Netflix launches a virtual HBCU boot camp with Norfolk State to increase exposure to the tech industry

Netflix is going back to school.

Working with Norfolk State University, the alma mater of one of the company’s senior software engineers, and the online education platform, 2U, Netflix is developing a virtual boot camp for students to gain exposure to the tech industry.

Starting today Netflix will open enrollment for 130 students to participate in a 16-week training program beginning in January.

That program will be divided into three tracks — Java Engineering, UX/UI Design and Data Science. Experts from Netflix will work with 2U to design each track and all courses will be led by faculty from Norfolk State University and feature guest lecturers from the tech industry, the company said.

Members from the company’s data science, engineering and design teams will serve as mentors — including Norfolk State alumnus Michael Chase.

Netflix will foot the bill for students accepted into the program, and they’ll get course credit for completing the boot camp, the company said.

“The goal is for participants to come away better equipped with industry-relevant skills to enter today’s workforce and with valuable, long-lasting relationships,” Kabi Gishuru, the company’s director of Inclusion Recruiting Programs wrote in a statement. “As we continue to invest in building the best service for our members, we want to invest in the best team to support it. Creating space in the industry for all voices will only make it stronger.”

Kite adds support for 11 new languages to its AI code completion tool

When Kite, the well-funded AI-driven code completion tool, launched in 2019, its technology looked very impressive, but it only supported Python at the time. Earlier this year, it added JavaScript, and today it is launching support for 11 new languages.

The new languages are Java, Kotlin, Scala, C/C++, Objective C, C#, Go, TypeScript, HTML/CSS and Less. Kite works in most popular development environments, including the likes of VS Code, JupyterLab, Vim, Sublime and Atom, as well as all JetBrains IntelliJ-based IDEs, including Android Studio.

This will make Kite a far more attractive solution for a lot of developers. Currently, the company says, it saves its most active developers from writing about 175 “words” of code every day. One thing that always made Kite stand out is that it ranks its suggestions by relevance — not alphabetically as some of its non-AI driven competitors do. To build its models, Kite fed its algorithms code from GitHub .

The service is available as a free download and as a server-powered paid enterprise version with a larger deep learning model that consequently offers more AI smarts, as well as the ability to create custom models. The paid version also includes support for multi-line code completion, while the free version only supports line-of-code completions.

Kite notes that in addition to adding new languages, Kite also spent the last year focusing on the user experience, which should now be less distracting and, of course, offer more relevant completions.

Image Credits: Kite

Standing by developers through Google v. Oracle

The Supreme Court will hear arguments tomorrow in Google v. Oracle. This case raises a fundamental question for software developers and the open-source community: Whether copyright may prevent developers from using software’s functional interfaces — known as APIs — to advance innovation in software. The court should say no — free and open APIs protect innovation, competition and job mobility for software developers in America.

When we use an interface, we don’t need to understand (or care) about how the function on the other side of the interface is performed. It just works. When you sit down at your computer, the QWERTY keyboard allows you to rapidly put words on the screen. When you submit an online payment to a vendor, you are certain the funds will appear in the vendor’s account. It just works.

In the software world, interfaces between software programs are called “application programming interfaces” or APIs. APIs date back to the 1950s and allow developers to write programs that reuse other program functionality without knowing how that functionality is performed. If your program needs to sort a list, you could have it use a sorting program’s API to sort the list for your program. It just works.

Developers have historically used software interfaces free of copyright concerns, and this freedom has accelerated innovation, software interoperation and developer job mobility. Developers using existing APIs save time and effort, allowing those savings to be refocused on new ideas. Developers can also reimplement APIs from one software platform to others, enabling innovation to flow freely across software platforms.

Importantly, reusing APIs gives developers job portability, since knowledge of one set of APIs is more applicable cross-industry. The upcoming Google v. Oracle decision could change this, harming developers, open-source software and the entire software industry.

Google v. Oracle and the platform API bargain

Google v. Oracle is the culmination of a decade-long dispute. Back in 2010, Oracle sued Google, arguing that Google’s Android operating system infringed Oracle’s rights in Java. After ten years, the dispute now boils down to whether Google’s reuse of Java APIs in Android was copyright infringement.

Prior to this case, most everyone assumed that copyright did not cover the use of functional software like APIs. Under that assumption, competing platforms’ API reimplementation allowed developers to build new yet familiar things according to the API bargain: Everyone could use the API to build applications and platforms that interoperate with each other. Adhering to the API made things “just work.”

