program, offered biased predictions for recidivism that wrongly forecast that Black defendants (incorrectly judged to be at higher risk of recidivism) would reoffend at a much greater rate than white defendants (incorrectly flagged as low-risk).

Organizations and governments have tried to establish guidelines to help AI developers refine technical aspects so that algorithmic decisions will be more interpretable — allowing humans to understand clearly how decisions were reached — and thus fairer. For example, Microsoft has launched programs that identify high-level principles such as fairness, transparency, accountability, and ethicality to guide computer scientists and engineers in their coding efforts. Similar efforts are underway on the government level, as demonstrated by the European Union’s Ethics Guidelines for Trustworthy AI and Singapore’s Model AI Governance Framework.

But neither the efforts of computer scientists to factor in technological features nor the efforts of companies and governments to develop principle-based guidelines quite solves the issue of trust. To do that, designers need to account for the information needs and expectations of the people facing the results of the models’ outputs. This is important ethically and also practically: An abundance of research in management shows that the fairer decisions are perceived to be, the more that employees accept them, cooperate with others, are satisfied with their jobs, and perform better. Fairness matters greatly to organizational functioning, and there’s no reason to think that will change when AI becomes the decision maker.

So, how can businesses that want to implement AI persuade users that they’re not compromising on fairness? Put simply, they need to stop thinking about fairness — a complicated concept — as something they can address with the right automated processes and start thinking about an interdisciplinary approach in which computer and social sciences work together. Fairness is a social construct that humans use to coordinate their interactions and subsequent contributions to the collective good, and it is subjective. An AI decision maker should be evaluated on how well it helps people connect and cooperate; people will consider not only its technical aspects but also the social forces operating around it. An interdisciplinary approach allows for identifying three types of solutions that are usually not discussed in the context of AI as a fair decision maker.

Solution 1: Treat AI fairness as a cooperative act.

Algorithms aim to reduce error rates as much as possible in order to reveal the optimal solution. But while that process can be shaped by formal criteria of fairness, algorithms leave the perceptual nature of fairness out of the equation and do not cover aspects such as whether people feel they have been treated with dignity and respect and have been taken care of — important justice concerns. Indeed, algorithms are largely designed to create optimal prediction models that factor in technical features to enhance formal fairness criteria, such as interpretability and transparency, despite the fact that those features do not necessarily meet the expectations and needs of the human end user. As a result, and as the Amazon example shows, algorithms may predict outcomes that society perceives as unfair.

There’s a simple way to address this problem: The model produced by AI should be evaluated by a human devil’s advocate. Although people are much less rational than machines and are to some extent blind to their own inappropriate behaviors, research shows that they are less likely to be biased when evaluating the behaviors and decisions of others. In view of this insight, the strategy for achieving AI fairness must involve a cooperative act between AI and humans. Both parties can bring their best abilities to the table to create an optimal prediction model adjusted for social norms.

Recommendation: Organizations need to invest significantly in the ethical development of their managers. Being a devil’s advocate for algorithmic decision makers requires managers to develop their common sense and intuitive feel for what is right and wrong.

Solution 2: Regard AI fairness as a negotiation between utility and humanity.

Algorithmic judgment is demonstrated to be more accurate and predictive than human judgment in a range of specific tasks, including the allocation of jobs and rewards on the basis of performance evaluations. It makes sense that in the search for a better-functioning business, algorithms are increasingly preferred over humans for those tasks. From a statistical point of view, that preference may appear valid. However, managing workflow and resource allocation in (almost) perfectly rational and consistent ways is not necessarily the same as building a humane company or society.

No matter how you may try to optimize their workdays, humans don’t work in steady, predictable ways. We have good and bad days, afternoon slumps, and bursts of productivity — all of which presents a challenge for the automated organization of the future. Indeed, if we want to use AI in ways that promote a humane work setting, we have to accept the proposition that we should not optimize the search for utility to the detriment of values such as tolerance for failure, which allows people to learn and improve — leadership abilities considered necessary to making our organizations and society humane. The optimal prediction model of fairness should be designed with a negotiation mindset that strives for an acceptable compromise between utility and humane values.

Recommendation: Leaders need to be clear about what values the company wants to pursue and what moral norms they would like to see at work. They must therefore be clear about how they want to do business and why. Answering those questions will make evident the kind of organization they would like to see in action.

Solution 3: Remember that AI fairness involves perceptions of responsibility.

Fairness is an important concern in most (if not all) of our professional interactions and therefore constitutes an important responsibility for decision makers. So far, organizations and governments — because of their adherence to matrix structures — have tackled the question of fair AI decision-making by developing checklists of qualities to guide the development of algorithms. The goal is to build AIs whose outputs match a certain definition of what’s fair.

That’s only half of the equation, however: AI’s fairness as a decision maker really depends on the choices made by the organization adopting it, which is responsible for the outcomes its algorithms generate. The perceived fairness of the AI will be judged through the lens of the organization employing it, not just by the technical qualities of the algorithms.

Recommendation:An organization’s data scientists need to know and agree with the values and moral norms leadership has established. At most organizations, a gap exists between what data scientists are building and the values and business outcomes organizational leaders want to achieve. The two groups need to work together to understand what values cannot be sacrificed in the use of algorithms. For example, if the inclusiveness of minority groups, which are usually poorly represented in available data, is important to the company, then algorithms need to be developed that include that value as an important filter and ensure that outliers, not just commonalities, are learned from.

