Digital transformation depends on diversity

Across industries, businesses are now tech and data companies. The sooner they grasp and live that, the quicker they will meet their customer needs and expectations, create more business value and grow. It is increasingly important to reimagine business and use digital technologies to create new business processes, cultures, customer experiences and opportunities.

One of the myths about digital transformation is that it’s all about harnessing technology. It’s not. To succeed, digital transformation inherently requires and relies on diversity. Artificial intelligence (AI) is the result of human intelligence, enabled by its vast talents and also susceptible to its limitations.

Therefore, it is imperative for organizations and teams to make diversity a priority and think about it beyond the traditional sense. For me, diversity centers around three key pillars.


People are the most important part of artificial intelligence; the fact is that humans create artificial intelligence. The diversity of people — the team of decision-makers in the creation of AI algorithms — must reflect the diversity of the general population.

This goes beyond ensuring opportunities for women in AI and technology roles. In addition, it includes the full dimensions of gender, race, ethnicity, skill set, experience, geography, education, perspectives, interests and more. Why? When you have diverse teams reviewing and analyzing data to make decisions, you mitigate the chances of their own individual and uniquely human experiences, privileges and limitations blinding them to the experiences of others.

One of the myths about digital transformation is that it’s all about harnessing technology. It’s not.

Collectively, we have an opportunity to apply AI and machine learning to propel the future and do good. That begins with diverse teams of people who reflect the full diversity and rich perspectives of our world.

Diversity of skills, perspectives, experiences and geographies has played a key role in our digital transformation. At Levi Strauss & Co., our growing strategy and AI team doesn’t include solely data and machine learning scientists and engineers. We recently tapped employees from across the organization around the world and deliberately set out to train people with no previous experience in coding or statistics. We took people in retail operations, distribution centers and warehouses, and design and planning and put them through our first-ever machine learning bootcamp, building on their expert retail skills and supercharging them with coding and statistics.

We did not limit the required backgrounds; we simply looked for people who were curious problem solvers, analytical by nature and persistent to look for various ways of approaching business issues. The combination of existing expert retail skills and added machine learning knowledge meant employees who graduated from the program now have meaningful new perspectives on top of their business value. This first-of-its-kind initiative in the retail industry helped us develop a talented and diverse bench of team members.


AI and machine learning capabilities are only as good as the data put into the system. We often limit ourselves to thinking of data in terms of structured tables — numbers and figures — but data is anything that can be digitized.

The digital images of the jeans and jackets our company has been producing for the past 168 years are data. The customer service conversations (recorded only with permissions) are data. The heatmaps from how people move in our stores are data. The reviews from our consumers are data. Today, everything that can be digitized becomes data. We need to broaden how we think of data and ensure we constantly feed all data into AI work.

Most predictive models use data from the past to predict the future. But because the apparel industry is still in the nascent stages of digital, data and AI adoption, having past data to reference is often a common problem. In fashion, we’re looking ahead to predict trends and demand for completely new products, which have no sales history. How do we do that?

We use more data than ever before, for example, both images of the new products and a database of our products from past seasons. We then apply computer vision algorithms to detect similarity between past and new fashion products, which helps us predict demand for those new products. These applications provide much more accurate estimates than experience or intuition do, supplementing previous practices with data- and AI-powered predictions.

At Levi Strauss & Co., we also use digital images and 3D assets to simulate how clothes feel and even create new fashion. For example, we train neural networks to understand the nuances around various jean styles like tapered legs, whisker patterns and distressed looks, and detect the physical properties of the components that affect the drapes, folds and creases. We’re then able to combine this with market data, where we can tailor our product collections to meet changing consumer needs and desires and focus on the inclusiveness of our brand across demographics. Furthermore, we use AI to create new styles of apparel while always retaining the creativity and innovation of our world-class designers.

Tools and techniques

In addition to people and data, we need to ensure diversity in the tools and techniques we use in the creation and production of algorithms. Some AI systems and products use classification techniques, which can perpetuate gender or racial bias.

For example, classification techniques assume gender is binary and commonly assign people as “male” or “female” based on physical appearance and stereotypical assumptions, meaning all other forms of gender identity are erased. That’s a problem, and it’s upon all of us working in this space, in any company or industry, to prevent bias and advance techniques in order to capture all the nuances and ranges in people’s lives. For example, we can take race out of the data to try and render an algorithm race-blind while continuously safeguarding against bias.

We are committed to diversity in our AI products and systems and, in striving for that, we use open-source tools. Open-source tools and libraries by their nature are more diverse because they are available to everyone around the world and people from all backgrounds and fields work to enhance and advance them, enriching with their experiences and thus limiting bias.

An example of how we do this at Levi Strauss & Company is with our U.S. Red Tab loyalty program. As fans set up their profiles, we don’t ask them to pick a gender or allow the AI system to make assumptions. Instead, we ask them to pick their style preferences (Women, Men, Both or Don’t Know) in order to help our AI system build tailored shopping experiences and more personalized product recommendations.

Diversity of people, data, and techniques and tools is helping Levi Strauss & Co. revolutionize its business and our entire industry, transforming manual to automated, analog to digital, and intuitive to predictive. We are also building on the legacy of our company’s social values, which has stood for equality, democracy and inclusiveness for 168 years. Diversity in AI is one of the latest opportunities to continue this legacy and shape the future of fashion.

