Posts tagged with "AI"

CES 2021: The Latest in Innovation

CES, the Consumer Electrionics Show, convened digitally this year. This is the first time that CES has ever been conducted fully virtually due to the ongoing Coronavirus pandemic. The event produced and owned by the Consumer Technology Association (CTA) featured technology related to Digital Health, Smart Cities, Immersive Entertainment, Lifestyle, 5G and Internet of Things (IOT), Artificial Intelligence, Automotive, Startups, Tech Enabling the New Normal, and more. Some notable brands who participated in the conference portion of the event include Sony, GM, Intel, Audi, Panasonic, Samsung, and Mercedes-Benz. Despite the virtual interface, the overall reception of this year’s event was positive.

More than 500 companies participated in the exhibition event, though some garnered more social media buzz than others. According to analytics from Talkwalker, a social listening and analytics company, Samsung had 11 thousand social media mentions related to the event. This was aided by the ever present social media monster that is the BTS army who blew up this tweet before the event had even started. Some other products that generated a lot of buzz are Razer’s N95 facemask with gaming lighting and over this Keurig-but-it’s-ice-cream contraption.

The next CES show will be January 5-8, 2022.

Gabrielle Illustrates a Rolls-Royce Article for 360 MAGAZINE

Rolls-Royce Art Programme

Rolls-Royce announced the shortlist of artists for its inaugural moving-image Dream Commission.

Sondra Perry, Beatriz Santiago Muñoz, Martine Syms and Zhou Tao all made the shortlist for Muse, Rolls-Royce’s Art Programme.

The Dream Commission finds mid-career artists to be best representative and the best creators of moving-image concept artwork, and these artists were all tasked with creating art depicting dreams. The winner will be declared in 2021.

Torsten MüllerÖtvös, chief executive for Rolls-Royce said he looks for artists who share Rolls-Royce’s passion for pushing technical and conceptual boundaries.

“We are delighted to announce four shortlisted moving-image artists for the inaugural Dream Commission, each of whom have outstanding reputations,” MüllerÖtvös said. “The art of moving-image is a creative and avant-garde genre and we are pleased to be supporting this medium at this critical time for the industry. To commission artists during a pandemic is an act of determination and faith in the power of culture to inform and transform our lives; a quality at Rolls-Royce that we fully endorse.”

All finalists were selected by an international jury, and each created a work of short-form moving-image art to explore the notion of “Dreams.”

Sondra Perry has a wide range of tools, including AI, animation, performance and video. She uses them to explore race, identity and technology. Her piece, titled “Lineage for a Phantom Zone,” looks at lineage, longing and memory through her own archived footage.

“The piece begins with me playing a theremin, using the touchless liminal instrument to conjure a dream space with multi-dimensional sound,” she said. “Growing up, my grandmother had a picture of herself on the land she was raised on in North Carolina on her dresser. I think about that picture often and I wanted to reflect on her history, that land, and my experience of it through images. I collapse time, space, and two generations of family to visualize a life dream that was mutated through imagination, images, and video and passed down through my DNA.”

Beatriz Santiago Muñoz is based in Puerto Rico and is an artist and filmmaker. She makes films after lengthy research, observation and documentation, and her piece, titled “The Source,” uses photos of people, places and experiences to strike the viewer.

She uses Puerto Rico to elicit the emotions and ideas of the piece, and it comes with a strong sense of history and culture. She said her vision for “The Source” begins with an experience.

“My son is standing at the source of the Río Caguitas. Something seems to stop. Is it the ancient rock, the loud cold water pushing, the slowness above us, the smell? An interval,” she said. “Reading his translation of Proust to Haitian Kreyòl, a project undertaken for its own sake and without readers in mind, my colleague writer and translator Guy Regis Junior had said the task had been both full of pleasure and sacrificial. From there, there is an opening- up in the interval, in that time/space between one language, history and sensorial world, and another”

Martine Syms is based in Los Angeles and combines grit, humor and social commentary in her art. Her entry, titled “Kita’s World,” dives into Syms as a person biologically, psychologically and sociologically.