The post What Does Building a Fair AI Really Entail? appeared first on Artificial Intelligence.

One curious side-effect of the work to digitize books and historical texts is the ability to search these databases for words, when they first appeared and how their frequency of use has changed over time.

The Google Books n-gram corpus is a good example (an n-gram is a sequence of n words). Enter a word or phrase and it’ll show you its relative usage frequency since 1800. For example, the word “Frankenstein” first appeared in the late 1810s and has grown in popularity ever since.

By contrast, the phrase “Harry Potter” appeared in the late 1990s, gained quickly in popularity but never overtook Frankenstein — or Dracula, for that matter. That may be something of surprise given the unprecedented global popularity of J.K. Rowling’s teenage wizard.

And therein lies the problem with a database founded on an old-fashioned, paper-based technology. The Google Books corpus records “Harry Potter” once for each novel, article and text in which it appears, not for the millions of times it is printed and sold. There is no way to account for this level of fame or how it leaves others in the shade.

Today that changes, thanks to the work of Thayer Alshaabi at the Computational Story Lab at the University of Vermont and a number of colleagues. This team has created a searchable database of over 100 billion tweets in more than 150 languages containing over a trillion 1-grams, 2-grams and 3-grams. That’s about 10 per cent of all Twitter messages since September 2008.

Data Visualization

The team has also developed a data visualization tool called Storywrangler that reveals the popularity of any words or phrases based on the number of times they have been tweeted and retweeted. The database shows how this popularity waxes and wanes over time.

“In building Storywrangler, our primary goal has been to curate and share a rich, language-based ecology of interconnected n-gram time series derived from Twitter,” say Alshaabi and co.

Storywrangler immediately reveals the “story” associated with a wide range of events, individuals and phenomenon. For example, it shows the annual popularity of words associated with religious festivals such as Christmas and Easter. It tells how phrases associated with new films burst into Twittersphere and then fade away, while TV series tend to live on, at least throughout the series’ lifetime. And it reveals the emergence of politico-social movements such as Brexit, Occupy #MeToo and Black Lives Matter.

The storylines can also be compared with other databases to provide more fine-grained insight and analysis. For example, the popularity of film titles on Twitter can be compared with the film’s takings at the box office; the emergence of words associated with disease can be compared with the number of infections recorded by official sources; and words associated with political unrest can be compared with incidents of civil disobedience.

That’s useful because this kind of analysis provides a new way to study society, potentially with predictive results. Indeed, computer scientists have long suggested that social media can be used to predict the future.

Cultural Significance

These storylines have social and cultural significance too. “Our collective memory lies in our recordings — in our written texts, artworks, photographs, audio and video — and in our retellings and reinterpretations of that which becomes history,” say Alshaabi and colleagues.

Now anyone can study it with Storywrangler. Try it, it’s interesting.

As for Harry Potter, Frankenstein and Dracula, the tale that Storywrangler tells is different from the Google Books n-gram corpus. Harry Potter is significantly more popular than his grim-faced predecessors and always has been on Twitter. In 2011, Harry Potter was the 44th most popular term on Twitter while Dracula has never risen higher than 2653rd. Frankenstein’s best rank is 3560th.

The post How Data Mining Visualizes Story Lines in the Twittersphere appeared first on Artificial Intelligence.

The work of MIT computer scientist Aleksander Madry is fueled by one core mission: “doing machine learning the right way.”

Madry’s research centers largely on making machine learning — a type of artificial intelligence — more accurate, efficient, and robust against errors. In his classroom and beyond, he also worries about questions of ethical computing, as we approach an age where artificial intelligence will have great impact on many sectors of society.

“I want society to truly embrace machine learning,” says Madry, a recently tenured professor in the Department of Electrical Engineering and Computer Science. “To do that, we need to figure out how to train models that people can use safely, reliably, and in a way that they understand.”

Interestingly, his work with machine learning dates back only a couple of years, to shortly after he joined MIT in 2015. In that time, his research group has published several critical papers demonstrating that certain models can be easily tricked to produce inaccurate results — and showing how to make them more robust.

In the end, he aims to make each model’s decisions more interpretable by humans, so researchers can peer inside to see where things went awry. At the same time, he wants to enable nonexperts to deploy the improved models in the real world for, say, helping diagnose disease or control driverless cars.

“It’s not just about trying to crack open the machine-learning black box. I want to open it up, see how it works, and pack it back up, so people can use it without needing to understand what’s going on inside,” he says.

For the love of algorithms

Madry was born in Wroclaw, Poland, where he attended the University of Wroclaw as an undergraduate in the mid-2000s. While he harbored interest in computer science and physics, “I actually never thought I’d become a scientist,” he says.

An avid video gamer, Madry initially enrolled in the computer science program with intentions of programming his own games. But in joining friends in a few classes in theoretical computer science and, in particular, theory of algorithms, he fell in love with the material. Algorithm theory aims to find efficient optimization procedures for solving computational problems, which requires tackling difficult mathematical questions. “I realized I enjoy thinking deeply about something and trying to figure it out,” says Madry, who wound up double-majoring in physics and computer science.