#articles, #artificial-general-intelligence, #artificial-intelligence, #column, #cybernetics, #dei, #diversity, #fashion, #machine-learning, #neural-networks, #science-and-technology, #tc

Researchers demonstrate that malware can be hidden inside AI models

This photo has a job application for Boston University hidden within it. The technique introduced by Wang, Liu, and Cui could hide data inside an image classifier rather than just an image.

Enlarge / This photo has a job application for Boston University hidden within it. The technique introduced by Wang, Liu, and Cui could hide data inside an image classifier rather than just an image. (credit: Keith McDuffy CC-BY 2.0)

Researchers Zhi Wang, Chaoge Liu, and Xiang Cui published a paper last Monday demonstrating a new technique for slipping malware past automated detection tools—in this case, by hiding it inside a neural network.

The three embedded 36.9MiB of malware into a 178MiB AlexNet model without significantly altering the function of the model itself. The malware-embedded model classified images with near-identical accuracy, within 1% of the malware-free model. (This is possible because the number of layers and total neurons in a convolutional neural network is fixed prior to training—which means that, much like in human brains, many of the neurons in a trained model end up being either largely or entirely dormant.)

Just as importantly, squirreling the malware away into the model broke it up in ways that prevented detection by standard antivirus engines. VirusTotal, a service that “inspects items with over 70 antivirus scanners and URL/domain blocklisting services, in addition to a myriad of tools to extract signals from the studied content,” did not raise any suspicions about the malware-embedded model.

Read 4 remaining paragraphs | Comments

#ai, #deep-learning, #machine-learning, #malware, #neural-networks, #steganography, #tech

Movie written by algorithm turns out to be hilarious and intense

Sunspring, a short science fiction movie written entirely by AI, debuted exclusively on Ars in June 2016. (video link)

Ars is excited to be hosting this online debut of Sunspring, a short science fiction film that’s not entirely what it seems. It’s about three people living in a weird future, possibly on a space station, probably in a love triangle. You know it’s the future because H (played with neurotic gravity by Silicon Valley‘s Thomas Middleditch) is wearing a shiny gold jacket, H2 (Elisabeth Gray) is playing with computers, and C (Humphrey Ker) announces that he has to “go to the skull” before sticking his face into a bunch of green lights. It sounds like your typical sci-fi B-movie, complete with an incoherent plot. Except Sunspring isn’t the product of Hollywood hacks—it was written entirely by an AI. To be specific, it was authored by a recurrent neural network called long short-term memory, or LSTM for short. At least, that’s what we’d call it. The AI named itself Benjamin.

Knowing that an AI wrote Sunspring makes the movie more fun to watch, especially once you know how the cast and crew put it together. Director Oscar Sharp made the movie for Sci-Fi London, an annual film festival that includes the 48-Hour Film Challenge, where contestants are given a set of prompts (mostly props and lines) that have to appear in a movie they make over the next two days. Sharp’s longtime collaborator, Ross Goodwin, is an AI researcher at New York University, and he supplied the movie’s AI writer, initially called Jetson. As the cast gathered around a tiny printer, Benjamin spat out the screenplay, complete with almost impossible stage directions like “He is standing in the stars and sitting on the floor.” Then Sharp randomly assigned roles to the actors in the room. “As soon as we had a read-through, everyone around the table was laughing their heads off with delight,” Sharp told Ars. The actors interpreted the lines as they read, adding tone and body language, and the results are what you see in the movie. Somehow, a slightly garbled series of sentences became a tale of romance and murder, set in a dark future world. It even has its own musical interlude (performed by Andrew and Tiger), with a pop song Benjamin composed after learning from a corpus of 30,000 other pop songs.

Read 10 remaining paragraphs | Comments

#ai, #gaming-culture, #neural-networks, #oscar-sharp, #ross-goodwin, #sunspring

Archaeologists train a neural network to sort pottery fragments for them

Archaeologists train a neural network to sort pottery fragments for them

(credit: Pawlowicz and Downum 2021)

Real archaeological fieldwork is seldom as exciting as it looks in the movies. You tend to get fewer reanimated mummies, deadly booby traps, and dramatic shootouts with Nazis. Instead, you’ll see pieces of broken pottery—a lot of them. Potsherds are ubiquitous at archaeological sites, and that’s true for pretty much every culture since people invented pottery. In the US Southwest in particular, museums have collected sherds by the tens of thousands.

Although all those broken bits may not look like much at first glance, they’re often the key to piecing together the past.

“[Potsherds] provide archaeologists with critical information about the time a site was occupied, the cultural group with which it was associated, and other groups with whom they interacted,” said Northern Arizona University archaeologist Chris Downum, who co-authored a new study with Leszek Pawlowicz.

Read 15 remaining paragraphs | Comments

#ancestral-pueblo, #ancient-north-america, #archaeology, #machine-learning, #neural-networks, #potsherds, #pottery, #pre-columbian-civilizations, #science

Google Photos update adds new Memories and a Locked Folder, previews Cinematic moments

Google announced a series of upgrades to its Google Photos service, used by over a billion users, at today’s Google I/O developer event, which was virtually streamed this year due to Covid. The company is rolling out Locked Folders, new types of photo “Memories” for reminiscing over past events, as well as a new feature called “Cinematic moments” that will animate a series of static photos, among other updates.

