TensorFlow Lite has emerged as a popular platform for running machine learning models on the edge. A microcontroller is a tiny low-cost device to perform the specific tasks of embedded systems.

In a workshop held as part of Google I/O, TensorFlow founding member Pete Warden delved deep into the potential use cases of TensorFlow Lite for microcontrollers.

Further, quoting the definition of TinyML from a blog, he said:

Tiny machine learning is capable of performing on-device sensor data analytics at extremely low power, typically in the mW range and below, and hence enabling a variety of ways-on-use-case and targeting battery operated devices.

A Venn diagram of TinyML showcasing the composition of TinyML (Source: Google I/O)

Most machine learning applications are resource-intensive, and expensive to deploy and maintain.

According to PH Data, $65K (INR 47 lakhs) is the bare minimum amount required to deploy and maintain a model over 5 years. As you build a scalable framework to support future modeling activities, the cost might escalate to $95K (INR 70 lakhs) over five years.

On the other hand, TinyML is quite flexible and simple and requires less power.

Each hardware component (mW and below) acts independently, and the storage capacity of most machine learning models barely exceeds 30kb. Also, data can be processed on the device locally, which inevitably reduces data latency and solves the data privacy issue. Arm, Arduino, Sparkfun, adafruit, Raspberry Pi, etc are the major players in TinyML.

TensorFlow Lite, an open-source library by Google, helps in designing and running tiny machine learning (TinyML) models across a wide range of low-power hardware devices, and does not require much coding or machine learning expertise, said Warden.

Benefits of TinyML:

The TinyML process works in four simple steps gather/collect data, design and train the model, quantise the model and deploy to the microcontroller.

In a blog post, TensorFlow Lite for Microcontrollers, Google has explained some of its latest projects that combine Arduino and TensorFlow to create useful tools:

To initiate the project, you need a TF4 Micro Motion Kit pre-installed on Arduino. Once you have installed the packages and libraries on your laptop or personalised computer, look for the red, green and blue flashing LED in the middle of the board. The details of the setup are found here.

Once the setup is complete, you need to connect the device via Bluetooth; the TF4Micro motion kit communicates with this website via BLE, giving you a wireless experience. Now, tap the button on your Arduino, then wait for the red, green, and blue LED pattern to return. After this, click the connect button as shown on the website, then select TF4Micro Motion Kit from the dialogue box. You are now good to go. Similar steps need to be followed for all three experiments Air snare, FUI and tiny motion trainer.

Note: Do not hold the button down as this will clear the board.

The above experiments will help you get a hang of TensorFlow Lite on microcontrollers. You can also submit your ideas to the TensorFlow Microcontroller Challenge and win exciting cash prizes.

As part of a TensorFlow Microcontroller Challenge, Sparkfun is giving out a free TensorFlow Lite for Microcontrollers Kit. Click here to get yours.

Amit Raja Naik is a senior writer at Analytics India Magazine, where he dives deep into the latest technology innovations. He is also a professional bass player.

Here is the original post:
How TensorFlow Lite Fits In The TinyML Ecosystem - Analytics India Magazine

Media and publishing used to be the domain of specialized companies who controlled the content. The internet broke that model, and today anyone can go online and publish a blog, a podcast, or star in their own video.

But the big tech companies want to take control, closing content into walled gardens. But thats not what the majority of publishers, big or small, want.

We get to hear the perspectives of the publishers at every scale, and they consistently tell us the same thing: They want to more directly connect to consumers; they dont want to be tied into these walled gardens which dictate how they must present their content and in some cases what content theyre allowed to present, said Dr. Peter Day (pictured, right), chief technology officer at Quantcast Corp.

Day and Shruti Koparkar (pictured, left), head of product marketing at Quantcast, spoke with John Furrier, host of theCUBE, SiliconANGLE Medias livestreaming studio, duringThe Cookie Conundrum: A Recipe for Success event. They discussed the importance of smart technology for the post-cookie future of digital marketing. (* Disclosure below.)

Quantcast has cast itself as a champion of the open internet as it sets out to find the middle ground between the ability to scale provided by walled gardens and access to individual-level user data. Urgency for the quest is provided by Goliath company Google, which announced it will no longer be supporting third-party cookies on its Chrome browser as of January 2022.

Our approach to a world without third-party cookies is grounded in three fundamental things, Koparkar stated. First is industry standards: We think its really important to participate and to work with organizations who are defining the standards that will guide the future of advertising, Koparkar said, naming IAB Technology Laboratorys Project Rearc and Prebid as open projects Quantcast is involved with.