When it came to delving deeper into algorithms in graduate school, he went to his first choice: MIT. Here, he worked under both Michel X. Goemans, who was a major figure in applied math and algorithm optimization, and Jonathan A. Kelner, who had just arrived to MIT as a junior faculty working in that field. For his PhD dissertation, Madry developed algorithms that solved a number of longstanding problems in graph algorithms, earning the 2011 George M. Sprowls Doctoral Dissertation Award for the best MIT doctoral thesis in computer science.

After his PhD, Madry spent a year as a postdoc at Microsoft Research New England, before teaching for three years at the Swiss Federal Institute of Technology Lausanne — which Madry calls “the Swiss version of MIT.” But his alma mater kept calling him back: “MIT has the thrilling energy I was missing. It’s in my DNA.”

Getting adversarial

Shortly after joining MIT, Madry found himself swept up in a novel science: machine learning. In particular, he focused on understanding the re-emerging paradigm of deep learning. That’s an artificial-intelligence application that uses multiple computing layers to extract high-level features from raw input — such as using pixel-level data to classify images. MIT’s campus was, at the time, buzzing with new innovations in the domain.

But that begged the question: Was machine learning all hype or solid science? “It seemed to work, but no one actually understood how and why,” Madry says.

Answering that question set his group on a long journey, running experiment after experiment on deep-learning models to understand the underlying principles. A major milestone in this journey was an influential paper they published in 2018, developing a methodology for making machine-learning models more resistant to “adversarial examples.” Adversarial examples are slight perturbations to input data that are imperceptible to humans — such as changing the color of one pixel in an image — but cause a model to make inaccurate predictions. They illuminate a major shortcoming of existing machine-learning tools.

Continuing this line of work, Madry’s group showed that the existence of these mysterious adversarial examples may contribute to how machine-learning models make decisions. In particular, models designed to differentiate images of, say, cats and dogs, make decisions based on features that do not align with how humans make classifications. Simply changing these features can make the model consistently misclassify cats as dogs, without changing anything in the image that’s really meaningful to humans.

Results indicated some models — which may be used to, say, identify abnormalities in medical images or help autonomous cars identify objects in the road — aren’t exactly up to snuff. “People often think these models are superhuman, but they didn’t actually solve the classification problem we intend them to solve,” Madry says. “And their complete vulnerability to adversarial examples was a manifestation of that fact. That was an eye-opening finding.”

That’s why Madry seeks to make machine-learning models more interpretable to humans. New models he’s developed show how much certain pixels in images the system is trained on can influence the system’s predictions. Researchers can then tweak the models to focus on pixels clusters more closely correlated with identifiable features — such as detecting an animal’s snout, ears, and tail. In the end, that will help make the models more humanlike — or “superhumanlike” — in their decisions. To further this work, Madry and his colleagues recently founded the MIT Center for Deployable Machine Learning, a collaborative research effort working toward building machine-learning tools ready for real-world deployment. 

“We want machine learning not just as a toy, but as something you can use in, say, an autonomous car, or health care. Right now, we don’t understand enough to have sufficient confidence in it for those critical applications,” Madry says.

Shaping education and policy

Madry views artificial intelligence and decision making (“AI+D” is one of the three new academic units in the Department of Electrical Engineering and Computer Science) as “the interface of computing that’s going to have the biggest impact on society.”

In that regard, he makes sure to expose his students to the human aspect of computing. In part, that means considering consequences of what they’re building. Often, he says, students will be overly ambitious in creating new technologies, but they haven’t thought through potential ramifications on individuals and society. “Building something cool isn’t a good enough reason to build something,” Madry says. “It’s about thinking about not if we can build something, but if we should build something.”

Madry has also been engaging in conversations about laws and policies to help regulate machine learning. A point of these discussions, he says, is to better understand the costs and benefits of unleashing machine-learning technologies on society.

“Sometimes we overestimate the power of machine learning, thinking it will be our salvation. Sometimes we underestimate the cost it may have on society,” Madry says. “To do machine learning right, there’s still a lot still left to figure out.”

The post “Doing machine learning the right way” appeared first on Artificial Intelligence.

Newswise — The future of commuter traffic probably looks something like this: ride-hailing companies operating fleets of autonomous electric vehicles alongside an increasing number of semi-autonomous EVs co-piloted by humans, all supported by a large infrastructure of charging stations. This scenario is particularly likely in California, which has committed to reducing carbon emissions to 40 percent below 1990 levels by 2030.

Computer scientists at Lawrence Livermore National Laboratory are preparing for this potential future by applying Deep Reinforcement Learning — the same kind of goal-driven algorithms that have defeated video game experts and world champions in the strategy game Go — to determine the most efficient strategy for charging and driving electric vehicles used for ride-hailing services. The goal of the strategy is to maximize service while reducing carbon emissions and the impact to the electrical grid, with an emphasis on autonomous EVs capable of 24-hour service.