She said her world is a combination of “core material, broken samples, seductive loops and heavy theory.”

“Symptomatic of the contemporary condition, I was inspired by an anecdote by a prominent theorist in which intimate technology appears to read our minds,” she said.  “Everything has a subtext, ulterior motives—but tech flattens everything out. It can speak our unconscious; we unravel equally in realms both real and digital. There is a dissolution of difference but no real plurality. I’m using Kita, an homage to an avatar from my childhood, to think through this tension.”

Zhou Tao of Guangzhou uses video, drawing and photography together. His submission, titled “Three Hundred Miles Southwest,” makes populated areas seem mythological.

“Three hundred miles from the dangerous peak to the southeast, those forgotten areas not covered by the high-speed network are at the end of the geography,” he said. “Between a wolf seeker with mountains as a companion and the 37th ‘remote style’ ecological model; between the giant reliefs in the narrow valley and the legendary gate into the four-dimensional; the engineering bases connected from one terminal to another scattered among the mountains, presenting a future fable that has long passed away of this mythological place.”

Following the nominations of 23 artists, these four were chosen unanimously by the jury, and the statement from that jury reads as follows:

“The Dream Commission offers an opportunity for artists to have a space to develop their aesthetics and to be able to delve deeply into an area where they can have an autonomy to make a work which can resonate. The quality of the long list that was presented to us made this an incredibly engaging, but also difficult selection process. The breadth of practice that was selected for us to consider was extraordinary – the sophistication of ideas and expression across this media was so inspiring. We have succeeded in selecting a variety of artists from different countries, cultures and different kinds of artistic thinking.”

The Dream Commission accepts entries from any medium used to create moving-image art, like experimental film, video, animation, immersive and participatory installations and content presented in non-screen formats.

Once this cycle concludes after two years, another cycle will begin to welcome in a brand new set of landmark artists and creators.

To stay up to date with Muse and the Dream Commission, you can follow them on Instagram.

You can also click right here to learn more.

Global AI Spending to surge by 120% and hit $110bn by 2024

By Jastra Kranjec

Recent years have witnessed a swell in the adoption of artificial intelligence solutions, revolutionizing industries, and helping businesses boost growth. The rising volume and complexity of business data are set to continue driving AI adoption in the following years, causing a surge in global AI spending.

According to data presented by BuyShares.co.nz, global artificial intelligence spending is expected to surge by 120% and hit $110bn by 2024.

Global AI Spending Jumped 33% YoY, Despite COVID-19 Crisis

Businesses across the world use AI technology to be innovative and scalable. Using automation, deep learning, and natural language processing can improve their decision-making, efficiency, speed, and help predict trends.

In 2015, companies and organizations worldwide spent $5bn on implementing AI systems in their business, revealed the IDC 2020 Worldwide Artificial Intelligence Systems Spending Guide. In the next three years, this figure jumped five times to $25bn. Statistics show that 2019 witnessed a $37.5bn worth of investments into AI business solutions, a 650% jump in four years.

Increased investments in AI technology continued in 2020, with organizations expected to invest $50.1bn in AI systems, despite the COVID-19 crisis. The following years are set to witness remarkable growth in global AI spending, with the figure surging by almost 120% to $110bn by 2023. 

Automated customer service, sales process automation, automated threat intelligence and prevention, and IT automation were the leading use cases for AI in 2020, accounting for nearly a third of total AI spending this year. However, the IDC data show that automated human resources, IT automation and pharmaceutical research and discovery are the fastest-growing use cases.

Life Sciences and Retail Lead in Adoption of AI

The IDC data indicate the retail industry and the banking sector are expected to spend the most on AI solutions in 2020. The retail companies primarily focused their AI investments on improving customer experience via chatbots and recommendation engines. Banks are expected to keep investing in AI-driven fraud prevention and program advisors. Discrete manufacturing, process manufacturing, and healthcare round out the top five industries for AI spending this year.