The companys engineering team also participates in meetings with the World Wide Web Consortium (W3C) to keep on top of what is happening with web browsers and to monitor what Google is up to with its Federated Learning of Cohorts (FLoC) project.

The second fundamental principle to Quantcasts strategy is interoperability. With multiple identity solutions from Unified ID 2.0 to FLoC already existing, and more on the way, We think it is important to build a platform that can ingest all of these signals, and so thats what weve done, Koparkar said referring to the release of Quantcasts intelligent audience platform.

Innovation is the third principle. Being able to take in multiple signals, not only IDs and cohorts, but also contextual first-party consent, time, language, geolocation and many others is increasingly important, according to Kopackar.

All of these signals can help us understand user behavior, intent and interests in absence of third-party cookies, she said.

But these signals are raw, messy, complex and ever-changing. What you need is technology like AI and machine learning to bring all of these signals together, combine them statistically, and get an understanding of user behavior, intent and interest, and then act on it, Koparkar stated. And the only way to bring them all together to obtain coherent understanding is through intelligent technologies such as machine learning, she added.

The foundation of our platform has always been machine learning from before it was cool, Day said. Many of the core team members at Quantcast have doctorate degrees in statistics and ML, which means it drives the companys decision-making.

Data is only useful if you can make sense of it, if you can organize it, and if you can take action on it, Day said. And to do that at this kind of scale its absolutely necessary to use machine learning technology.

Watch the complete video interview below, and be sure to check out more of SiliconANGLEs and theCUBEs coverage of The Cookie Conundrum: A Recipe for Success event. (* Disclosure: TheCUBE is a paid media partner for The Cookie Conundrum: A Recipe for Success event. Neither Quantcast Corp., the sponsor for theCUBEs event coverage, nor other sponsors have editorial control over content on theCUBE or SiliconANGLE.)

Show your support for our mission with our one-click subscription to our YouTube channel (below). The more subscribers we have, the more YouTube will suggest relevant enterprise and emerging technology content to you. Thanks!

Support our mission: >>>>>> SUBSCRIBE NOW >>>>>> to our YouTube channel.

Wed also like to tell you about our mission and how you can help us fulfill it. SiliconANGLE Media Inc.s business model is based on the intrinsic value of the content, not advertising. Unlike many online publications, we dont have a paywall or run banner advertising, because we want to keep our journalism open, without influence or the need to chase traffic.The journalism, reporting and commentary onSiliconANGLE along with live, unscripted video from our Silicon Valley studio and globe-trotting video teams attheCUBE take a lot of hard work, time and money. Keeping the quality high requires the support of sponsors who are aligned with our vision of ad-free journalism content.

If you like the reporting, video interviews and other ad-free content here,please take a moment to check out a sample of the video content supported by our sponsors,tweet your support, and keep coming back toSiliconANGLE.

See the article here:
Quantcast uses machine learning and AI to take on walled garden giants in the fight for the open internet - SiliconANGLE News

For the past several years, tech giants have been trying to make artificial intelligence in its many guises HPC, data analytics, and other advanced workloads more available and easier to use for enterprises.

Traditional OEMs such as Hewlett Packard Enterprise, Dell Technologies, and Lenovo are using a combination of hardware, software, and services to make technologies that in years gone by would only be employed by research institutions and the largest of corporations more widely accessible.

The public clouds, particularly those that have their own hyperscale applications driven by machine learning and scale, also function a bit like OEMs when it comes to these workloads.

They have exposed the tools they build for their own use through their clouds, giving customers another option to be a modern computing organization. In many cases, this reduces time to market for AI-driven applications and can also reduce costs particularly huge capital outlays for buying GPU-laden infrastructure but also for high salaried AI experts who are in short supply and high demand.

The global AI space is expected to grow from $27.23 billion in 2019 to almost $267 billion by 2027, according to a report from Fortune Business Insights. While on-premises deployments will grab revenue share, the cloud deployment segment [will] gain traction owing to less implementation expenses, the report states. Also, the cloud offers tools and pre-trained networks, which makes building AI applications convenient.

Amazon Web Services the largest of the hyperscale cloud providers offers a range of services, from Fraud Detector and Forecast (for predicting demand) to Kendra (enterprise search) and CodeGuru (automating code reviews). Microsoft Azure offers an AI platform that includes services reaching from machine learning to knowledge search to various apps and agents.