In a paper published and presented at the recent NeurIPS 2019 Workshop on Tackling Climate Change with Machine Learning, LLNL computer scientists applied deep reinforcement learning to data gathered from ride-hailing services and utility providers to determine when EV drivers or autonomous electric cars should charge their vehicles and when they should pick up customers. The researchers hope to eventually create a robust tool that could provide ride-hailing drivers or autonomous cars with an optimal driving policy based on surge pricing, wait times at charging stations, carbon emissions released while charging, the current cost of energy and other factors that can change throughout the day.

“This project is a simple environment to train autonomous agents to improve their driving behavior,” said LLNL principal investigator and machine learning researcher Ruben Glatt. “We wanted to build a simulation with input from the ride-sharing and energy data so we could simulate typical rides, including costs and energy implications given a certain location or time. We wanted to know: How can we balance ecological factors like the carbon footprint, which is important for the society, while at the same time optimizing revenue that benefits the individual?”

While EVs are clearly a major step to reducing carbon emissions, there are downsides when compared to combustion engine vehicles, the researchers explained. Currently, drivers who use electric vehicles for ride-hailing companies must constantly weigh numerous options in determining when to offer a ride and when to charge their cars, they said.

“It’s hard to be a ride-share driver with a normal EV because you don’t get as much range with your car when it’s fully charged as you would with a full tank of gas. And waiting times at charging stations can be very high, compared to a couple of minutes to fill your gas car,” said main author and LLNL machine learning researcher Jacob Pettit. “There’s a lot of opportunity cost involved; if you drive for a ride-sharing company you might waste a lot of time just recharging and not providing as much service.”

In training the deep reinforcement learning algorithms, each time the agent (representing an autonomous EV or driver of a shared EV capable of driving 24 hours per day) dropped off a customer, it faced a decision to either charge the vehicle or give a ride to a customer. It was rewarded for successfully completing trips with an expected fare amount and penalized for producing carbon emissions when charging or attempting to provide a ride with insufficient battery power.

The agent learned a beneficial strategy was to charge the vehicle when energy costs are cheap or have low carbon emissions, there is fewer demand for rides and waiting times at charging stations is low. Overall, the agent discovered how to optimize the number of rides it provided (revenue) while at the same time minimizing charging wait times and reducing emissions.

“It learned to look at time of day and extrapolate that, given the time, it wouldn’t have to wait long, and wouldn’t pay much money to charge the car, said co-author and LLNL machine learning researcher Brenden Petersen. “The surprising thing was that although we were mostly optimizing for money, the policy also produced less emissions per mile. Even though the agents were acting selfishly it still helped the environment, which is basically a win-win.”  

The researchers are seeking to collaborate with both ride-hailing companies and energy providers to ensure the infrastructure that will eventually support autonomous EV ride-hailing services will be more stable and increase adoption of EVs generally for such services. They envision a machine-learning based tool that could help utilities and city planners decide where to place future electric vehicle charging stations and build an electrical infrastructure to accommodate autonomous EV traffic.

The team has applied for a Technology Commercialization Fund grant from the Department of Energy to expand the simulation to include multiple agents and alternative scenarios.

“We want to make the simulation closer to real life,” Glatt said. “Here we only investigated for a single agent. We want to see what happens if we can control a fleet of agents and if additional networking effects evolve that we can benefit from.”

The work was funded by the Department of Energy. Former LLNL researcher John Donadee initiated the project and also contributed to the publication.  

The post LLNL computer scientists explore deep learning to improve efficiency of ride-hailing and autonomous electric vehicles appeared first on Artificial Intelligence.

Carnegie Mellon University computer scientists have taken a deep learning method that has revolutionized face recognition and other image-based applications in recent years and redirected its power to explore the relationship between genes.

The trick, they say, is to transform massive amounts of gene expression data into something more image-like. Convolutional neural networks (CNNs), which are adept at analyzing visual imagery, can then infer which genes are interacting with each other. The CNNs outperform existing methods at this task.

The researchers’ report on how CNNs can help identify disease-related genes and developmental and genetic pathways that might be targets for drugs is being published today in the Proceedings of the National Academy of Science. But Ziv Bar-Joseph, professor of computational biology and machine learning, said the applications for the new method, called CNNC, could go far beyond gene interactions.

The new insight described in the paper suggests that CNNC could be similarly deployed to investigate causality in a wide variety of phenomena, including financial data and social networking, said Bar-Joseph, who co-authored the paper with Ye Yuan, a post-doctoral researcher in CMU’s Machine Learning Department.

“CNNs, which were developed a decade ago, are revolutionary,” Bar-Joseph said. “I’m still in awe of Google Photos, which uses them for facial recognition,” he added as he scrolled through photos on his smartphone, showing how the app could identify his son at different ages, or identify his father based on an image of the rear right side of his head. “We sometimes take this technology for granted because we use it all the time. But it’s incredibly powerful and is not restricted to images. It’s all a matter of how you represent your data.”

In this case, he and Yuan were looking at gene relationships. The approximately 20,000 genes in humans work in concert, so it’s necessary to know how genes work together in complexes or networks to understand human development or diseases.

One way to infer these relationships is to look at gene expression—which represents the activity levels of genes in cells. Generally, if gene A is active at the same time gene B is active, that’s a clue that the two are interacting, Yuan said. Still, it’s possible that this is a coincidence or that both are activated by a third gene C. Several previous methods have been developed to tease out these relationships.

