Raspberry Pi is all set to ramp up its machine learning abilities, co-founder Eton Upton said at tinyML Summit 2021. Upton said the in-house chip-development team has been working towards improving the hardware’s machine learning skills. 

The Raspberry Pi 400 is a computer integrated into a single compact keyboard that runs on Raspberry Pi 4, released in 2019. The Raspberry Pi 4 came up with a significant upgrade in its hardware, architecture, and operating system. 

Pi 4 can run basic machine learning algorithms using its built-in camera input for image recognition. It can accomplish basic tasks such as recognising objects, observing movement or running basic inference tasks. It allows for better load and run-time algorithms since code can be compiled quickly; thanks to its faster CPU and RAM. Machine learning tasks also perform twice as much better on Pi 4 as it is vastly more powerful than its previous iterations.

Going by Upton’s tinyML summit presentation, most of their work appears to be focused on developing lightweight accelerators for ultra-low power machine learning applications. 

For this, Raspberry Pi will use either of the three current generations of ‘Pi Silicon’ boards or low-cost, high-performance boards, two of which – SparkFun’s MicroMod RP2040 and Arduino’s Nano RP2040 Connect – are from board partners.

The last one works on an ArduCam Pico4ML from ArduCam that incorporates a camera, microphone, screen, and machine learning into the Pico package. It is a standalone microcontroller that does not need a CPU.

He also said the future chips might have lightweight accelerators, possibly 4-8 multiply-accumulates (MACs) per clock cycle compared to less than one MACs per clock cycle of RP2040.

The post Raspberry Pi To Get Machine Learning Boost appeared first on Artificial Intelligence.

Many retail stores remain closed due to COVID-19 restrictions, forcing countless outlets to move their operation online, many for the first time. With eCommerce retail becoming an increasingly competitive space, data science is playing a key role in giving retailers a competitive advantage, particularly for those businesses hoping to make a longer-term investment in their online presence.

Data-driven decision making uses facts, metrics and data to inform strategic business decisions that align with a company’s goals, objectives and initiatives. It subsequently enables companies to create new business opportunities, generate more revenue, predict future trends, optimise current operational efforts and produce actionable insights.

There are many ways in which data science can revolutionise eCommerce businesses and, in this article, we take a look at some of the most important.

Shopping Cart Abandonment

The shopping cart abandonment rate is an important metric for eCommerce sites to track because a high abandonment rate could signal a poor user experience or broken sales funnel. A sales funnel should run seamlessly from marketing to product selection to checkout, bringing potential customers to a purchase through a series of marketing actions such as automated emails, videos, articles and landing pages. Today, the average cart abandonment rate in online retail is 69.57%, which is $18bn lost every year.¹

There are many possible causes of cart abandonment, making it a complex problem to tackle. Beyond simply improving and optimising the shopping experience through A/B testing, a key strategy for dealing with cart abandonment is shopping cart recovery.

The following methods can be used to entice shoppers to recover items in their cart.

1. Abandoned cart emails or text messages – If the user entered their email address or phone number during the checkout process before leaving the website, then there is the opportunity to send them an abandonment message. This usually takes the form of an offer or discount code to entice the user to return to the site and complete the purchase.
2. Abandoned Cart Retargeting – Ad retargeting is another powerful tactic in cart recovery. With retargeting, retailers place an ad pixel on their checkout page and then can remarket to those users on platforms such as social media and Google. The advantage of retargeting is that it works even in the absence of personal information such an email address.

Sentiment Analysis

Sentiment analysis tools help eCommerce retailers derive valuable insights mined from unstructured customer comments on feedback forms and social media platforms about a given product or brand.  A customer experience strategy that does not integrate sentiment analysis as a core functionality will not capture the overall customer journey in a holistic manner.

Using sophisticated text mining techniques, eCommerce businesses can identify and resolve issues in products or services and enhance the overall user experience. Natural language processing techniques can identify words bearing a negative or positive attitude towards the brand and this feedback helps retailers to improve their products and services in direct response to consumer needs.

Loyalty Cards

Customer loyalty cards, while rewarding shoppers with discounts and deals, are an effective way for retailers to collect data on a large scale.