Today, Google Photos stores over 4 trillion photos and videos, but the majority of those are never viewed. To change that, Google has been developing A.I.-powered features to help its users reflect on meaningful moments from their lives. With Memories, launched in 2019, Google Photos is able to resurface photos and videos focused on people, activities, and hobbies as well as recent highlights from the week prior.

At Google I/O, the company announced it’s adding a new type of Memory, which it’s calling “little patterns.” Using machine learning, little patterns looks for a set of three or more photos with similarities, like shape or color, which it then highlights as a pattern for you.

Image Credits: Google

For example, when one of Google’s engineers traveled the world with their favorite orange backpack, Google Photos was able to identify a pattern where that backpack was featured in photos from around the globe. But patterns may also be as simple family photos that are often snapped in the same room with an identifiable piece of furniture, like the living room couch. On their own, these photos may not seem like much, but when they’re combined over time, they can produce some interesting compilations.

Google will also be adding Best of Month Memories and Trip highlights to the your photo grid, which you’ll now be able to remove or rename, as well as Memories featuring events you celebrate, like birthdays or holidays. These events will be identified based on a combination of factors, Google says. This includes by identifying objects in the photos — like a birthday cake or a Hanukkah menorah, for example — as well as by matching up the date of the photo with known holidays.

Image Credits: Google

Best of Month and Trip highlight Memories will start to roll out today and will be found in the photo grid itself. Later this year, you’ll begin to see Memories related to the events and moments you celebrate.

Image Credits: Google

Another forthcoming addition is Cinematic Moments, which is somewhat reminiscent of the “deep nostalgia” technology from MyHeritage that went viral earlier this year, as users animated the photos of long-past loved ones. Except in Google’s case, it’s not taking an old photo and bringing it to life, it’s stitching together a series of photos to create a sense of action and movement.

Google explains that, often, people will take multiple photos of the same moment in order to get one “good” image they can share. This is especially true when trying to capture something in motion — like a small child or a pet who can’t sit still.

Image Credits: Google

These new Cinematic moments build on the Cinematic photos feature Google launched in December 2020, which uses machine learning to create vivid, 3D version of your photos. Using computational photography and neural networks to stitch together a series of near-identical photos, Google Photos will be able to create vivid, moving images by filling in the gaps in between your photos to create new frames. This feature doesn’t have a launch date at this time.

Of course, not all past moments are worthy of revisiting for a variety of reasons. While Google already offered tools to hide certain photos and time periods from your Memories, it’s continuing to add new controls and, later this summer, will make it easier to access its existing toolset. One key area of focus has been working with the transgender community, who have said that revisiting their old photos can be painful.

Soon, users will also be able to remove a single photo from a Memory, remove their Best of Month Memories, and rename and remove Memories based on the events they celebrate, too.

Image Credits: Google

Another useful addition to Google Photos is the new Locked Folder, which is simply a passcode-protected space for private photos. Many users automatically sync their phone’s photos to Google’s cloud, but then want to pull up photos to show to others through the app on their phone or even their connected TV. That can be difficult if their galleries are filled with private photos, of course.

Image Credits: Google

This particular feature will launch first on Pixel devices, where users will have the option to save photos and videos directly from their Camera to the Locked folder. Other Android devices will get the update later in the year.

#ai, #artificial-intelligence, #google, #google-photos, #google-search, #machine-learning, #neural-networks, #photos, #tc

AI is ready to take on a massive healthcare challenge

Which disease results in the highest total economic burden per annum? If you guessed diabetes, cancer, heart disease or even obesity, you guessed wrong. Reaching a mammoth financial burden of $966 billion in 2019, the cost of rare diseases far outpaced diabetes ($327 billion), cancer ($174 billion), heart disease ($214 billion) and other chronic diseases.

Cognitive intelligence, or cognitive computing solutions, blend artificial intelligence technologies like neural networks, machine learning, and natural language processing, and are able to mimic human intelligence.

It’s not surprising that rare diseases didn’t come to mind. By definition, a rare disease affects fewer than 200,000 people. However, collectively, there are thousands of rare diseases and those affect around 400 million people worldwide. About half of rare disease patients are children, and the typical patient, young or old, weather a diagnostic odyssey lasting five years or more during which they undergo countless tests and see numerous specialists before ultimately receiving a diagnosis.

No longer a moonshot challenge

Shortening that diagnostic odyssey and reducing the associated costs was, until recently, a moonshot challenge, but is now within reach. About 80% of rare diseases are genetic, and technology and AI advances are combining to make genetic testing widely accessible.

Whole-genome sequencing, an advanced genetic test that allows us to examine the entire human DNA, now costs under $1,000, and market leader Illumina is targeting a $100 genome in the near future.

The remaining challenge is interpreting that data in the context of human health, which is not a trivial challenge. The typical human contains 5 million unique genetic variants and of those we need to identify a single disease-causing variant. Recent advances in cognitive AI allow us to interrogate a person’s whole genome sequence and identify disease-causing mechanisms automatically, augmenting human capacity.

A shift from narrow to cognitive AI

The path to a broadly usable AI solution required a paradigm shift from narrow to broader machine learning models. Scientists interpreting genomic data review thousands of data points, collected from different sources, in different formats.