IBM Cloud hosts a range of capabilities based on the companys Watson AI technology and Oracle Cloud includes an array of AI services and optimized infrastructure.

For about a decade, Google has focused on AI and machine learning, seeing such technologies as keys for advancing the capabilities throughout its ever-expanding array of services. That has been on display this week during the companys virtual Google I/O 2021 developer conference. In his keynote address, Sundar Pichar, CEO of both Google and its parent company, Alphabet, spoke about how Google continues to infuse AI and machine learning into everything from search to security to Android-based devices.

Even a new facility aimed at accelerating Googles quantum computing capabilities includes AI in its name: the Quantum AI campus in Santa Barbara, California, which will be down the road from the University of California campus where Urs Hotzle, senior vice president for technical infrastructure at Google, was a professor of computer science before joining the search engine giant as one of its earliest employees.

At the same time, Google Cloud took steps to make it easier for data scientists and developers to pull together AI-based applications and for enterprises to get those applications deployed. Vertex AI is a platform that encompasses a range of existing machine learning services with a unified user interface and API. Developers using Vertex AI can train an AI model using almost 80 percent fewer lines of code than platforms from other cloud providers, which opens up the development of such models and the management of machine learning projects to a wider range of data scientists and machine learning engineers with varying levels of skill, according to Google.

Today, data scientists grapple with the challenge of manually piecing together ML point solutions, creating a lag time in model development and experimentation, resulting in very few models making it into production, Craig Wiley, director of product for Vertex AI and AI applications at Google Cloud, wrote in a blog post. To tackle these challenges, Vertex AI brings together the Google Cloud services for building ML under one unified UI and API, to simplify the process of building, training, and deploying machine learning models at scale. In this single environment, customers can move models from experimentation to production faster, more efficiently discover patterns and anomalies, make better predictions and decisions, and generally be more agile in the face of shifting market dynamics.

Andrew Moore, vice president and general manager of cloud AI and industry solutions at Google Cloud, said the goals of Vertex AI were to remove orchestration burdens from data scientists and engineers and create an industry-wide shift that would make everyone get serious about moving AI out of pilot purgatory and into full-scale production.

Organizations using Vertex AI will get access to the same AI toolkit which includes such capabilities as computer vision, language and conversation as well as structured data Google engineers use internally for the companys own operations, as well as new MLOps features like Vertex Vizier to speed up experimentation, Vertex Feature Store (a fully managed feature where data scientists and developers can offer, share and reuse machine learning features) and Vertex Experiments to use faster model selection to accelerate the deployment of machine learning models into deployment.

An experimental application called Vertex ML Edge Manager will enable organizations to deploy and monitor models on the edge via automated processes and APIs, enabling the data to stay on the device or on site. Other tools, such as Vertex Model Monitoring, Vertex ML Metadata and Vertex Pipelines are designed to streamline machine learning workflows.

AutoML enables developers and engineers with little machine learning experience to train models targeted at specific business needs and includes a central managed registry for all datasets across vision, natural language and tabular data types, while enterprises use BigQuery ML to export datasets from Googles managed BigQuery cloud data warehouse into Vertex AI. Vertex Data Labeling enables accurate labels for data collection.

Vertex AI also integrates with such open-source frameworks as TensorFlow, PyTorch and scikit-learn.

Google is promising more innovations around Vertex AI, which will be important for the company as it tries to gain ground on AWS and Azure, who together accounted for more than half of global cloud revenues in the first quarter in a market that saw spending reach more than $39 billion, a 37 percent year-over-year increase, according to Synergy Research Group.

Google Cloud is the third on the list and was among several other companies those being Alibaba, Tencent and Baidu as cloud providers that saw growth rates that surpassed the overall growth in the market.

Read the original here:
Google Cloud Mainstreams AI With Vertex Platform - The Next Platform

Data driven analytics and machine learning predictions are key to business success, but they are being commoditized. How well they work is all about the data. This is the position startupExploriumtook a bet on in 2019, and it seems to be paying off.

After a $19M Round A in 2019 and a $31M Series B in 2020, today Explorium announced it has closed a $75M Series C funding round, led by global venture capital and private equity firm Insight Partners, with existing investors Zeev Ventures, Emerge, F2 Venture Capital, 01 Advisors and Dynamic Loop Capital also participating.

Wecovered Explorium when it first set out to conquer the world in 2019. We caught up again with co-founder and CEO Maor Shlomo to discuss progress made, and the state of external data acquisition in 2021.