The life sciences sector, including biotech, pharma and biomedical companies, has the most significant share of organizations that have adopted AI, revealed the Capgemini`s AI-Powered Enterprise survey.

Statistics show that 67% of organizations operating in this market adopted AI at scale, while another 33% launched AI pilots that are still undeployed in production. The retail industry ranked second, with 51% of companies utilizing artificial intelligence technology. The consumer products sector follows with a 44% share.

The Capgemini data show the automotive industry represents the fourth-leading sector, with 17% of companies successfully using AI in production. Another 49% of automotive companies have deployed a few use cases in production on a limited scale. The telecom industry follows, with a 14% and 57% share, respectively.

The full story can be read here: https://buyshares.co.nz/2020/10/20/global-artificial-intelligence-spending-to-surge-by-120-and-hit-110bn-by-2024/

AI brain chip illustration by UMD featured in 360 MAGAZINE.

UMD iSchool to Investigate AI Tech for Intelligence Analysts

University of Maryland College of Information Studies (UMD iSchool) researchers, led by principal investigator Dr. Susannah Paletz, have been awarded a three-year $616,700 grant funded by the Army Research Office (ARO). ARO Program Manager Dr. Edward Palazzolo overseas it. This project examines how teams of intelligence analysts can work together and with artificial intelligence (AI). AI has the potential to support intelligence analysts in reviewing potentially hundreds of thousands of source documents, pulling out key findings, and assembling them into actionable intelligence. AI can also aid in the flow of information and projects among members of the intelligence team, improving the efficiency and accuracy of their work.

“AI-driven technology has sometimes been touted as a replacement for human intelligence,” said Dr. Adam Porter, the project’s co-principal investigator, professor at the UMD Department of Computer Science, and Executive and Scientific Director of the Fraunhofer USA Center for Experimental Software Engineering (CESE). “In practice, however, AI doesn’t always work, or gives limited or biased answers. Human oversight is still required, and it’s therefore critical that we deeply understand how humans and AI can work best together.”

The Human-Agent Teaming on Intelligence Tasks project coordinated through the iSchool will focus on two particular research areas; 1.) how interactive AI agents, such as chatbots, have the ability to mitigate or exacerbate the communication and coordination problems that can occur with shift handovers of intelligence work, such as inaccuracy blindness and overlooking potentially relevant information, and 2.) examining how humans could potentially deal with these blind spots, biases, or inaccuracies.

The Experiment

The research team plans to develop an experimental infrastructure to help test team cognition challenges within the work completed by intelligence analysts consisting of task-relevant input materials, such as mission descriptions and source documents, activity recording tools, experimental monitoring capabilities, and different AI supports for human analysts, such as chatbots offering advice on a particular task.

“We want to develop a task that can raise the problems with asynchronous team cognition in intelligence tasks, but is simple enough to be used by research participants with minimal training,” said Dr. Susannah B.F. Paletz, research professor at the UMD iSchool, and affiliate at the UMD Applied Research Laboratory for Intelligence and Security (ARLIS).

This task will substantially increase insight into the strengths and weaknesses of AI technology to support intelligence task. In addition, it will help shed light on how and when human analysts can safely place their trust in AI technology. Also, how they can proactively identify problems in AI-generated input. It will also aid teams of humans, including asynchronous teams, working together in situations that include AI-generated input.

“This basic research is an important step in the early process of learning how humans and agents can collaboratively become a single team with considerably greater capacity and productivity than human only teams,” Palazzolo said. “Moreover, this research has broad implications into the work of many teams focused on knowledge work and information management such as medical teams involved in shift work, collaborative software development teams, and research teams.”