But if the Google v. Oracle decision indicates that API reimplementation requires copyright permission, the bargain falls apart. Nothing “just works” unless platform makers say so; they now dictate rules for interoperability — charging developers huge prices for the platform or stopping rival, compatible platforms from being built.

Free and open APIs are essential for modern developers

If APIs are not free and open, platform creators can stop competing platforms from using compatible APIs. This lack of competition blocks platform innovation and harms developers who cannot as easily transfer their skills from project to project, job to job.

MySQL, Oracle’s popular database, reimplemented mSQL’s APIs so third-party applications for mSQL could be “ported easily” to MySQL. If copyright had restricted reimplementation of those APIs, adoption of MySQL, reusability of old mSQL programs and the expansion achieved by the “LAMP” stack would have been stifled, and the whole ecosystem would be poorer for it. This and other examples of API reimplementation — IBM’s BIOS, Windows and WINE, UNIX and Linux, Windows and WSL, .NET and Mono, have driven perhaps the most amazing innovation in human history, with open-source software becoming critical digital infrastructure for the world.

Similarly, a copyright block on API-compatible implementations puts developers at the mercy of platform makers say so — both for their skills and their programs. Once a program is written for a given set of APIs, that program is locked-in to the platform unless those APIs can also be used on other software platforms. And once a developer learns skills for how to use a given API, it’s much easier to reuse than retrain on APIs for another platform. If the platform creator decides to charge outrageous fees, or end platform support, the developer is stuck. For nondevelopers, imagine this: The QWERTY layout is copyrighted and the copyright owner decided to charge $1,000 dollars per keyboard. You would have a choice: Retrain your hands or pay up.

All software used by anyone was created by developers. We should give developers the right to freely reimplement APIs, as developer ability to shift applications and skills between software ecosystems benefits everyone — we all get better software to accomplish more.

I hope that the Supreme Court’s decision will pay heed to what developer experience has shown: Free and open APIs promote freedom, competition, innovation and collaboration in tech.

A founder’s guide to recession planning for startups

We are living through one of the nation’s longest periods of economic growth. Unfortunately, the good times can’t last forever. A recession is likely on the horizon, even if we can’t pinpoint exactly when. Founders can’t afford to wait until the midst of a downturn to figure out their game plans; that would be like initiating swim lessons only after getting dumped in the open ocean.

When recession inevitably strikes, it will be many founders’ — and even many VCs’ — first experiences navigating a downturn. Every startup executive needs a recession playbook. The time to start building it is now.

While recessions make running any business tough, they don’t necessitate doom. I co-founded two separate startups just before downturns struck, yet I successfully navigated one through the 2000 dot-com bust and the second through the 2008 financial crisis. Both companies not only survived but thrived. One went public and the second was acquired by Mastercard.

I hope my lessons learned prove helpful to building your own recession game plan.

Recession is an opportunity to leapfrog the competition

In entrepreneurship, the goal isn’t just to survive; it’s to win. Some founders think that surviving recession amounts to hoarding cash and sitting out the financial winter. While there’s wisdom in hoarding cash (see below), I strongly recommend against sitting idly when that time could be actively leveraged to strengthen competitive advantage.

I founded my first startup, Yodlee, in a strong economy with almost 20 competitors. Ten years and a painful recession later, we were the only game in town. Critical to our success was acquiring our largest competitor, something we never could have done in a strong economy because they never would have been willing to sell. The recession made it untenable for them to fundraise, enabling us not only to buy them, but to do so without cash in an all-equity deal. I recommend thinking ahead of time about which companies you would want to buy if the opportunity arose, and your goals for doing so, such as consolidating competition, acquiring customers or engineering talent, entering new markets or strengthening product offerings or distribution channels.

Recession is also an opportunity to improve

You can’t rebuild a plane when you’re traveling 500 miles per hour. During a strong economy, companies spend most of their energy on sales and growth. During a weaker economy, it’s easier to justify the investment in infrastructure and technical debt. Yodlee was built on PERL, which we knew would eventually need upgrading. Once the downturn hit, we took advantage of the slower sales cycles to totally retool in Java, an enterprise-class programming language capable of scale. And we didn’t stop there — we created six new products during the downturn.

Make yourself indispensable to customers and partners

The precipice of a recession is not the time to over-index on top-line revenue. You never want to be on your customers’ top five lists of easiest-to-cut products and services. Instead, take the time to understand your customers’ needs, embed yourself deeply in their operations or customer experience and invest significantly in top-notch customer success.