Customer loyalty cards extend beyond the obvious function of purchase tracking by establishing potential links between online and in store customer behaviour. This helps retailers to understand and shape purchase decisions by targeting advertising and organising products to encourage more sales.

Predictive Forecasting

Enabling personalised product recommendation is one way in which data science is transforming eCommerce businesses. Predictive forecasting uses different data sources to make predictions of a customer’s budget and preferences, including the history of previous sales, economic indicators, customer searches and demographic data. Predictive intelligence technology is used serve what online shoppers need even before they look for a product.

A predictive model can be trained using a historical dataset which classifies customers according to their possession of various characteristics, and the degree to which these characteristics tend to indicate certain product purchases. We would then customise our product suggestions to new customers based on the combination of price and product characteristics the model suggests will be most likely to lead to a purchase. As a further extension of this idea, we can also create metrics such as customer lifetime value (CLV), or incorporate a marketing mix model to understand how exactly we should target each customer.

Pricing Optimisation

Selling a product at the optimal price for each customer can be done with the help of machine learning algorithms. The algorithm analyses a number of parameters from the data at a highly granular level, such as flexibility of prices, location of the customer, the buying attitude of an individual customer and competitor pricing. The resulting price point is optimised to benefit all parties. This is a powerful and important tool for retailers to market their product using customer-specific and location-specific parameters.

Upselling and Cross-Selling

Ecommerce is a particularly rich environment for upselling and cross-selling. Retailers can make offers and recommendations that are truly personalised through insights gained from data science. In doing so, retailers not only increase revenue and profit, but also strengthen customer relationships.

With the help of customer data and product performance analytics, retailers can see what products a person is buying and track the different products they frequently purchase to learn how to optimise their marketing for each customer based on their previous purchases. For example, if a customer frequently buys apple juice and bottled water separately, it may be advantageous for the retailer to market these products together as a bundle, to increase the purchase frequency.

Inventory Management

In a supply chain, the warehousing function is critical to link the material flows between the supplier and customer. It is important for retailers to stock the right goods, in the right quantities and the right locations to meet customer demand for products. To achieve this, the stock and supply chain must be analysed thoroughly.

Powerful machine learning algorithms can analyse the data between supply and demand in great detail to detect patterns and correlations among purchases. This data is then analysed and informs a strategy to increase sales, confirm timely delivery and manage the inventory stock. This can be used to predict ahead of time whether periods of very low or no demand for a product are indicators of mistakes in the data, for example miss-stored or misclassified items, or genuine low demand.

Warehouse management software can also dictate how and where stock should be stored to optimise picking routes. Ultimately, by applying intelligence to big data, these systems can recommend stock movements within the warehouse so the flow of goods is constantly optimised.

Reducing Churn Rate

For subscription-based digital products, machine learning models can be used to predict whether a customer may churn. Such models are usually discriminative classifiers, using deep neural networks, tree-based methods or logistic regression. Generative models or recurrent neural networks can also be used. Both kinds of models can provide a probabilistic assessment of whether a customer is likely to take an action and is appropriate for targeting.

The digital world is in a constant state of flux, and to keep up with the competition and move with the ever-changing landscape, retailers must leverage data to make more informed and powerful data-driven business decisions. Data-driven decision-making can help retailers to improve and personalise user experience, predict purchases, optimise inventory management and, ultimately, drive profits. Data-driven insights can enable retailers to increase their agility, compete more effectively and gain a serious competitive advantage over other eCommerce businesses.

The post Give Your ECommerce Operation a Data Science Boost appeared first on Artificial Intelligence.

Ken Goldberg thinks artificial intelligence will enable surgical robots to achieve their best function — not replacing surgeons but augmenting their work by reducing the monotony of specific subtasks like suturing.

The William S. Floyd Jr. Distinguished Chair in Engineering at UC Berkeley, Goldberg and his research team have continued to advance the field.

The group — which includes postdoctoral researchers Minho Hwang and Jeffrey Ichnowski and PhD student Brijen Thananjeyan — has demonstrated how a deep neural network plus 3D-printed depth-sensing markers can train a da Vinci surgical robot to automatically perform peg transfer slightly faster and more accurately than an experienced surgical resident.