Collaborators

In addition to Porter, the Fraunhofer USA team also includes Dr. Madeline Diep, Senior Scientist, and Jeronimo Cox, Software Developer, at Fraunhofer USA CESE. The Fraunhofer USA team will lead the effort to create configurable AI agents used in the experimental tasks. Also, it will create a data collection and analysis infrastructure for capturing and understanding participant behaviors.

The UMD iSchool team includes graduate students Tammie Nelson, a fourth year PhD student, Melissa Carraway, incoming first year PhD student, and Sarah Valhkamp, incoming first year PhD, in Information Science.

The grant proposal team includes UMD Office of Research Administration Contract Manager, Stephanie Swann; iSchool Business Manager, Jacqueline Armstrong. Also, former iSchool Business Manager, Lisa Geraghty.

Outside of UMD, Dr. Aimee Kane, the Harry W. Witt Faculty Fellow and an Associate Professor of Management in the Palumbo-Donahue School of Business at Duquesne University, will be a consultant and an intellectual contributor on this project.

ARO is an element of the U.S. Army Combat Capabilities Development Command’s (CCDC) Army Research Laboratory. The Human-Agent Teaming on Intelligence Tasks project (grant no. W911NF-20-1-0214) runs through June 30, 2023.

About the University of Maryland College of Information Studies

Founded in 1965 and located just outside of Washington, D.C., the University of Maryland College of Information Studies (UMD iSchool) is top-ten ranked research and teaching college in the field of information science. UMD iSchool faculty, staff, and students are expanding the frontiers of how people access and use information and technology in an evolving world – in government, education, business, social media, and more. The UMD iSchool is committed to using information and technology to empower individuals and communities, create opportunities, ensure equity and justice, and champion diversity.

For more information click here.

Uzy Rozenthal Botswana

Botswana eVisa Applications

Travelers to Botswana will soon be able to complete visa applications online and ease their entry into the country with Pangea IT. The new service will be hugely beneficial to travelers as well as reduce the government’s investment of funds, time, paperwork, and manpower allocated for visa applications.

Israel’s Pangea IT, which specializes in digital transformation of government and business services, has been selected by the government of Botswana to integrate its eVisa solution. This will enable visitors to Botswana to obtain a visa online and have a seamless entry upon arrival at all the land and air arrival points in the country.

Online visas are becoming more popular as both governments and travelers are increasingly embracing the digital age. The Covid-19 global health crisis is expected to accelerate the trend, as tourists will be looking at reducing their contacts in their home countries and at the travel destination. In the case of Botswana, this is extremely important since the country has a limited number of diplomatic offices abroad so the eVisa option will make traveling to the popular safari destination easier for visitors from various countries.

Pangea has been active in Botswana since 2003 when the company installed a digital birth and death registration system which is still in operation. In 2010, the Israeli company deployed a computerized immigration and citizenship system (ICS) in Botswana.

“We are in the process of integrating our eVisa solution and expect to have it operational till the end of the year,” said Uzy Rozenthal, Pangea EVP and general manager government division. “eVisa will dramatically speed up the entry process in Botswana and will translate into a substantial savings in both administration costs and time needed without compromising border security.”

Vaughn Lowery, 360 MAGAZINE

Why you need visibility into invoices

(and how AI can help you get it)

We’ve covered how having visibility into employee expense reports helps organizations catch errors, waste, and fraud, and streamline process so that employees get reimbursed faster. But there’s another area of business spend where visibility might be even more critical: invoices. 

Enterprise AP departments pay thousands of invoices every month. Overworked AP teams may be hard pressed to scrutinize every invoice manually for adherence to contract terms, pricing anomalies, gradual increases in invoice amounts, or patterns that could indicate fraud, such as repeated invoices that fall just below the PO limit. Even worse, criminals can exploit an AP department’s lack of bandwidth by sending invoices for products that were never delivered or services that were never rendered, sometimes from companies that don’t even exist. 

Of course, the vast majority of vendors are trustworthy and want to earn and retain their customers’ trust, but with humans in the payment processing equation, honest mistakes are bound to happen. 