At my second startup, Truaxis, once recession struck, we pivoted from credit card customer acquisition for banks (which requires no help during a recession) to helping banks address churn. Our revised offering yielded a tremendous ROI for banks — a 10X increase in profit. Our product also became the cornerstone to their online consumer banking experience. If you figure out how to make your product indispensable or core to the customer experience, it won’t get cut, even during a recession.

Lock-in long-term customer contracts

Both of my companies started out with B2C business models. After each recession hit, I quickly pivoted to B2B2C. Here’s why: While consumers can react immediately to economic jitters, businesses must keep spending in order to keep operating. Plus, they work on annual budget cycles. Even when businesses want to reduce their costs, they typically can’t react very quickly because they have to wait out their contracts.

In a bull economy, short-term contracts are popular because they enable companies to keep raising prices. Don’t be tempted by short-term cash. B2B and B2B2C firms should take the potential revenue hit by locking in long-term contracts now while budgets and buyers are flush.

Consider diversifying revenue streams and customer segments

While the economy is still healthy, explore options for diversifying your revenue streams and customer bases to more recession-resistant segments. If your business is consumer-focused, consider a different distribution model via businesses or new consumer segments like affluent populations, which are less sensitive to economic fluctuations. If you have an enterprise-focused business, transition more of your revenue to larger enterprises, which are more financially resilient than smaller ones, or to enterprises that need your service for survival, especially in a down market.

Key to the diversification strategy is plotting your axis ahead of time. You don’t want to start your exploration when the market has already turned and you’re burning cash faster than you can get it. Upon exploration, you may find that no pivot is necessary — perhaps only the need to slow down. Now is the time to look for and deeply understand the signals in your business, though you may not need to act on them for a while — or perhaps even ever.

Raise a lot of money — and stash away more than you think you’ll need

It’s obviously a lot easier to raise money in a healthy economy than a weak one. If your coffers aren’t full going into a downturn, it doesn’t matter what you do; you’ve lost the game right there. Having enough cash can make the difference between emerging as the market leader (i.e. the only one still with cash in the bank) and going out of business — even if your company would have thrived in a strong economy. Be conservative when projecting how much money you’ll need to stay afloat. Many leaders underestimate how much elongated sales cycles, diminished average deal sizes and dwindling total sales transactions weaken total revenue.

Be thoughtful about valuations for your employees’ sake

I’m supportive of founders seeking aggressive valuations, but it’s important to realize the potential downside. Valuations soften during recessions, which can lead to corrections or recapitalizations. Recapitalizations create new companies in which the old stock is worth nearly nothing, leaving many employees’ options under water.

I learned this the hard way at Yodlee after raising a lot of money at a high valuation in 1999. We banked enough money that we could have lasted through most downturns without fundraising. Alas, while the average recession lasts 11 months, the dot-com crash lasted several years. Even though we were strong enough to fundraise during the recession, our high valuation forced us to recapitalize. This was crushing for the employees whose equity was suddenly worthless.

In a weak economy, startups struggle to retain their strongest employees who often retreat for safer work environments and more predictable incomes. Recapitalizations deliver an unwanted shove out the door to demoralized employees who feel they have no reason to stay. Inevitably after recapitalizations the people who are strong enough to get hired elsewhere do so. Surviving a downturn is challenging enough. Doing so without a strong, motivated team is nearly impossible.

While times are strong, choose the board you’ll want when things go bad

When my Yodlee board members suggested we pivot from B2C to B2B2C, I thought they were crazy. We had acquired 1 million users through word of mouth in only two-three months. I couldn’t believe they advocated such a significant pivot when things were going so well. I eventually came to understand that these seasoned board members were actually saving my business.

As my colleague Karan Mehandru said, “investors are your war partners, not your beer buddies.” When fundraising, think carefully about who you want around the table if the economy goes south. I recommend asking potential investors if they’ve weathered downturns before and how they’d help you navigate one. I’d ask the same questions of the firm’s other partners to look for consistency of answers and to gauge your investors’ standing and seniority within the partnership. All too many board members are lovely when companies grow rapidly, but challenging when speed bumps arise. Will your board members actively help you address these challenges or stand in passive judgment?

Being a founder is hard enough, but leading a startup through a recession catapults an already challenging job to a whole different level. Whether the recession begins tomorrow or in four years, I hope you’ll learn from my experience and be prepared either way.

LaunchDarkly CEO Edith Harbaugh explains why her company raised another $54M

This week, LaunchDarkly announced that it has raised another $54 million. Led by Bessemer Venture Partners and backed by the company’s existing investors, it brings the company’s total funding up to $130 million.