They also published a paper in Science Robotics last November about new AI software that allows robots to learn how to rapidly plan smooth motions that increase speed and reduce wear in factories, warehouses…and operating rooms.

Their work encompasses so much more, too. (Here’s a list of published research.)

Goldberg recently consulted with his research team to provide some insights to Medical Design & Outsourcing and MassDevice about where the surgical robotics space could advance in coming years:

MDO: You’ve spoken before of an emerging generation of surgical robots. Tell us more.

Goldberg: Many of the early patents on surgical-assist robots are expiring, so there are several new companies entering the market. For example, Johnson and Johnson purchased Auris, and Medtronic purchased Mazor in the past 2 years. Many new companies that are emerging in Asia — including China and Korea — are also developing a new generation of surgical-assist robots at lower cost and working to introduce some supervised autonomy.

Rather than replacing human surgeons, an emerging new generation of robots will assist surgeons by performing tedious subtasks such as suturing and debridement to improve consistency, reduce fatigue and open the door to long-distance tele-surgery.  Advances in AI can be applied to data collected from surgical systems such as Intuitive’s da Vinci to learn underlying control policies for subtasks including cutting, suturing, palpation, dissection, retraction and debridement.

MDO: How can artificial intelligence advance robot-assisted surgery? Give us an example.

Goldberg: We were recently able to automate peg transfer, a common training procedure for minimally invasive surgery, with 99.4% accuracy (357/360 trials) [on a da Vinci Research Kit] — even when the robot tools are switched. This task is challenging because it requires high accuracy, and the cables that drive surgical robots’ joints stretch during motion, which can significantly reduce their accuracy.[We’ve learned] to correct errors at key points in the task visually by using demonstrations from a human teleoperator.

We also present an alternate approach to this task in two papers, one published and one preprint. These approaches use fiducial markers and depth-sensing with deep learning to learn a model of how the robot moves as a function of its past motions, which is a complex function of its cabling properties. Using this learned model for control, we achieve 94-100% accuracy on bilateral and unilateral versions of the task, respectively. Using a trajectory optimization procedure, we are able to roughly match or outperform a human surgeon in terms of speed as well.

MDO:  TransEnterix executives see AI as an edge when it comes to competing better with its Senhance system. What do you think of their Intelligent Surgical Unit?

Goldberg: Learning how to position and move the surgical camera without manual adjustment is a very interesting problem.  It is tedious for the surgeon to do this herself, so it would be helpful if a robot could anticipate, for example, as a suture advances beyond the field of view and track the camera to move accordingly.  This is also subtle because erroneous camera motions could be frustrating for surgeons.

MDO: Overall, what are you most excited about in the robot-assisted surgery space?

Goldberg: I’m hopeful that we can develop systems that can learn to perform specific subtasks such as suturing or debridement faster and more accurately than human surgeons, thus relieving human surgeons of tedium, allowing them to focus on more nuanced aspects of surgery and also to reduce time in the operating room.

This research field is growing rapidly: There are now 30 labs worldwide doing experiments with a research version of the Intuitive surgical-assistant robot, and new hardware and algorithms are being published every month.

The post Could artificial intelligence boost surgical robotics to new heights? appeared first on Artificial Intelligence.

Anyone looking for evidence that insurance advertisements have become a mainstay of popular culture need look no further than Geico’s YouTube channel, which currently boasts 1.88 million subscribers and is home to videos with tens of millions of views. Geico’s competitors might not have as robust a presence on the platform, but their brand representatives (Jake from State Farm, Allstate’s Mayhem, Progressive’s Flo) are as familiar to the public as the Geico Gecko.

For these insurance companies, spending vast amounts of money on advertising on both traditional and digital channels is a must. However, having such universal brand recognition also makes it difficult for brands to know whether a particular person requires more exposure to advertising messaging in order to convert or whether that person has already been shown sufficient messaging to make a decision.

At the moment, most insurance companies are erring on the side of caution, preferring to inundate audiences with additional digital advertising instead of strategically focusing their spending on the people with whom it would make the most difference. With deep learning, insurance companies can drive down the costs of customer acquisition while improving the effectiveness of their advertising, enabling them to win away more new customers in a highly competitive market.