The risk hiding in vendor invoices

We recently reviewed the aggregated, anonymized data from billions of audit transactions across hundreds of customer accounts in a variety of industries, and summarized the results in our quarterly report, The State of AI in Business Spend. We found that, for the average enterprise, invoices comprise 96% of their business spend. (In comparison, employee expenses for travel and entertainment (T&E) make up only 3.7% of spend). The average company processes 60,354 invoices every month, but only audits or reviews at most 10% of them, usually after payment. In other words, most companies only find mistakes after the money is out the door, when clawing it back is expensive and time consuming.

Our report also revealed that 4% of invoices could be considered high risk. These invoices generally fell into three main categories:

•Prices, discounts or terms didn’t match the contract. Procurement teams may work hard to negotiate great terms with vendors, but if AP doesn’t ensure that invoices reflect those terms, that effort is wasted.

•Vendors billed inflated prices compared to the market. Most AP teams don’t have time to see whether better rates are available elsewhere.

•Duplicates. We uncovered double billing that may or may not have been accidental, as well as amounts duplicated on expense reports. 

Why visibility into high-risk invoices is critical

We probably don’t have to work that hard to convince you that incorrect invoices hurt your bottom line. But you might be surprised at how small mistakes, intentional or not, can add up. 

For example, we’ve found that many invoices don’t align with the signed contract, and the most frequent error is net payment terms: A contract may list payment terms as Net 60, but the invoices list Net 15 or Net 30. This difference can have a huge impact on your cash flow…and even profit.

Duplicate charges or payments happen with surprising frequency. Often, after an invoice is held up, an employee may intervene and approve manual payment of the invoice, but when the system clears the hold, the invoice is paid again. Even if the vendor notifies you about the duplicate payment, the time and energy everyone will spend trying to fix the mistake could be better spent elsewhere. 

But it’s the big mistakes, like fraud and non-compliance, that can cost your organization not only money but something that’s hard to replace: its reputation. Invoice fraud is real, and even large companies fall victim to it: A Lithuanian man recently bilked Facebook and Google out of more than $100 million by impersonating a vendor with which the tech titans do business.

What’s more, our report found that for every 10,000 invoices, at least one contains a regulatory violation. For example, a regional sales director might funnel payments to a distributor for fake “logistics services” that are actually a bribe to a government official who influences reimbursement policy for your company’s product. In a real-life story that illustrates the potential consequences, a large multinational retailer will have to pay the U.S. government $282 million for violating the Foreign Corrupt Practices Act (FCPA), in part because it failed to institute sufficient internal accounting controls related to third-party payments.

How AI can help

Given the value at risk, many enterprises are embracing AI as a way to get visibility into invoices—before they pay vendors—for errors, fraud, and out-of-compliance spend. Companies that use AI achieve 100% visibility into their invoices; companies that don’t use AI get at most 10% visibility.

AI extracts and analyzes key pieces of information on every invoice to catch duplicate charges, enforce payment terms, identify missing discounts, eliminate overcharges, catch suspicious activity, and flag compliance issues. This frees your AP team to focus on tasks that will add more value to your financial processes—and helps improve your bottom line by stopping unnecessary outflows. 

To understand the magnitude of the issue and see what 100% visibility into business spend means to you, download our latest research report, The State of AI in Business Spend. The findings focus on spend visibility, value at risk in invoices, insights on streamlining the spend audit process, recommendations for finance teams, and more.

Claire Chen is a Business Operations Analyst at AppZen, where she’s passionate about providing simple solutions for complicated data.

Roybi Robot, 360 MAGAZINE, ai, tech, kids, children, youth, school

ROYBI ROBOT – AI-powered EdTech

A growing number of states say their schools will stay closed for the rest of the 2019-20 academic year to stem the Coronavirus outbreak. At Roybi Robot, a leader in AI-powered EdTech and personalized education, they know first hand the importance of AI in connection with remote education and learning.