For the unfamiliar, LaunchDarkly builds a platform that allows companies to easily roll out new features to only certain customers, providing a dashboard for things like “canary launches” (pushing new stuff to a small group of users to make sure nothing breaks) or launching a feature only in select countries or territories. By productizing an increasingly popular development concept (“feature flagging”) and making it easier to toggle new stuff across different platforms and languages, the company is quickly finding customers in companies that would rather not spend time rolling their own solutions.

I spoke with CEO and co-founder Edith Harbaugh, who filled me in on where the idea for LaunchDarkly came from, how their product is being embraced by product managers and marketing teams and the company’s plans to expand with offices around the world. Here’s our chat, edited lightly for brevity and clarity.

MicroEJ is taking over IoT on Earth and beyond

The internet of things (IoT) market is expanding at a rate where distinguishing it as a separate category is beginning to seem a bit absurd. Increasingly, new products — and updates of existing ones — are smart and/or connected. One company is changing the fundamental calculus behind this shift by lowering the barrier considerably when it comes to what it costs to make something ‘smart,’ both in terms of the upfront bill of materials, along with subsequent support and development costs.

MicroEJ CEO Fred Rivard took me through his company’s history from its founding in 2004 until now. Much of those earlier years were spent in development, but since around 2012 or so, the French company has been deploying for IoT devices what Android is to smartphones — a flexible, extensible platform that can operate on a wide range of hardware profiles while being relatively easy to target for application and feature developers. MicroEJ takes the ‘code once, deploy anywhere’ maxim to the extreme, since its platform is designed from the ground up to be incredibly conservative when it comes to resource consumption, meaning it can run on hardware with as little as one-tenth or more the bill of materials cost of running more complex operating platforms — like Android Things, for instance.

“We take category of device where currently, Android is too big,” Rivard said. “So it doesn’t fit, even though you would like to have the capability to add software easily devices, but you can’t because Android is too big. The cost of entry is roughly $10 to $15 per unit in hardware and bill of material — that’s the cost of Android […] So it would be great to be able to run an Android layer, but you can’t just because of the cost. So we managed to reduce that cost, and to basically design a very small layer that’s1000 times smarter than Android.”

Universal Acceptance is the first-mover advantage that may be worth billions

Your expanding global business is leaving money on the table if its systems aren’t compatible with web addresses that have extensions such as .世界 or .ОНЛАЙН (.world and .online in Chinese and Russian, respectively). This missed opportunity has been growing for some time; a 2017 study concluded that an ecommerce market worth nearly $10 billion dollars annually is up for grabs — and that is a conservative estimate.

To understand why, consider these two facts:

First, the version of the Latin alphabet you are reading now is used by just more than a third of the world population. That number is dwarfed by the billions of people who read and write every day in Chinese, Arabic, Cyrillic, Devanagari or other scripts. These are being used in regions where population growth, economic growth and internet adoption all outpace global averages.

Second, recent innovations in how we navigate the internet have made domain names in diverse alphabets available to the majority of the world who use them. In 2012, there were only 22 so-called generic domain names (with familiar extensions like .com or .edu). That number now stands at more than 1,500.
Such innovation effectively brings an end to the era in which, say, a Japanese web surfer needs to toggle their keyboard to type a “www” or “.com,” because the entire domain name can now be written in Japanese.

This change is a big deal across rapidly growing markets worldwide, but particularly in Asia, where multilingualism is not widespread and new users on smartphones are key drivers of digital and economic growth. Even today, only a tiny percentage of all web addresses are expressed in Chinese characters, though Mandarin Chinese speakers make up nearly one-fifth of the world’s internet user population.

Even more relevant for the next wave of online consumers is that along with new domain names come email addresses in different scripts. A growing online population is using these addresses to sign up for services or sign into platforms.

This is why smart, global companies — and companies with global aspirations — are taking action to eliminate a major blind spot. Many software developers and corporate leaders reside in the English speaking or “Latin alphabet” parts of the world and the internet works pretty smoothly for them; therefore, they have not taken the important step of upgrading their software applications to accept all domain names equally. This step is a best practice referred to as “Universal Acceptance” of domain names and email addresses.

When systems are not Universal Acceptance ready, people using domain names or email addresses in different scripts cannot successfully use these systems, because the domain names and email addresses are not recognized as “valid.” This means lost business opportunities. Code libraries already exist in programming languages like Java and Python, often making this task the equivalent of a “bug fix;” however, it is a fix with huge implications.