An analysis of a human genome can take as long as eight hours, and there are only a few thousand qualified scientists worldwide. When we reach the $100 genome, analysts are expecting 50 million-60 million people will have their DNA sequenced every year. How will we analyze the data generated in the context of their health? That’s where cognitive intelligence comes in.

#artificial-intelligence, #cognitive-computing, #column, #cybernetics, #ec-column, #ec-consumer-health, #emerging-technologies, #genomics, #health, #machine-learning, #natural-language-processing, #neural-networks, #obesity, #precision-medicine

Deeplite raises $6M seed to deploy ML on edge with fewer compute resources

One of the issues with deploying a machine learning application is that it tends to be expensive and highly compute intensive.  Deeplite, a startup based in Montreal, wants to change that by providing a way to reduce the overall size of the model, allowing it to run on hardware with far fewer resources.

Today, the company announced a $6 million seed investment. Boston-based venture capital firm PJC led the round with help from Innospark Ventures, Differential Ventures and Smart Global Holdings. Somel Investments, BDC Capital and Desjardins Capital also participated.

Nick Romano, CEO and co-founder at Deeplite, says that the company aims to take complex deep neural networks that require a lot of compute power to run, tend to use up a lot of memory, and can consume batteries at a rapid pace, and help them run more efficiently with fewer resources.

“Our platform can be used to transform those models into a new form factor to be able to deploy it into constrained hardware at the edge,” Romano explained. Those devices could be as small as a cell phone, a drone or even a Raspberry Pi, meaning that developers could deploy AI in ways that just wouldn’t be possible in most cases right now.

The company has created a product called Neutrino that lets you specify how you want to deploy your model and how much you can compress it to reduce the overall size and the resources required to run it in production. The idea is to run a machine learning application on an extremely small footprint.

Davis Sawyer, chief product officer and co-founder, says that the company’s solution comes into play after the model has been built, trained and is ready for production. Users supply the model and the data set and then they can decide how to build a smaller model. That could involve reducing the accuracy a bit if there is a tolerance for that, but chiefly it involves selecting a level of compression — how much smaller you can make the model.

“Compression reduces the size of the model so that you can deploy it on a much cheaper processor. We’re talking in some cases going from 200 megabytes down to on 11 megabytes or from 50 megabytes to 100 kilobytes,” Davis explained.

Rob May, who is leading the investment for PJC, says that he was impressed with the team and the technology the startup is trying to build.

“Deploying AI, particularly deep learning, on resource-constrained devices, is a broad challenge in the industry with scarce AI talent and know-how available. Deeplite’s automated software solution will create significant economic benefit as Edge AI continues to grow as a major computing paradigm,” May said in a statement.

The idea for the company has roots in the TandemLaunch incubator in Montreal. It launched officially as a company in mid-2019 and today has 15 employees with plans to double that by the end of this year. As it builds the company, Romano says the founders are focused on building a diverse and inclusive organization.

“We’ve got a strategy that’s going to find us the right people, but do it in a way that is absolutely diverse and inclusive. That’s all part of the DNA of the organization,” he said.

When it’s possible to return to work, the plan is to have offices in Montreal and Toronto that act as hubs for employees, but there won’t be any requirement to come into the office.

“We’ve already discussed that the general approach is going to be that people can come and go as they please, and we don’t think we will need as large an office footprint as we may have had in the past. People will have the option to work remotely and virtually as they see fit,” Romano said.

#artificial-intelligence, #developer, #edge-computing, #funding, #machine-learning, #neural-networks, #recent-funding, #rob-may, #startups, #tc

MIT researchers develop a new ‘liquid’ neural network that’s better at adapting to new info

A new type of neural network that’s capable of adapting its underlying behavior after the initial training phase could be the key to big improvements in situations where conditions can change quickly – like autonomous driving, controlling robots, or diagnosing medical conditions. These so-called ‘liquid’ neural networks were devised by MIT Computer Science and Artificial Intelligence Lab’s Ramin Hasani and his team at CSAIL, and they have the potential to greatly expand the flexibility of AI technology after the training phase, when they’re engaged in the actual practical inference work done in the field.

Typically, after the training phase, during which neural network algorithms are provided with a large volume of relevant target data to hone their inference capabilities, and rewarded for correct responses in order to optimize performance, they’re essentially fixed. But Hasani’s team developed a means by which his ‘liquid’ neural net can adapt the parameters for ‘success’ over time in response to new information, which means that if a neural net tasked with perception on a self-driving car goes from clear skies into heavy snow, for instance, it’s better able to deal with the shift in circumstances and maintain a high level of performance.

The main difference in the method introduced by Hasani and his collaborators is that it focuses on time-series adaptability, meaning that rather than being built on training data that is essentially made up of a number of snapshots, or static moments fixed in time, the liquid networks inherently considers time series data – or sequences of images rather than isolated slices.

Because of the way the system is designed, it’s actually also more open to observation and study by researchers, when compared to traditional neural networks. This kind of AI is typically referred to as a ‘black box,’ because while those developing the algorithms know the inputs and the the criteria for determining and encouraging successful behavior, they can’t typically determine what exactly is going on within the neural networks that leads to success. This ‘liquid’ model offers more transparency there, and it’s less costly when it comes to computing because it relies on fewer, but more sophisticated computing nodes.