At ROYBI, they’re already noticing a big shift towards remote learning due to the recent circumstances and headlines. And throughout this all, one thing seems inevitable: school settings, as they stand today, will change. Online and remote learning will be systems that educational institutions will adopt for future emergencies. They envision a future where the new culture of learning begins at home through devices with sophisticated AI technology such as Roybi Robot. Artificial Intelligence allows educators to follow the child’s progress in a smarter way and provides a personalized approach to each child individually. Additionally, it provides a closer collaboration between parents and educators, because it can connect in a joint force to education.

With many uncertainties around the school closures, many educators have already started approaching distance and remote learning in the long term, but lack of personalized attention and progress tracking has been a major challenge for them. The role of artificial intelligence becomes even more significant for a modern world as it can monitor each child individually and provide feedback to educators more accurately than traditional approaches.

At Roybi, they are NOT saying to eliminate school and the classroom. They are saying that to save time and cost, we can be educating children more at home (by the educators) and use AI to personalize the educational experience for each child. They envision a future where they can connect learners, parents, educators, and even their Roybi Robots together while creating an engaging and interactive learning experience.

Aiways, 360 MAGAZINE

AIWAYS RE-STARTS PRODUCTION

AIWAYS, the Shanghai-based personal mobility provider, is to start taking online orders for its U5 all-electric SUV from European consumers from the end of April. Secured via a small deposit, the U5 will be offered exclusively via a direct-to-customer sales model, and not retailed or leased through traditional dealerships. European pre-sale markets and the required deposit amount will be announced by AIWAYS in April.

Alexander Klose, Executive VP Overseas Operation at AIWAYS, commented: “Online pre-sales represents the next important phase of AIWAYS’ entry into the European market. It’s our promise to customers that for only a small deposit they can be among the first to receive the U5 and start enjoying the benefits of a long range, high-tech and well equipped electric SUV.”

Meanwhile, AIWAYS has re-started production of the U5 at its manufacturing facility in Shangrao, China, following the interruption caused by COVID-19 (coronavirus). Production of the European U5 will start in July, with the first deliveries now slated for August 2020.

Making the most of its agility and flexibility as a startup, AIWAYS is adapting its pre-sale marketing activities to better suit the enforced period of ‘contactless’ retail because of COVID-19. By introducing new online platforms and seamless digital experiences, AIWAYS will give European car buyers the confidence to order the U5 online. More details to follow soon.

Microsoft AI x “Bode Vault”

Emily Adams Bode and Microsoft AI introduce the ‘Bode Vault:’ the first-ever digital quilt expert

Microsoft develops a custom-built digital library of thousands of hand-sourced materials, streamlining the NYC-based designer’s approach to creative process and production

Highly celebrated New York fashion designer Emily Adams Bode has partnered with Microsoft to develop the world’s first AI-powered digital quilt expert that draws upon an interactive archive of antique patterns as a comprehensive resource for design, production, inventory management, and storytelling. As the platform grows and develops its own internal intelligence, the Bode Vault is set to become an industry-changing tool for pairing the rich tapestry of human history with the best of machine insight.

Emily Adams Bode, the founder of eponymous luxury menswear brand Bode, was the first female designer to show at Men’s New York Fashion Week, the winner of the 2019 CFDA Award for Emerging Designer of the Year, and most recently granted the Inaugural Karl Lagerfeld Award for Innovation . Passionate about traditional garment making, her clothing is a nostalgic nod to American craftsmanship and the pieces she designs are deeply personal, each employing the timeless and historic garment-making technique: quilting. Marrying modern workwear silhouettes with the female-centric traditional techniques of quilting and applique, each piece is laden with its own history. All of the garments in the Bode line have a unique narrative that are based on a specific person or moment in history.

“I am excited to partner with Microsoft to help tell and share the stories of American quilts. This partnership uses technology to help us identify quilts in the field and in our stores. I aim to reinvigorate the rich history behind these textiles for my design team as we continue to grow with the potential of it becoming public facing.” says Emily Adams Bode.