To get a sense of the importance of Universal Acceptance, consider India. It has one of the fastest growing internet user populations on the planet and provides an illuminating case study.

As fast as internet adoption may be in India, in rural India it is faster still. The internet user base in India recently exceeded 500 million and is likely to reach 627 million by 2020. Two-fifths of users are located in rural areas. Consider also that India has 22 official languages and most users are on mobile devices.

In the Indian state of Rajasthan, the state government recently offered each of its 69 million residents free email addresses in both Hindi and English, while directing online public services to be Universal Acceptance ready (i.e. 100% available in Hindi). This required an intensive, 30-day push by developers to be compliant, and now residents can use their same Hindi email addresses to access an array of online platforms and services. Are some of these residents of Rajastan your future customers?

Microsoft is among the companies in the forefront of such compatibility. Last year it announced Email Address Internationalization (EAI) across most of its email platforms in an impressive 15 languages spoken across India. As Meetul Patel, COO of Microsoft India said:

“Ensuring that language is not a barrier to the adoption of technology is key for digital inclusion and growth. Making email addresses available in 15 languages is an exciting step to making modern communications and collaboration tools more accessible and easier to use for all – something we have been relentlessly working towards. We’re making technology use the language of people, and not requiring people to first learn the traditional language of technology.”

Despite such advantages, there is still a lot of work ahead. A recent review of the top 1,000 websites around the world found that only about five percent accepted all of the email address variations now in use.

Bringing systems up to date with Universal Acceptance is an easy way to make the internet more accessible for the billions of people whose languages are written in different scripts, making it a treasured cause of digital inclusion advocates. However, for any business seeking new global markets, it is a key competitive differentiator in an era of global online platforms, from direct e-commerce to the sharing economy. This is one first-mover advantage that may be worth billions.

HackerRank acquires Mimir, an online platform for computer science courses

HackerRank, a popular platform for practicing and hosting online coding interviews, today announced that it has acquired Mimir, a cloud-based service that provides tools for teaching computer science courses. Mimir, which is HackerRank’s first acquisition, is currently in use by a number of universities, including UCLA, Purdue, Oregon State and Michigan State, as well as by corporation like Google.

HackerRank says it will continue to support Mimir’s classroom product as a stand-alone product for the time being. By Q2 2020, the two companies expect to have an initial release of a combined product offering.

HackerRank will work closely with professors, students and customers to help student developers learn, improve and assess their skills from course work to career,” Vivek Ravisankar, the co-founder and CEO of HackerRank, told me. “Ultimately, we envision a combined product that allows students to obtain both a formal academic education as well as practical skills assessments which can help build a strong and successful career.”

The two companies did not disclose the financial details of the acquisition, but Indiana-based Mimir previously raised a total of $2.5 million and had eight employees at the time of the acquisition, including the three-person executive team.

As the companies stress, both focus on allowing developers for a variety of backgrounds to successfully vie for jobs, no matter where they went to school. HackerRank argues that the combination of its existing services and Mimir’s classroom tools will “provide computer science classrooms with the most comprehensive developer assessment platform on the market; allowing students to better prepare for real-world programming and universities to more accurately evaluate student progress.” The idea here clearly is to expand HackerRank’s reach into the world of academia and expand the talent pool for its customers who are looking to recruit from its users, but Ravisankar also noted that he hopes the combines strengths of HackerRank and Mimir will allow students to combine their academic learning with market learning. “This will ensure that they’re equipped with the skills that their future workplaces require,” he said.

Mimir isn’t so much a tool for massive online courses but instead focuses on helping teachers and students manage programming projects and assignments. To do so, it offers a full online IDE, as well as support for Jupyter notebooks, as well as more traditional teaching tools for creating quizzes and assignments. The built-in IDE supports 40 programming languages, including Python, Java and C. There’s also a tool for detecting plagiarism.

Currently, about 15,000 to 20,000 students are using Mimir’s platform for their course work. That’s dwarfed by the 7 million developers who have signed up for HackerRank so far, but not all of those are active, while, almost by default, all of Mimir’s users will be on the job market sooner or later.

“Mimir has made a name for itself by becoming a secret weapon for computer science programs – Mimir equips them with the tools to make a real difference in the education of developers,” said Prahasith Veluvolu, co-founder and CEO of Mimir. “Working with HackerRank is a natural evolution of our mission, allowing our customers to scale their programs while simultaneously giving students an unmatched classroom experience to prepare them for the careers of tomorrow.”