To employ CNNs to help analyze gene relationships, Yuan and Bar-Joseph used single-cell expression data—experiments that can determine the level of every gene in a single cell. The results of hundreds of thousands of these single-cell analyses were then arranged in the form of a matrix or histogram so that each cell of the matrix represented a different level of co-expression for a pair of genes.

Presenting the data in this way added a spatial aspect that made the data more image-like and, thus, more accessible to CNNs. By using data from genes whose interactions already had been established, the researchers were able to train the CNNs to recognize which genes were interacting and which weren’t based on the visual patterns in the data matrix, Yuan said.

“It’s very, very hard to distinguish between causality and correlation,” Yuan said, but the CNNC method proved statistically more accurate than existing methods. He and Bar-Joseph anticipate CNNC will be one of several techniques that researchers will eventually deploy in analyzing large datasets.

“This is a very general method that could be applied to a number of analyses,” Bar-Joseph said. The main limitation is data—the more data there is, the better CNNs work. Cell biology is well-suited for using CNNC, as a typical experiment can involve tens of thousands of cells and generate a massive amount of data.

The post Deep learning helps tease out gene interactions appeared first on Artificial Intelligence.

The University will lead one of four successful bids under the initiative, set up to harness the benefits of advanced data science and ensure strong data science collaborations in New Zealand and internationally.

The focus of the project, Beyond Prediction: explanatory and transparent data science for life and social sciences, is to develop new methods that discover, gather and integrate useful data that needs minimal human intervention.

The project is a collaboration between the Universities of Auckland, Otago, Canterbury, Victoria and Massey and will involve computer scientists and statisticians working alongside scientists in fields such as computational biology and ecology.

It aims to improve the application of data science methods in complex research settings, make processing more efficient and create transparent and computationally-reproducible workflows that are published, open and easily reused.

Much of the budget for the project will go towards training and equipping doctoral and post-doctoral researchers who will successfully apply data science methods to unique New Zealand datasets that improve knowledge and understanding in specific fields.

Dean of the Faculty of Science at the University of Auckland Professor John Hosking welcomed the funding.

“We know that data science has an important role to play in our ability to create good policy and to target specific areas such as healthcare, so significant funding in this area is vital to New Zealand,” he said.

Research, Science and Innovation Minister Megan Woods says the successful bids were chosen for their excellence and their potential to help New Zealand position itself at the forefront of emerging data science technologies.

Other projects chosen for funding ranged from teaching Siri to speak in Te Reo to crunching large environmental datasets collected via satellite.

The post Major funding boost for data science appeared first on Artificial Intelligence.

Turns out after they’ve been trained on enormous datasets, algorithms can not only tell what a picture is such as knowing a cat is a cat but can also generate absolutely original images. The artificial intelligence that makes this possible has matured significantly in recent years and in some applications is very proficient, but in other ways, still has a long way to go.

AI Recognizes What an Image Is

It’s taken two decades for computer scientists to train and develop machines that can “see” the world around them—another example of an everyday skill humans take for granted yet one that is quite challenging to train a machine to do.

Facial recognition technology, used both in retail and security, is one way AI and its ability to “see” the world is starting to be commonplace. Retailers use facial recognition technology to better market and sell to their target audience. In one particularly intriguing use case, some Chinese office complexes have vending machines that identify shoppers through facial recognition technology and track the items they take from the machine to ultimately bill the shoppers’ accounts. Even anonymous data about shoppers collected from cameras such as age, gender, and body language can help retailers improve their marketing efforts and provide a better customer experience.

Retail giant Walmart deployed a fleet of stock-monitoring robots that can identify through computer vision when a shelf needs more product and alert humans to ensure it happens. Target is also testing out similar technology.

Seeing AI, an iPhone app uses artificial intelligence to help blind and partially-sighted people navigate their environment by using computer vision to identify and speak its observations of the scenes and objects in its field of vision. From Face ID to unlock the iPhone X to cameras on the street used to identify criminals as well as the algorithms that allow social media platforms to identify who is in photos, AI image recognition is everywhere.

AI Image Generation

Now that artificial intelligence is able to understand, for the most part, what an image represents and can tell the difference between a stop sign and a dog, a dog from an elephant and more, the next frontier to perfect is AI image generation.

One of the former barriers to having AI generate believable images was the need for enormous datasets for training. With today’s significant computing power and the incredible amount of data we now collect, AI has breached that barrier.

How does AI know how to execute when tasked with creating an image? It uses Generative Adversarial Network or Nets (GAN), invented in 2014 by Ian Goodfellow, who was a Google researcher. It uses two neural networks; one that creates an image and another one that judges, based on real-life examples of the target image, how close the image is to the real thing. After scoring the image for accuracy, it sends that info back to the original AI system. That system learns from the feedback and returns an altered image for the next round of scoring. This process continues until the scoring machine determines the AI-generated image matches the “control” image.

Today, AI can create realistic images and videos of cats and hamburgers, representations of your words, faces that aren’t of real people and even original works of art.

In fact, there’s even a market for AI’s original artwork—Google hosted an art show to benefit charity and to showcase work created by its software DeepDream. It sold an AI-generated piece of art that was a collaboration between human and machine for $8,000 plus others. The human creator (or artist) that was part of this collaboration, Memo Akten explained that Google made a better “paintbrush” as a tool, but the human artist was still critical to creating art that would command an $8K price tag. Another AI-generated piece of art, Portrait of Edmond de Belamy was auctioned by Christie’s for $610,000.