Meanwhile, performance results indicate that it’s better than other alternatives for accuracy in predicting the future values of known data sets. Th next step for Hasani and his team are to determine how best to make the system even better, and ready it for use in actual practical applications.

#articles, #artificial-intelligence, #artificial-neural-networks, #computing, #emerging-technologies, #neural-network, #neural-networks, #science, #science-and-technology, #self-driving-car, #tc

Tesla is willing to license Autopilot and has already had “preliminary discussions” about it with other automakers

Tesla is open to licensing its software, including its Autopilot highly-automated driving technology, and the neural network training it has built to improve its autonomous driving technology. Tesla CEO Elon Musk revealed those considerations on the company’s Q4 earnings call on Wednesday, adding that the company has in fact already “had some preliminary discussions about licensing Autopilot to other OEMs.”

The company began rolling out its beta version of the so-called ‘full self-driving’ or FSD version of Autopilot late last year. The standard Autopilot features available in general release provide advanced driver assistance (ADAS) which provide essentially advanced cruise control capabilities designed primarily for use in highway commutes. Musk said on the call that he expects the company will seek to prove out its FSD capabilities before entering into any licensing agreements, if it does end up pursuing that path.

Musk noted that Tesla’s “philosophy is definitely not to create walled gardens” overall, and pointed out that the company is planning to allow other automakers to use its Supercharger networks, as well as its autonomy software. He characterized Tesla as “more than happy to license” those autonomous technologies to “other car companies,” in fact.

One key technical hurdle required to get to a point where Tesla’s technology is able to demonstrate true reliability far surpassing that of a standard human driver is transition the neural networks operating in the cars and providing them with the analysis that powers their perception engines is to transition those to video. That’s a full-stack transition across the system away from basing it around neural nets trained on single cameras and single frames.

To this end, the company has developed video labelling software that has had “a huge effect on the efficiency of labeling,” with the ultimate aim being enabling automatic labeling. Musk (who isn’t known for modesty around his company’s achievements, it should be said) noted that Tesla believes “it may be the best neural net training computer in the world by possibly an order of magnitude,” adding that it’s also “something we can offer potentially as a service.”

Training huge quantities of video data will help Tesla push the reliability of its software from 100% that of a human driver, to 200% and eventually to “2,000% better than the average human,” Musk said, while again suggesting that it won’t be a technological achievement the company is interested into keeping to themselves.

#automation, #automotive, #car, #ceo, #driver, #elon-musk, #neural-network, #neural-networks, #self-driving-car, #tc, #tesla, #tesla-autopilot, #transport, #transportation

AI’s next act: Genius chips, programmable silicon and the future of computing

If only 10% of the world had enough power to run a cell phone, would mobile have changed the world in the way that it did?

It’s often said the future is already here — just not evenly distributed. That’s especially true in the world of artificial intelligence (AI) and machine learning (ML). Many powerful AI/ML applications already exist in the wild, but many also require enormous computational power — often at scales only available to the largest companies in existence or entire nation-states. Compute-heavy technologies are also hitting another roadblock: Moore’s law is plateauing and the processing capacity of legacy chip architectures are running up against the limits of physics.

If major breakthroughs in silicon architecture efficiency don’t happen, AI will suffer an unevenly distributed future and huge swaths of the population miss out on the improvements AI could make to their lives.

The next evolutionary stage of technology depends on completing the transformation that will make silicon architecture as flexible, efficient and ultimately programmable as the software we know today. If we cannot take major steps to provide easy access to ML we’ll lose unmeasurable innovation by having only a few companies in control of all the technology that matters. So what needs to change, how fast is it changing and what will that mean for the future of technology?

An inevitable democratization of AI: A boon for startups and smaller businesses

If you work at one of the industrial giants (including those “outside” of tech), congratulations — but many of the problems with current AI/ML computing capabilities I present here may not seem relevant.

For those of you working with lesser caches of resources, whether financially or talent-wise, view the following predictions as the herald of a new era in which organizations of all sizes and balance sheets have access to the same tiers of powerful AI and ML-powered software. Just like cell phones democratized internet access, we see a movement in the industry today to put AI in the hands of more and more people.

Of course, this democratization must be fueled by significant technological advancement that actually makes AI more accessible — good intentions are not enough, regardless of the good work done by companies like Intel and Google. Here are a few technological changes we’ll see that will make that possible.

From dumb chip to smart chip to “genius” chip

For a long time, raw performance was the metric of importance for processors. Their design reflected this. As software rose in ubiquity, processors needed to be smarter: more efficient and more commoditized, so specialized processors like GPUs arose — “smart” chips, if you will.

Those purpose-built graphics processors, by happy coincidence, proved to be more useful than CPUs for deep learning functions, and thus the GPU became one of the key players in modern AI and ML. Knowing this history, the next evolutionary step becomes obvious: If we can purpose-build hardware for graphics applications, why not for specific deep learning, AI and ML?

There’s also a unique confluence of factors that makes the next few years pivotal for chipmaking and tech in general. First and second, we’re seeing a plateauing of Moore’s law (which predicts a doubling of transistors on integrated circuits every two years) and the end of Dennard scaling (which says performance-per-watt doubles at about the same rate). Together, that used to mean that for any new generation of technology, chips doubled in density and increased in processing power while drawing the same amount of power. But we’ve now reached the scale of nanometers, meaning we’re up against the limitations of physics.