Optimizing digital advertising

What factors might induce someone to choose one insurance provider over another? The amount of the quote and the quality of the services provided absolutely play a part, but in the crowded world of insurance, where individuals have a multitude of companies to choose from, each of which offers a relatively similar roster of products for more or less the same price point, advertising and branding can play a crucial role in swaying someone to choose one provider over another.

Insurance companies understand this, hence the plethora of witty, exciting, and just plain weird ads that inundate the airwaves regularly. But what they lack is the ability to tell which type of person is more likely to pick their product over another. Considering that anybody possessing a home, car, motorcycle, etc. is a potential target for an insurance provider, being able to hone in on the specific traits that differentiate a future Progressive customer from an Allstate devotee is an incredibly valuable skill that helps brands save money while improving the likelihood of a successful conversion.

Thanks to technological advances, insurance companies can now recruit deep learning’s analytical capabilities — its ability to find discrete patterns within customer data that might hitherto have remained undetected by human marketers — to optimize their digital advertising and targeting. Deep learning allows insurance companies to differentiate between people who are already customers, people who are not yet customers but are amenable to conversion, and people who will not be swayed regardless of how many Limu Emu ads they are shown, thus enabling brands to focus their advertising spend on the most likely prospects and avoid wasting money on fruitless pursuits.

In other words, insurance providers can use deep learning to focus their advertising on the types of people who are most likely to be attracted by a brand’s unique proposition, whether this happens to be the particular value an insurance company is able to offer or the brand ethos the company has carefully cultivated.

Insurance providers can enlist a deep learning-enabled algorithm to examine all of the existing customer data they have on hand and combine it with available third-party data like demographics. From there, the algorithm will find similarities between existing customers, then identify people in the general population whose characteristics mark them out as good candidates for conversion. The benefit of relying on deep learning to carry out this search, as opposed to more manual marketing methods, is that the deep learning algorithm is not limited to a single understanding of what a customer looks like; instead, the algorithm is capable of identifying infinite combinations of characteristics that might distinguish someone as a potential customer. In some cases, what the algorithm identifies as a valuable audience cluster might run counter to what marketers believe to be their core consumer base, thus presenting brands with the opportunity to reach a previously untapped audience.

A solution for common insurer problems

As has already been mentioned, deep learning offers insurance companies the ability to hone in on those who are still on the fence and figure out the best way to convert them. Cognitiv’s research has found that the implementation of deep learning, in addition to delivering higher rates of incremental lift than other solutions currently on the market, is also capable of increasing conversion rates and improving ROI on incremental customers. For the insurance industry in particular, which has long struggled with incrementality and cost of acquisition, the existence of such a solution will help providers map out more effective, sophisticated marketing strategies that focus on reaching the right audiences in the manner most likely to lead to an efficient conversion.

Insurers spend so much money on advertising, often without the assurance that their blanket of ads is having the desired effect of swaying undecided insurance seekers. Many companies have tried for years without success to accurately measure incrementality and attain incremental lift on the scale they require. Deep learning finally offers a solution to those problems. By training a deep learning algorithm on insurance providers’ own first-party data, marketers can gain a better understanding of what their customers look like, and target others with similar characteristics while avoiding sending ads to existing customers or people who have already made their mind up. Diehard Geico fans need not fear, though: the Geico YouTube channel will always be there when they need it.

The post How insurers can use deep learning to boost performance marketing appeared first on Artificial Intelligence.

A new distance learning project that will enable more people to develop skills in robotics and autonomous systems, to help close the skills gap and drive forward productivity in the UK, is being launched by researchers from the University of Sheffield.

The project, led by Professor Tony Prescott from the University of Sheffield’s Department of Computer Science, will use cloud computing and state-of-the-art robots to develop distance learning activities in robotics for people across all educational levels.

Piloting a new way of learning in robotics education, the project will enable students to write programs from home, test them using a simulation, then run the program on a remote robot to see if it works in the real-world – a first for robotics teaching.

The project will also focus on addressing inequalities in the robotics and autonomous systems industry. It will provide positive and diverse role models together with the resources needed to help young people from disadvantaged backgrounds to go on and develop the skills to study and establish a career in the industry. 