Explorium is growing alongside the demand for external data in organizations. As business conditions evolved and regulations restricted access to crucial sources of information, teams went hunting outside their organizations for data to support machine learning and other mission-critical analytics.

Arecent surveyfound enterprises are hungry for external data, but they don't necessarily have a clear idea how to get it. Here are some of the key findings.

Organizations value data acquisition, but that doesn't mean they have a clear strategy. Survey respondents overwhelmingly indicated that the acquisition and onboarding of external data was important to their business.

Interestingly, less than a third of respondents actually have a strategy in place, with 26% relying on ad-hoc practices or an informal process for data acquisition. Seven percent of respondents find data acquisition so challenging, they don't do it at all.

Most organizations are increasing their data acquisition investment for 2021, and allocating significant budgets for this. In 2020, 81% of companies spent more than $100k each month on external data acquisition, and 31% spent more than $500k.

Demand for external data acquisition is growing in organizations, and it's a complicated issue. Image: Explorium

Almost half of respondents said they spend over 50 hours per month on external data acquisition. In 2021, 78% of respondents are planning to increase their budgets for external data acquisition, and only 1% do not have a budget in place at all.

Multiple vendor relationships are slowing down external data acquisition. Nearly all respondents engage with at least two data providers in their external data acquisition strategy.

Sixty-nine percent of companies engage with three or more vendors, and 7% have five or more vendor relationships. Expectedly, the more data vendors organizations engage with, the more time and money they spend on data acquisition.

Most organizations need to discover and onboard external data at scale. When asked about the types of external data the organization was purchasing, over 50% of respondents said they are purchasing three or four types of external data.

This means the efforts and costs of purchasing a single data source must be multiplied in order to fulfill their business needs. As the number of use cases, data sources, and providers evaluated for each data source grows, this increases efforts exponentially, and so organizations must find a way to make the process more scalable.

Concluding, the survey found that 77% of respondents simply don't know what to look for in data acquisition, despite considering it valuable. Nearly all respondents said that finding relevant external data, as well as deriving insights from data, is medium to high effort for their organization.

Granted, that survey was sponsored by Explorium, and it points towards a single end-to-end platform, which Explorium would very much like to be. Regardless, the need for data external to organizations seems to have legs. As Shlomo emphasized, COVID-19 has exacerbated this by showing how diversified datasets can lead to better predictions.

Explorium touts itself as the external data platform that automatically discovers thousands of relevant data signals and uses them to improve analytics and machine learning. Explorium's platform works in three stages: Data enrichment, feature engineering, and predictive modeling.

The platform acts as a data marketplace and a predictive model hub at the same time. It aggregates and provides access to an array of data sources external to organizations. When the most relevant datasets have been chosen, Explorium uses them to generate candidate features for machine learning models.

Explorium then evaluates hundreds of models on different subsets of features and sources to introduce automated feedback on the data sources and features. Eventually, Explorium converges into the best subset of features given a specific model.

Explorium's platform works in a 3-step process: Data enrichment - Feature engineering - Predictive modeling. Image: Explorium

The basic recipe has not changed, Shlomo said, because it works well. Explorium is seeing exponential growth, he went on to add. Both its team and is customer base are growing. In the last year, Explorium has doubled its customer base and more than quadrupled revenue.

Companies like BlueVine, GlassesUSA.com, Melio and PepsiCo use Explorium to enhance AI models for use cases including lead scoring, identifying default risk and fraud and upleveling analytics such as demand forecasting and customer lifetime value.

On the technical side, Explorium has progressed by expanding the breadth and depth of its offering. Recently it introducedSignal Studio, which Shlomo described as a streamlined user interface to help users who are not necessarily data scientists build their own custom data feeds.

Explorium has also builtintegrations with a number of 3rd party data vendors, such as AWS, Azure, Google Cloud, SAP and Snowflake. Shlomo referred to those as deep integrations, mentioning for example how Explorium enables Snowflake users to integrate external data without having to code.

It's all about removing friction, as per Shlomo. Explorium did not invent data marketplaces, or AutoML. But it came up with a way to bring them together and make them easy to access, and that's the road to success.

Explorium will use the funding to grow its ecosystem and offering further, and there will be more announcements soon, Shlomo concluded. The round brings Explorium's total investment to more than $127M, and it entails George Mathew, Managing Director at Insight Partners and former President & COO of Alteryx, joining Explorium's board of directors.

See original here:
It's all about the data: Explorium's bet pays off in $75M Series C funding - ZDNet