With increasing demand, Bode is faced with the challenge of finding ways of organizing and categorizing an extensive inventory of bespoke raw materials — she looked to Microsoft as a long-time technical innovator within the industry to develop a solution for streamlining growth and efficiency. Initially trained on Bode’s existing quilt collection that was compiled by her team, as well as Bode’s extensive network of quilt experts and vendors to establish a rich historical reference toolset from which the Microsoft AI technologies could evolve. The result is a powerful design-to-production platform that draws from history’s most comprehensive library of textiles that is reshaping the company’s internal process, with the potential to do so for the fashion industry at large. Built with Microsoft Custom Vision AI and an integrated database system called Cosmos DB an extension of Microsoft Azure, the archive is trained on hundreds of images to make up a nuanced library that contains an archive of historic quilts of various patterns. From keeping track of inventory to offering up niche institutional knowledge, the archive is also an educational tool that it serves a host of essential functions for the business.

“The partnership with Bode demonstrates how Microsoft technology is unlocking creative and enterprise innovation for fashion designers and the industry,” says Maruschka Loubser, Director of Brand Partnerships at Microsoft. “Particularly for smaller companies, this allows them to rethink their process and structure with technology as a collaborative tool.”

For more information visit bodenewyork.com and microsoft.com/inculture.

Emily Adams Bode,  Bode Vault, Microsoft, Vaughn Lowery, Sam Berman, 360  Magazine,
Emily Adams Bode, Bode Vault, Microsoft,  Vaughn Lowery, 360 Magazine

Rice University x SLIDE

Deep learning rethink overcomes major obstacle in AI industry SLIDE is first algorithm for training deep neural nets faster on CPUs than GPUs

Rice University computer scientists have overcome a major obstacle in the burgeoning artificial intelligence industry by showing it is possible to speed up deep learning technology without specialized acceleration hardware like graphics processing units (GPUs). Scientists from Rice, supported by collaborators from Intel, will present their results today at the Austin Convention Center as a part of the machine learning systems conference MLSys.

Many companies are investing heavily in GPUs and other specialized hardware to implement deep learning, a powerful form of artificial intelligence that’s behind digital assistants like Alexa and Siri, facial recognition, product recommendation systems and other technologies. For example, Nvidia, the maker of the industry’s gold-standard Tesla V100 Tensor Core GPUs, recently reported a 41% increase in its fourth quarter revenues compared with the previous year.

Rice researchers created a cost-saving alternative to GPU, an algorithm called “sub-linear deep learning engine” (SLIDE) that uses general-purpose central processing units (CPUs) without specialized acceleration hardware.

“Our tests show that SLIDE is the first smart algorithmic implementation of deep learning on CPU that can outperform GPU hardware acceleration on industry-scale recommendation datasets with large fully connected architectures,” said Anshumali Shrivastava, an assistant professor in Rice’s Brown School of Engineering who invented SLIDE with graduate students Beidi Chen and Tharun Medini SLIDE doesn’t need GPUs because it takes a fundamentally different approach to deep learning. The standard “back-propagation ” training technique for deep neural networks requires matrix multiplication, an ideal workload for GPUs. With SLIDE, Shrivastava, Chen and Medini turned neural network training into a search problem that could instead be solved with hash tables, his radically reduces the computational overhead for SLIDE compared to back-propagation training. For example, a top-of-the-line GPU platform like the ones Amazon, Google and others offer for cloud-based deep learning services has eight Tesla V100s and costs about $100,000, Shrivastava said.

“We have one in the lab, and in our test case we took a workload that’s perfect for V100, one with more than 100 million parameters in large, fully connected networks that fit in GPU memory,” he said. “We trained it with the best (software) package out there, Google’s TensorFlow, and it took 3 1/2 hours to train.