The AI stack that’s changing retail personalization

Consumer expectations are higher than ever as a new generation of shoppers look to shop for experiences rather than commodities. They expect instant and highly-tailored (pun intended?) customer service and recommendations across any retail channel.

To be forward-looking, brands and retailers are turning to startups in image recognition and machine learning to know, at a very deep level, what each consumer’s current context and personal preferences are and how they evolve. But while brands and retailers are sitting on enormous amounts of data, only a handful are actually leveraging it to its full potential.

To provide hyper-personalization in real time, a brand needs a deep understanding of its products and customer data. Imagine a case where a shopper is browsing the website for an edgy dress and the brand can recognize the shopper’s context and preference in other features like style, fit, occasion, color etc., then use this information implicitly while fetching similar dresses for the user.

Another situation is where the shopper searches for clothes inspired by their favorite fashion bloggers or Instagram influencers using images in place of text search. This would shorten product discovery time and help the brand build a hyper-personalized experience which the customer then rewards with loyalty.

With the sheer amount of products being sold online, shoppers primarily discover products through category or search-based navigation. However, inconsistencies in product metadata created by vendors or merchandisers lead to poor recall of products and broken search experiences. This is where image recognition and machine learning can deeply analyze enormous data sets and a vast assortment of visual features that exist in a product to automatically extract labels from the product images and improve the accuracy of search results. 

Why is image recognition better than ever before?

retail and artificial intelligence

 

While computer vision has been around for decades, it has recently become more powerful, thanks to the rise of deep neural networks. Traditional vision techniques laid the foundation for learning edges, corners, colors and objects from input images but it required human engineering of the features to be looked at in the images. Also, the traditional algorithms found it difficult to cope up with the changes in illumination, viewpoint, scale, image quality, etc.

Deep learning, on the other hand, takes in massive training data and more computation power and delivers the horsepower to extract features from unstructured data sets and learn without human intervention. Inspired by the biological structure of the human brain, deep learning uses neural networks to analyze patterns and find correlations in unstructured data such as images, audio, video and text. DNNs are at the heart of today’s AI resurgence as they allow more complex problems to be tackled and solved with higher accuracy and less cumbersome fine-tuning.

How much training data do you need?

The AI stack that’s changing retail personalization

Consumer expectations are higher than ever as a new generation of shoppers look to shop for experiences rather than commodities. They expect instant and highly-tailored (pun intended?) customer service and recommendations across any retail channel.

To be forward-looking, brands and retailers are turning to startups in image recognition and machine learning to know, at a very deep level, what each consumer’s current context and personal preferences are and how they evolve. But while brands and retailers are sitting on enormous amounts of data, only a handful are actually leveraging it to its full potential.

To provide hyper-personalization in real time, a brand needs a deep understanding of its products and customer data. Imagine a case where a shopper is browsing the website for an edgy dress and the brand can recognize the shopper’s context and preference in other features like style, fit, occasion, color etc., then use this information implicitly while fetching similar dresses for the user.

Another situation is where the shopper searches for clothes inspired by their favorite fashion bloggers or Instagram influencers using images in place of text search. This would shorten product discovery time and help the brand build a hyper-personalized experience which the customer then rewards with loyalty.

With the sheer amount of products being sold online, shoppers primarily discover products through category or search-based navigation. However, inconsistencies in product metadata created by vendors or merchandisers lead to poor recall of products and broken search experiences. This is where image recognition and machine learning can deeply analyze enormous data sets and a vast assortment of visual features that exist in a product to automatically extract labels from the product images and improve the accuracy of search results. 

Why is image recognition better than ever before?

retail and artificial intelligence

 

While computer vision has been around for decades, it has recently become more powerful, thanks to the rise of deep neural networks. Traditional vision techniques laid the foundation for learning edges, corners, colors and objects from input images but it required human engineering of the features to be looked at in the images. Also, the traditional algorithms found it difficult to cope up with the changes in illumination, viewpoint, scale, image quality, etc.

Deep learning, on the other hand, takes in massive training data and more computation power and delivers the horsepower to extract features from unstructured data sets and learn without human intervention. Inspired by the biological structure of the human brain, deep learning uses neural networks to analyze patterns and find correlations in unstructured data such as images, audio, video and text. DNNs are at the heart of today’s AI resurgence as they allow more complex problems to be tackled and solved with higher accuracy and less cumbersome fine-tuning.

How much training data do you need?