Thirdly, compounding the physical challenge, the computing demands of next-gen AI and ML apps are beyond what we could have imagined. Training neural networks to within even a fraction of human image recognition, for example, is surprisingly hard and takes huge amounts of processing power. The most intense applications of machine learning are things like natural language processing (NLP), recommender systems that deal with billions or trillions of possibilities, or super high-resolution computer vision, as is used in the medical and astronomical fields.

Even if we could have predicted we’d have to create and train algorithmic brains to learn how to speak human language or identify objects in deep space, we still could not have guessed just how much training — and therefore processing power — they might need to become truly useful and “intelligent” models.

Of course, many organizations are performing these sorts of complex ML applications. But these sorts of companies are usually business or scientific leaders with access to huge amounts of raw computing power and the talent to understand and deploy them. All but the largest enterprises are locked out of the upper tiers of ML and AI.

That’s why the next generation of smart chips — call them “genius” chips — will be about efficiency and specialization. Chip architecture will be made to optimize for the software running on it and run altogether more efficiently. When using high-powered AI doesn’t take a whole server farm and becomes accessible to a much larger percentage of the industry, the ideal conditions for widespread disruption and innovation become real. Democratizing expensive, resource intensive AI goes hand-in-hand with these soon-to-be-seen advances in chip architecture and software-centered hardware design.

A renewed focus on future-proofing innovation

The nature of AI creates a special challenge for the creators and users of AI hardware. The amount of change itself is huge: We’re living through the leap from humans writing code to software 2.0 — where engineers can train machine learning programs to eventually “run themselves.” The rate of change is also unprecedented: ML models can be obsolete in months or even weeks, and the very methods through which training happens are in constant evolution.

But creating new AI hardware products still requires designing, prototyping, calibrating, troubleshooting, production and distribution. It can take two years from concept to product-in-hand. Software has, of course, always outpaced hardware development, but now the differential in velocity is irreconcilable. We need to be more clever about the hardware we create for a future we increasingly cannot predict.

In fact, the generational way we think about technology is beginning to break down. When it comes to ML and AI, hardware must be built with the expectation that much of what we know today will be obsolete by the time we have the finished product. Flexibility and customization will be the key attributes of successful hardware in the age of AI, and I believe this will be a further win for entire market.

Instead of sinking resources into the model du jour or a specific algorithm, companies looking to take advantage of these technologies will have more options for processing stacks that can evolve and change as the demands of ML and AI models evolve and change.

This will allow companies of all sizes and levels of AI savvy to stay creative and competitive for longer and prevent the stagnation that can occur when software is limited by hardware — all leading to more interesting and unexpected AI applications for more organizations.

Widespread adoption of real AI and ML technologies

I’ll be the first to admit to tech’s fascination with shiny objects. There was a day when big data was the solution to everything and IoT was to be the world’s savior. AI has been through the hype cycle in the same way (arguably multiple times). Today, you’d be hard pressed to find a tech company that doesn’t purport to use AI in some way, but chances are they are doing something very rudimentary that’s more akin to advanced analytics.

It’s my firm belief that the AI revolution we’ve all been so excited about simply has not happened yet. In the next two to three years however, as the hardware that enables “real” AI power makes its way into more and more hands, it will happen. As far as predicting the change and disruption that will come from widespread access to the upper echelons of powerful ML and AI — there are few ways to make confident predictions, but that is exactly the point!

Much like cellphones put so much power in the hands of regular people everywhere, with no barriers to entry either technical or financial (for the most part), so will the coming wave of software-defined hardware that is flexible, customizable and future-proof. The possibilities are truly endless, and it will mark an important turning point in technology. The ripple effects of AI democratization and commoditization will not stop with just technology companies, and so even more fields stand to be blown open as advanced, high-powered AI becomes accessible and affordable.

Much of the hype around AI — all the disruption it was supposed to bring and the leaps it was supposed to fuel — will begin in earnest in the next few years. The technology that will power it is being built as we speak or soon to be in the hands of the many people in the many industries who will use their newfound access as a springboard to some truly amazing advances. We’re especially excited to be a part of this future, and look forward to all the progress it will bring.

#artificial-general-intelligence, #artificial-intelligence, #column, #hardware, #machine-learning, #ml, #natural-language-processing, #neural-networks, #science

Nvidia developed a radically different way to compress video calls

Instead of transmitting an image for every frame, Maxine sends keypoint data that allows the receiving computer to re-create the face using a neural network.

Enlarge / Instead of transmitting an image for every frame, Maxine sends keypoint data that allows the receiving computer to re-create the face using a neural network. (credit: Nvidia)

Last month, Nvidia announced a new platform called Maxine that uses AI to enhance the performance and functionality of video conferencing software. The software uses a neural network to create a compact representation of a person’s face. This compact representation can then be sent across the network, where a second neural network reconstructs the original image—possibly with helpful modifications.

Nvidia says that its technique can reduce the bandwidth needs of video conferencing software by a factor of 10 compared to conventional compression techniques. It can also change how a person’s face is displayed. For example, if someone appears to be facing off-center due to the position of her camera, the software can rotate her face to look straight instead. Software can also replace someone’s real face with an animated avatar.