With the UK economy needing to improve productivity to help recover from Covid-19 and remain competitive on an international level, the Sheffield-led team is developing a state-of-the-art robotic arm and manufacturing cell that engineering students at apprentice and degree-levels can use to develop skills in using robotics and autonomous systems in manufacturing. 

Engineering apprentices at the University of Sheffield AMRC Training Centre and the University’s degree-level computer science and systems engineering students will be able to reprogramme the robotic arm and manufacturing cell remotely.

Beginners can program a robotic simulation using the graphical language Google Blockly while more advanced students can program with the widely-used Python language. Students will be able to test their programs on physical hardware, remotely and in real-time using robot platforms located at the University. This is a new approach to robotics teaching that is being piloted for the first time in Sheffield.

The project will work with the University of Sheffield Advanced Manufacturing Research Centre (AMRC) to understand the skills gap experienced by industry in the Sheffield City Region and beyond in order to develop appropriate training strategies. 

Resources that will enable people outside of education to retrain and learn new skills in robotics and autonomous systems are also being developed by the researchers.

Professor Tony Prescott, Professor of Cognitive Robotics at the University of Sheffield, said: “The Covid-19 pandemic has left the UK with an urgent need to kick start its economy and boost productivity. A key to doing this will be increased automation and deploying the next generation of robotics and autonomous systems in the industry. 

“If we are to use more robots and autonomous systems then we need more people who have the skills to use these technologies. We need to encourage people from a young age to consider studying and developing careers in robotics while also providing the resources and the systems that will inspire young people and give them the platform they need to succeed. We also need to help the current workforce retrain and develop the skills they need to secure employment after the COVID-19 pandemic.” 

For primary and secondary schools, the project will trial a cloud-based simulation of the robot MiRo – a fully programmable autonomous robotic pet, created by the University spin-out company Consequential Robotics. MiRo is currently being used across the UK and abroad to pilot new applications for robotics in education and healthcare. 

The cloud-based simulation, MiRoCloud, will enable school pupils at both primary and high school levels to program a MiRo robot from home with the help of distance learning materials, tests, resources and lesson plans that are being developed by the University of Sheffield-led team. 

The project is working with schools that have high numbers of children from disadvantaged backgrounds and pupils with special educational needs to help widen access to the robotics industry. 

With equality and diversity being a key priority for the project, the educational resources are being developed with the aim of helping more of those who are currently under-represented in robotics to see where they can access learning opportunities and make their own contribution.

The research team will work with DiscoveryStem.org, a Sheffield-based educational consultancy that is at the forefront in developing schools teaching in robotics, to develop and share learning resources with primary and secondary schools across the UK. In the Sheffield City Region, the researchers will work in partnership with Beck Primary School – a large inclusive Sheffield school. 

Dr Becky Parry from the University’s School of Education is leading on ensuring the inclusivity outcomes of the project. Dr Parry said: “As a society, and in response to the rapid development of new technologies, we need new voices to help us imagine the potential value they will have to our lives. This project will be designed so that children can experience working in robotics as coders, designers or researchers and therefore imagine themselves in these roles in the future. 

“The researchers and students involved in running the project will seek to provide positive and diverse role models which we know are key to children making decisions about the subjects they are interested in. In the current climate, it is heartening to be part of a project which attempts to address structural inequalities in the short-term but also in terms of imagining the potential of robotics in the future to help us address the big issues of our times.” 

Professor Prescott added: “We hope that the new distance learning project will help us to develop scalable approaches for plugging the skills gap, then we can work together with schools, universities, industry and policymakers to roll this out across the UK.”

The distance learning project is linked to a national task group – Skills and Education in Robotics and Autonomous Systems (www.seras.org.uk), supported by the UK EPSRC Robotics and Autonomous Systems Network, that is preparing a White Paper on how to transform the UK’s skills in robotics and autonomous systems.

Led by the University of Sheffield, the White Paper is being developed in partnership with universities across the UK, the Institute of Coding, the Royal Academy of Engineering and the National STEM Centre.

The post Robotics Distance Learning Project Could Close Skills Gap And Boost Economy appeared first on Artificial Intelligence.