“We then showed that our new algorithm can do the training in one hour, not on GPUs but on a 44-core Xeon-class CPU,” Shrivastava said.

Deep learning networks were inspired by biology, and their central feature, artificial neurons, are small pieces of computer code that can learn to perform a specific task. A deep learning network can contain millions or even billions of artificial neurons, and working together they can learn to make human-level, expert decisions simply by studying large amounts of data. For example, if a deep neural network is trained to identify objects in photos, it will employ different neurons to recognize a photo of a cat than it will to recognize a school bus. “You don’t need to train all the neurons on every case,” Medini said. “We thought, ‘If we only want to pick the neurons that are relevant, then it’s a search problem.’ So, algorithmically, the idea was to use locality-sensitive hashing to get away from matrix multiplication.”

Hashing is a data-indexing method invented for internet search in the 1990s. It uses numerical methods to encode large amounts of information, like entire webpages or chapters of a book, as a string of digits called a hash. Hash tables are lists of hashes that can be searched very quickly.

“It would have made no sense to implement our algorithm on TensorFlow or PyTorch because the first thing they want to do is convert whatever you’re doing into a matrix multiplication problem,” Chen said. “That is precisely what we wanted to get away from. So we wrote our own C++ code from scratch.”

Shrivastava said SLIDE’s biggest advantage over back-propagation is that it is data parallel.

“By data parallel I mean that if I have two data instances that I want to train on, let’s say one is an image of a cat and the other of a bus, they will likely activate different neurons, and SLIDE can update, or train on these two independently,” he said. “This is much a better utilization of parallelism for CPUs.

“The flipside, compared to GPU, is that we require a big memory,” he said. “There is a cache hierarchy in main memory, and if you’re not careful with it you can run into a problem called cache thrashing, where you get a lot of cache misses.” Shrivastava said his group’s first experiments with SLIDE produced significant cache thrashing, but their training times were still comparable to or faster than GPU training times. So he, Chen and Medini published the initial results on arXiv in March 2019 and uploaded their code to GitHub. A few weeks later, they were contacted by Intel.

“Our collaborators from Intel recognized the caching problem,” he said. “They told us they could work with us to make it train even faster, and they were right. Our results improved by about 50% with their help.” Shrivastava said SLIDE hasn’t yet come close to reaching its potential.

“We’ve just scratched the surface,” he said. “There’s a lot we can still do to optimize. We have not used vectorization, for example, or built-in accelerators in the CPU, like Intel Deep Learning Boost. There are a lot of other tricks we could still use to make this even faster.”

He said SLIDE is important because it shows there are other ways to implement deep learning.

“The whole message is, ‘Let’s not be bottlenecked by multiplication matrix and GPU memory,'” Shrivastava said. “Ours may be the first algorithmic approach to beat GPU, but I hope it’s not the last. The field needs new ideas, and that is a big part of what MLSys is about.”

Additional co-authors include James Farwell, Sameh Gobriel and Charlie Tai, all of Intel Labs.

The research was supported by the National Science Foundation (NSF-1652131, NSF-BIGDATA 1838177), the Air Force Office of Scientific Research (FA9550-18-1-0152), Amazon and the Office of Naval Research.

MLSys paper
Rice University News on Twitter

About Rice Univerisity

Located on a 300-acre forested campus in Houston, Rice University is consistently ranked among the nation’s top 20 universities by U.S. News & World Report. Rice has highly respected schools of Architecture, Business, Continuing Studies, Engineering, Humanities, Music, Natural Sciences and Social Sciences and is home to the Baker Institute for Public Policy. With 3,962 undergraduates and 3,027 graduate students, Rice’s undergraduate student-to-faculty ratio is just under 6-to-1. Its residential college system builds close-knit communities and lifelong friendships, just one reason why Rice is ranked No. 1 for lots of race/class interaction and No. 4 for quality of life by the Princeton Review. Rice is also rated as the best value among private universities by Kiplinger’s Personal Finance.