Akten’s sentiment echoes how other industries have used AI to complement and enhance the work of humans rather than make human involvement completely unnecessary. It can inspire artists to go in directions they may not have seen without computer collaboration.

Ready to collaborate with AI on your own original work of art? Check out Depp Dream Generator or Depart that can transform any image you upload and create a new one following a particular art style. Need inspiration for what others have done? Just check out the users’ gallery. If you like what you make, you can even get them printed to display on your wall!

Artificial intelligence and computer vision are continuing to alter the way we work, live and even create. Do you see the potential or only the peril?

The post Artificial Intelligence Can Now Generate Amazing Images — What Does The Mean For Humans? appeared first on Artificial Intelligence.

In Serengeti National Park, there are 225 hidden cameras constantly photographing the creatures that roam this Tanzanian wilderness.

To date, these camera traps have captured more than three million images. For a small team of scientists living in the park, it’s a treasure trove. Through Serengeti Snapshot, as the program is called, they’ve studied everything from the migrations of the region’s herbivores to the surprising co-existence of lions, hyenas, and cheetahs.

It’s work that wouldn’t have been possible without an army of 30,000 citizen scientists, who manually sorted the collection, identifying and naming the species in each frame. It’s time-consuming work, and the volunteers are doing it for kicks. The project lives on Zooniverse, a Web platform where citizen scientists can help various scientific projects.

Now, a team of computer scientists at the University of Wyoming has found a way to automate this manual task, using artificial intelligence to identify the animals in the images quickly and accurately. By “training” the computer with the images that have been manually labeled by the volunteers, the team’s model has learned to process 99.3 percent of the images just as accurately as the human brain.

Getting these images manually classified presents “a very big bottleneck for ecologists,” says Mohammad Sadegh Norouzzadeh, a Ph.D. student at the University of Wyoming, and the first author of the paper. “[Artificial intelligence techniques] can save a tremendous amount of time for ecologists collecting the data they need.”

The benefits of this automation for science are clear: More data combined with more efficient analysis equals better scientific results, and hopefully better conservation of the natural environment.

But where does it leave the citizen scientist? A paper published on Wednesday highlights how, as people migrate to cities and lose touch with nature, citizen science “can increase emotional and cognitive connections to nature” and make participants more supportive of conservation efforts. The United States National Park Service has also found that engaging the public in hands-on work can spur further environmental action. As machines become more intelligent, and the need for human input declines, will the public’s engagement with science also suffer?

“Even though we’re hopeful that machine learning will reduce the number of images that we need people to look at, I don’t think there’s any risk that we’ll end up reducing the engagement of volunteers,” says Ali Swanson, founder of Serengeti Snapshot. “There are so many ecology and conservation projects producing so much data that I think the demand for volunteer effort is nowhere near being met.”

Rather than eradicating the need for input from enthusiastic volunteers, researchers will continue to depend on such volunteers for training their artificially intelligent models. Currently, only “supervised learning”—where machines are taught to identify animals through human-classified examples—can produce sufficiently accurate results. Until the machines can be weaned off this method, humans will still be responsible for the initial labeling process.

The easiest work is often the dreariest. Once machines have learned to classify the most obvious images, the valuable skills of a human workforce can be better utilized, Sadegh Norouzzadeh says—tasks such as classifying the 0.7 percent of images that the computer couldn’t.

“One of the problems was that the [Serengeti] project was getting boring for the citizen scientists,” Norouzzadeh says, emphasizing that the work slowed over time as volunteers dropped out. “If we use machine learning and it automatically processes most of the images, then the task could be more challenging for the citizen scientists, and more challenges means more interest in the project.”

Still, many of the current opportunities for everyday people to contribute involve working out the kinks in this new technology. As it improves, and as more information leads to better-trained models, it seems possible that this army of volunteer labelers could be rendered obsolete. I wanted to know if there is something out there—a task so intrinsically human—that it can’t be done by a computer.

“People bring a lot to the table that computers can’t,” Swanson says. “We talk a lot about ‘serendipitous discovery’ at the Zooniverse, and how volunteers often make really cool discoveries that are actually tangential to the main research question, such as the discovery of a new type of astronomical object, or unexpectedly tracking the outbreak of the Spanish Flu across the Royal Navy while looking at old ship logs for climatological data.”

Scientists are also relying more and more on the people who know their environment most intimately, such as indigenous footprint trackers. One such initiative is Footprint Identification Technology, a new approach developed by group called WildTrack based in North Carolina. It depends on indigenous people collecting photographs of animal footprints and identifying the species—something that most people, even most citizen scientists, can’t do. With this knowledge, WildTrack hopes to train an algorithm to automatically classify much more data than the group could handle manually, in hopes of revealing new insights into the various creatures roaming the planet.

Even if the need for human classification does decline, artificial intelligence is creating new opportunities for citizens to engage with nature more easily than ever before, generating an important stream of data for scientists in the process.