Maxine is a software development kit, not a consumer product. Nvidia is hoping third-party software developers will use Maxine to improve their own video conferencing software. And the software comes with an important limitation: the device receiving a video stream needs an Nvidia GPU with tensor core technology. To support devices without an appropriate graphics card, Nvidia recommends that video frames be generated in the cloud—an approach that may or may not work well in practice.

Read 27 remaining paragraphs | Comments

#generative-adversarial-networks, #machine-learning, #maxine, #neural-networks, #nvidia, #nvidia-maxine, #science, #tech

Disney Research neural face-swapping technique can provide photorealistic, high-resolution video

A new paper published by Disney Research in partnership with ETH Zurich describes a fully automated, neural network-based method for swapping faces in photos and videos — the first such method that results in high-resolution, megapixel resolution final results according, to the researchers. That could make it suited for use in film and TV, where high-resolution results are key to ensuring that the final product is good enough to reliably convince viewers as to their reality.

The researchers specifically intend this tech for use in replacing an existing actor’s performance with a substitute actor’s face, for instance when de-aging or increasing the age of someone, or potentially when portraying an actor who has passed away. They also suggest it could be used for replacing the faces of stunt doubles in cases where the conditions of a scene call for them to be used.

This new method is unique from other approaches in a number of ways, including that any face used in the set can be swapped with any recorded performance, making it possible to relatively easily re-image the actors on demand. The other is that it kindles contrast and light conditions in a compositing step to ensure the actor looks like they were actually present in the same conditions as the scene.

You can check out the results for yourself in the video below (as the researchers point out, the effect is actually much better in moving video than in still images). There’s still a hint of “uncanny valley” effect going on here, but the researchers also acknowledge that, calling this “a major step toward photo-realistic face swapping that can successfully bridge the uncanny valley” in their paper. Basically it’s a lot less nightmare fuel than other attempts I’ve seen, especially when you’ve seen the side-by-side comparisons with other techniques in the sample video. And, most notably, it works at much higher resolution, which is key for actual entertainment industry use.

The examples presented are a super small sample, so it remains to be seen how broadly this can be applied. The subjects used appear to be primarily white, for instance. Also, there’s always the question of the ethical implication of any use of face-swapping technology, especially in video, as it could be used to fabricate credible video or photographic “evidence” of something that didn’t actually happen.

Given, however, that the technology is now in development from multiple quarters, it’s essentially long past the time for debate about the ethics of its development and exploration. Instead, it’s welcome that organizations like Disney Research are following the academic path and sharing the results of their work, so that others concerned about its potential malicious use can determine ways to flag, identify and protect against any bad actors.

#artificial-intelligence, #disney-research, #film, #graphics, #media, #neural, #neural-networks, #tc, #television, #video, #video-production, #visual

TinyML is giving hardware new life

Aluminum and iconography are no longer enough for a product to get noticed in the marketplace. Today, great products need to be useful and deliver an almost magical experience, something that becomes an extension of life. Tiny Machine Learning (TinyML) is the latest embedded software technology that moves hardware into that almost magical realm, where machines can automatically learn and grow through use, like a primitive human brain.

Until now building machine learning (ML) algorithms for hardware meant complex mathematical modes based on sample data, known as “training data,” in order to make predictions or decisions without being explicitly programmed to do so. And if this sounds complex and expensive to build, it is. On top of that, traditionally ML-related tasks were translated to the cloud, creating latency, consuming scarce power and putting machines at the mercy of connection speeds. Combined, these constraints made computing at the edge slower, more expensive and less predictable.

But thanks to recent advances, companies are turning to TinyML as the latest trend in building product intelligence. Arduino, the company best known for open-source hardware is making TinyML available for millions of developers. Together with Edge Impulse, they are turning the ubiquitous Arduino board into a powerful embedded ML platform, like the Arduino Nano 33 BLE Sense and other 32-bit boards. With this partnership you can run powerful learning models based on artificial neural networks (ANN) reaching and sampling tiny sensors along with low-powered microcontrollers.

Over the past year great strides were made in making deep learning models smaller, faster and runnable on embedded hardware through projects like TensorFlow Lite for Microcontrollers, uTensor and Arm’s CMSIS-NN. But building a quality dataset, extracting the right features, training and deploying these models is still complicated. TinyML was the missing link between edge hardware and device intelligence now coming to fruition.

Tiny devices with not-so-tiny brains

#arduino, #artificial-intelligence, #artificial-neural-networks, #biotech, #cloud, #column, #coronavirus, #covid-19, #deep-learning, #drug-development, #embedded-systems, #extra-crunch, #gadgets, #hardware, #machine-learning, #manufacturing, #market-analysis, #ml, #neural-networks, #open-source-hardware, #robotics, #saas, #wearables

Immunai wants to map the entire immune system and raised $20 million in seed funding to do it

For the past two years the founding team of Immunai had been working stealthily to develop a new technology to map the immune system of any patient.

Founded by Noam Solomon, a Harvard and MIT-educated postdoctoral researcher, and former Palantir engineer, Luis Voloch, Immunai was born from the two men’s interest in computational biology and systems engineering. When the two were introduced to Ansuman Satpathy, a professor of cancer immunology at Stanford University, and Danny Wells, who works as a data scientist at the Parker Institute for Cancer Immunotherapy the path forward for the company became clear.