I discovered this for myself quite recently. It was Friday evening, and my boyfriend and I were out of milk. On our way to Target, I came across a tree that grows by my bus stop and that I’d finally managed to identify as a honeylocust. I’d recently stumbled across iNaturalist, an app that uses deep learning techniques to identify plants and animals. Based on the app’s description, it sounded as though it would be able to draw upon previously uploaded images of other honeylocusts to classify the particular tree in question.

Even though I’ve spent the last five years of my life writing about the environment, I’m no botanist. My knowledge of the plant world peaked early, when I developed a passion for snapdragons, mainly because they were the coolest plant growing in my grandmother’s garden. My ignorance beyond this one flower is something I’ve long wanted to remedy, but it was difficult to know where to begin. Manually identifying the honeylocust had required me to individually Google every species on the city government’s official tree spreadsheet. It wasn’t a sustainable solution.

Could iNaturalist save me hours of botanical procrastination? I pointed the camera at the honeylocust and clicked, “What did you see?” A word popped up on the screen: “honeylocust.” In the 10-minute walk to Target, we added oak and elm to our collection. Since then, just in our neighborhood, we’ve spotted wild blue indigo, meadow anemone, western columbine, and hairy beardtongue, among others.

That’s not to say that the app works perfectly. It thinks our orchid is a radish. But that’s where humans take over. And the more images that are labeled by citizen scientists, the more the app will learn and improve.

The purpose of iNaturalist is explicitly to connect people more closely to nature. Additionally, individuals can build communal projects that transcend personal curiosity. For instance, Boston has invited locals to document the various species of oyster living in the harbor, with an eye toward determining whether invasive European oysters are encroaching on native varieties.

Developers at iNaturalist have also built a page for “research-grade observations,” which can be used in scientific studies. Data collected by citizen scientists through the app has been used to investigate topics including the effect of invasive pests(such as the Asian longhorned beetle) on carbon sequestration in forests, and the transmission of viruses via fruit bats in parts of Asia.

I’m not sure how useful our documentation of the wildflowers of Southside Chicago will be, but it has added texture to our own lives. Our streets are suddenly cloaked in color and diversity. Our grocery trips are an opportunity for exploration and discovery. In fact, they always were; we just didn’t know it.

The post WILL ARTIFICIAL INTELLIGENCE MAKE CITIZEN SCIENTISTS OBSOLETE? appeared first on Artificial Intelligence.

New England’s groundfish season is in full swing, as hundreds of dayboat fishermen from Rhode Island to Maine take to the water in search of the region’s iconic cod and haddock. But this year, several dozen of them are hauling in their catch under the watchful eye of video cameras as part of a new effort to use technology to better sustain the area’s fisheries and the communities that depend on them.

Video observation on fishing boats—electronic monitoring—is picking up steam in the Northeast and nationally as a cost-effective means to ensure that fishing vessels aren’t catching more fish than allowed while informing local fisheries management. While several issues remain to be solved before the technology can be widely deployed—such as the costs of reviewing and storing data—electronic monitoring is beginning to deliver on its potential to lower fishermen’s costs, provide scientists with better data, restore trust where it’s broken, and ultimately help consumers gain a greater understanding of where their seafood is coming from.

“Electronic monitoring is a tremendous tool,” says Brett Alger, national electronics technology coordinator for NOAA Fisheries. “It isn’t necessarily for everyone or every fishery,” but “we’re working collaboratively in all of our regions with fishermen on the ground to understand their needs. I expect it to grow.”

The technology is required for highly migratory longline species in the Atlantic (swordfish). It’s thriving in the Pacific coast groundfish industry, and dozens of other fisheries regions have pilot initiatives.

“I was dead set against electronic monitoring when it first started,” says Nick Muto, a veteran commercial fisherman based in Cape Cod. “I wasn’t in favor of having Big Brother over my shoulder all the time. There was a lot of gray area about what we were calling data. There were concerns about confidentiality and who would have access to the data.”

Once the fishery’s stakeholders were able to work through those issues; however, Muto decided to give it a go. Last year he installed cameras on his 45-foot vessel, the Dawn T. He opted to try the technology because he says he “was upset with the broken system of human monitors.”

Human observers are widely used to monitor catch in quota-managed fisheries, and they’re expensive: It costs roughly $700 a day for an observer in New England.

New England uses a “catch shares” system to regulate the groundfish fishery. Fishermen are allotted permits for the total amount of fish they can catch in one year and discards (e.g., juveniles) count towards that quota. Human observers serve as monitors to determine whether fishermen are accurately reporting their discards at sea. Fishermen are required to notify regulators about their trips 48 hours before they leave the dock.

While human observers are required this year on only 15 percent of trips under the Northeast’s At-Sea Monitoring (ASM) Program, the costs add up in an industry with very tight margins. Many fishermen also dislike having another person on board a small vessel.

“It’s an unsafe situation,” says Muto, a first-generation fisherman who got his start as a deck hand, cleaning lobster pots and sorting gear. “Sure, an observer has insurance, but on top of all the other headaches, I now have responsibility for this other person, with all their scales and baskets. It makes a small boat even smaller,” he adds.

Regulators began allowing cameras on New England’s dayboats (smaller fishing vessels that typically spend only a few days out at sea) as a substitute for human observers in 2016. Two other programs are still in the pilot phase in New England, including one for larger groundfish trawlers and one for the herring fishery.

Muto’s vessel was outfitted with cameras, at a cost of about $8,000, through a collaborative venture between NOAA’s regional office and science center, The Nature Conservancy (TNC), the Gulf of Maine Research Institute, and the Cape Cod Commercial Fishermen’s Alliance. Camera costs are currently subsidized by NOAA Fisheries and its partners.

The cameras run the entire time Muto and his crew are out on the water. They record how the fisherman handle their discards, the fish they’re not allowed to keep because of size or species type, but that count towards their quotas. The cost is lower than what he’d pay for an in-person monitor.

One way to cut those costs is to spot-check the video footage and compare it to fishermen’s reports, in a “trust but verify” system, says Christopher McGuire, marine program director for TNC in Massachusetts. That approach is working in other fisheries, such as British Columbia’s groundfish fishery, where regulators view about 10 percent of the footage.

Using Machine Learning and Artificial Intelligence

Another way to cut costs is to use computers to review the footage. McGuire says there’s been a lot of talk about automating the review, but the common refrain is that it’s still five years off.

To spur faster action, TNC last year spearheaded an online competition, offering a $50,000 prize to computer scientists who could crack the code—that is, teach a computer how to count fish, size them, and identify their species.

“We created an arms race,” says McGuire. “That’s why you do a competition. You’ll never get the top minds to do this because they don’t care about your fish. They all want to work for Google, and one way to get recognized by Google is to win a few of these competitions.”

The contest exceeded McGuire’s expectations. “Winners got close to 100 percent in count and 75 percent accurate on identifying species,” he says. “We proved that automated review is now. Not in five years. And now all of the video-review companies are investing in machine leaning.” It’s only a matter of time before a commercial product is available, McGuire believes.

Big Brother or Better Trust?

With the rise of surveillance in society, from dashboard cameras to Facebook, it’s no surprise that fishermen may be leery about how the video footage that’s collected on their boats will be used.

“The biggest concern is that as soon as you give up your data to a third party, you’re vulnerable immediately. Your data could be used for unintended purposes,” says Jungwoo Ryoo, professor of information sciences and technology at Pennsylvania State University, on the larger trend. But, he says, “blind denial of these useful technologies is not the way to go. Big data helps us do our jobs more efficiently.”

Ryoo cautions fishermen to make sure that privacy and security are in the terms and conditions of electronic monitoring agreements, and to seek protections as a group. Fortunately, fishermen have group power. Electronic monitoring program details are typically worked out by local fishery councils where fishermen have a voice. That’s why New England industry experts think that electronic monitoring could actually help to reverse legendary mistrust between fishermen and scientists.

Fishermen are required to record everything they catch and discard. However, “Scientists don’t typically trust what fishermen write on their trip reports,” says McGuire. “There are a lot of incentives for people to misreport. Even though fisherman are required to fill them out, the management and science largely don’t use them.”

“Electronic Monitoring is trying to flip that on its head,” he says. “A fisherman’s catch record would be accepted as an accurate accounting of the trip unless proven otherwise. It’s one of the first ways you start to change the dynamic in the fishery.”

Muto agrees. “We look at the scientists and say, ‘you’re full of shit,’ and they look at us and say the same thing,” he says. But now that he has video footage to point to, he adds, “when I show up to a regulatory meeting or a hearing, I can say, here’s what I’m seeing on the water and here is my evidence to prove it. ”

Electronic monitoring also provides scientists and regulators with a more timely, more accurate assessment of the fishery’s health. And that makes it easier to link catch limits more closely to actual populations and improve the effectiveness of conservation restrictions.

Connecting to Consumer Traceability

As for consumers, one day, video footage of fishing-vessel activity could be connected to traceability movements that seek to ensure that seafood is ethically and legally harvested.

Ecotrust Canada, the video review partner for New England’s program, is piloting the first such traceability program. ThisFish allows consumers to enter a code online that tells them who caught their fish, how and where, and possibly even watch an online clip. So far it’s focused on Canadian fisheries.

McGuire thinks that the tuna industry is ripe for more of this sort of deck-to-dish traceability because of its high value. But he cautions that it may take a while to catch on. “Electronic monitoring is tough to explain in the three seconds it takes people to walk past the seafood case.”

A Future Without Human Observers?

Thus far, Muto says he’s fairly satisfied with how the cameras are working. Recently, he was rewarded when regulators granted greater flexibility to the fishermen using electronic monitoring. Now he can respond to what he sees on the water, and shift from fishing for groundfish to fishing for Atlantic Bluefin tuna without having to first return to port and notify regulators.

Alger says that sort of flexibility appeals to fishermen. “A lot of industry groups are starting to see electronic monitoring as a way to go back to the way it was: ‘I’m just going to go fishing.’ They’re seeing electronic monitoring as a way to be more opportunistic when they go fishing, and to provide more info about what’s occurring in the ocean.”

While today, only a fraction of New England dayboat fishermen are using cameras, McGuire is hopeful that the industry will reach a tipping point where use of the technology becomes the norm. “Fishermen talk a lot,” he says. “We just need one or two in every port. Others will find out.”

The post The Future of Fishing Is Big Data and Artificial Intelligence appeared first on Artificial Intelligence.