“Together we said we bring the understanding of all the technology and machine learning that needs to be brought into the work and Ansu and Danny bring the single-cell biology,” said Solomon. 

Now as the company unveils itself and the $20 million in financing it has received from investors including Viola Ventures and TLV Partners, it’s going to be making a hiring push and expanding its already robust research and development activities. 

Immunai already boasts clinical partnerships with over ten medical centers and commercial partnerships with several biopharma companies, according to the company. And the team has already published peer-reviewed work on the origin of tumor-fighting T cells following PD-1 blockade, Immunai said.

“We are implementing a complicated engineering pipeline. We wanted to scale to hundreds of patients and thousands of samples,” said Wells. “Right now, in the world of cancer therapy, there are new drugs coming on the market that are called checkpoint inhibitors. [We’re] trying to understand how these molecules are working and find new combinations and new targets. We need to see the immune system in full granularity.”

That’s what Immunai’s combination of hardware and software allows researchers to do, said Wells. “It’s a vertically integrated platform for single cell profiling,” he said. “We go even further to figure out what the biology is there and figure that out in a new combination design for the trial.”

Cell therapies and cancer immunotherapies are changing the practice of medicine and offering new treatments for conditions, but given how complex the immune system is, the developers of those therapies have few insights into how their treatments will effect the immune system. Given the diversity of individual patients variations in products can significantly change the way a patient will respond to the treatment, the company said.

Photo: Andrew Brookes/Getty Images

Immunai has the potential to change the way these treatments are developed by using single-cell technologies to profile cells by generating over a terabyte of data from an individual blood sample. The company’s proprietary database and machine learnings tools map incoming data to different cell types and create profiles of immune responses based on differentiated elements. Finally, the database of immune profiles supports the disvovery of biomarkers that can then be monitored for potential changes.

“Our mission is to map the immune system with neural networks and transfer learning techniques informed by deep immunology knowledge,” said Voloch, in a statement. “We developed the tools and knowhow to help every immuno-oncology and cell therapy researcher excel at their job. This helps increase the speed in which drugs are developed and brought to market by elucidating their mechanisms of action and resistance.”

Pharmaceutical companies are already aware of the transformational potential of the technology, according to Solomon. The company is already in the process of finalizing a seven-figure contract from a Fortune 100 company, according to Solomon. 

One of the company’s earliest research coups was using research to show the way that immune systems function when anti-PD1 molecules are introduced. Typically the presence of PD-1 means that t-cell production is being suppressed. What the research from ImmuneAI revealed was that the response wasn’t happening with T-cells within the tumor. There were new t-cells that were migrating to the tumor to fight it off, according to Wells.

“This whole approach that we have around looking at all of these indications — we believe that the right way and most powerful way to study these diseases is to look at the immune system from the top down,” said Voloch, in an interview. “Looking at all of these different scenarios. From the top, you see these patterns than wouldn’t be available otherwise.” 

#cancer, #cancer-immunotherapy, #data-scientist, #engineer, #harvard, #machine-learning, #mit, #neural-networks, #palantir, #stanford-university, #tc, #tlv-partners, #tumor

Hailo raises $60M Series B for its AI chips

Israeli AI chipmaker Hailo today announced that it has raised a $60 million Series B funding round led by its existing investors, who were joined by new strategic investor ABB Technology Ventures, the venture arm of the Swiss-based multination ABB, NEC Corporation and Londons’ Latitude Ventures. The new funding will help Hailo to roll out its Hailo-8 Deep Learning chip and to get into new markets and industries.

“This immense vote of confidence from our new strategic and financial investors, along with existing ones, is a testimony to our breakthrough innovation and market potential,” said Orr Danon, CEO and co-founder of Hailo. “The new funding will help us expedite the deployment of new levels of edge computing capabilities in smart devices and intelligent industries around the world, including areas such as mobility, smart cities, industrial automation, smart retail and beyond.”

I last met with the Hailo team at CES in January. At the time, the company was showing off a number of impressive demos, mostly around real-time image recognition. What makes the Hailo chip stand out is its innovative architecture, which can automatically adapt resources to best run its users’ custom neural networks. With this, the chip doesn’t just run faster but is also far more energy efficient. The company promises 26 tero operations per second in performance from its chip, which it says “outperforms all other edge processors with its small size, high performance, and low power consumption.”

With this round, Hailo’s total funding is now $88 million. In part, the investor enthusiasm for Hailo is surely driven by the success of other Israeli chip startups. Mobileye, after all, went to Intel for $15.3 billion, which also recently acquired Habana Labs. And the time, of course, is ripe for deep learning chips at the edge now that AI/ML technology is quickly becoming table stakes.

“Hailo is poised to become a defining player in the rapidly emerging market for AI processors,” said Julian Rowe, partner at Latitude Ventures. “Their Deep Learning edge chip can be disruptive to so many sectors today, while the new, innovative use cases Hailo’s chips can unlock are just starting to reveal themselves. We’re thrilled to join the team for what lies ahead.”

#artificial-intelligence, #deep-learning, #habana-labs, #hailo, #hardware, #intel, #mobileye, #neural-networks